Full Report

The Semiconductor Arena

The semiconductor industry runs roughly $700 billion of annual revenue in 2025 and is widely modeled to cross $1 trillion by 2030 — the silicon layer under every server, phone, car, and AI model. Three things to internalize before reading the rest of this report:

1. The industry is cyclical, but right now in a generational supercycle. AI training and inference demand has pulled forward years of capacity build, lifted the price of leading-edge logic, and concentrated profits in a small handful of suppliers.
2. Profits do not pool evenly across the value chain. The economic rents sit with chip designers (fabless), the leading-edge foundry (TSMC), specialty memory (a global oligopoly), and the equipment makers gating it. Everyone else competes for residual margin.
3. AMD's reportable segments map to three different sub-industries — server CPUs, GPUs and AI accelerators, PCs and consoles, and industrial/automotive embedded — each with its own competitive structure, cycle, and margin profile.

Vocabulary — the seven terms to know

Seven definitions cover most of what is needed to read this report:

The value chain — where the money actually pools

A chip starts as a design in EDA software, gets fabricated on a wafer at a foundry, is packaged and tested by an OSAT vendor, then ships to an OEM. The profit pool is not evenly distributed. The table below shows gross-margin bands typical of pure-play participants in each stage, alongside the dominant firms — and where AMD sits.

The chart below distils that into a stylised view of gross-margin bands by stage.

Two takeaways drive nearly every strategic decision in this industry:

• Design and EDA earn the highest, most durable margins because their cost base is engineers, not factories. NVIDIA's roughly 75% gross margin reflects pricing power on AI GPUs, not generosity from customers; Synopsys and Cadence earn 65 to 80 because their software is a toll-road on every chip designed.
• The foundry layer is structurally a duopoly-plus. TSMC controls roughly 90% of the leading-edge node (3nm and below). Every fabless designer in the world — AMD, NVIDIA, Qualcomm, Apple, Broadcom — relies on that single supplier. The foundry earns mid-50s gross margins on those wafers; the designer who can charge a steep premium on the resulting chip earns far more.

AMD's position is unambiguously fabless: it designs chips, contracts TSMC (and historically GlobalFoundries) to build the wafers, ships them to ATMP partners in Asia for packaging, and sells to hyperscalers, OEMs, and distributors. This model trades the capital intensity of a fab (Intel's perennial drag) for dependency risk on TSMC's roadmap and capacity allocation.

Market size and growth — heterogeneity inside a $1 trillion story

The aggregate semiconductor market grew from roughly $574 billion in 2022 to about $700 billion in 2025. McKinsey models a 13% CAGR through 2030 (taking the industry past $1 trillion), but the headline disguises a wildly uneven split: AI-relevant segments grow 20%+ annually while mature/analog grows 2 to 4%.

This is the chart that drives AMD's investment thesis. The two segments where AMD has its biggest competitive position — server CPUs and AI accelerators — sit at the top of the growth distribution. Management has raised its server CPU TAM estimate to over $120 billion by 2030 (from a prior ~$60 billion), and it now expects data center AI revenue to compound at over 80% over the next three to five years.

The cost of that growth is also extreme. SEMI estimates that fab capital expenditure on advanced process equipment will rise from $26 billion in 2024 to over $50 billion by 2028 (a 94% increase). For AMD, every additional dollar of TSMC capacity for leading-edge nodes is a dollar it has to compete for against NVIDIA, Apple, Qualcomm, and others.

Cyclicality — the rhythm inside the growth story

Semiconductors are one of the most cyclical industries in the economy. Each cycle has a recognisable structure: demand spikes → fabless designers over-order from foundries → inventory builds at customers → demand normalises → orders crash → inventory clears → cycle resets. The cycle is typically 3 to 4 years peak-to-peak.

The historical revenue path of AMD is a near-perfect chart of where the chip industry has been over two decades. The annotations below identify the cycle phases.

Three observations worth holding onto:

• The 2007 to 2015 lost decade was structural, not cyclical. AMD spent eight years roughly flat near $5 to 6 billion of revenue while losing share to Intel in CPUs and to NVIDIA in GPUs. The recovery, beginning with Zen architecture in 2017, was an execution story, not a market story.
• The 2022 to 2023 dip was a normal cyclical downturn driven by the post-COVID PC reset and gaming console mid-cycle. PCs (now Client) and Gaming, which together make up 42% of 2025 revenue, are still cyclical — the AI tailwind cushions but does not eliminate that.
• The 2024 to 2025 inflection is unprecedented in size. Revenue jumped 34% in 2025 ($34.6B), with Data Center up 32% and Client and Gaming up 51%. Q1 2026 revenue ran at $10.3 billion (up 38% YoY), with Data Center now 56% of the company.

Competitive structure — concentrated, segment-by-segment

The semi industry looks fragmented in aggregate (hundreds of fabless designers, dozens of analog vendors), but at the product-category level it is one of the most concentrated industries in technology. Most categories are duopolies or three-player oligopolies. The table below maps the relevant arenas for AMD.

This is the most important table in the deck for understanding AMD's investment story. Each row is a different competitive game with different economics:

• Server CPU is a duopoly where AMD is the share gainer. Recent industry data (per latest market trackers) has AMD at roughly one-third of server CPU shipments and 46% of revenue — record highs. Intel's foundry struggles have left rival supply stalled, accelerating the shift.
• AI accelerator is a duopoly-plus where NVIDIA enjoys ~90%+ revenue share. AMD's Instinct MI300, MI350, and forthcoming MI450 are the credible Western alternative, with binding multi-gigawatt commitments from Meta and OpenAI announced in 2025.
• Game consoles is the quietest part of AMD's moat: it has effectively all semi-custom console silicon (PlayStation 5, Xbox Series, the new Valve Steam Machine), which is a high-volume, predictable, lower-margin business but a steady cash anchor.

The bubble chart positions AMD against listed peers. Bubble size = market cap; x-axis = most recent fiscal-year revenue; y-axis = trailing growth.

Two extremes anchor the chart. NVIDIA is in a class of its own — $4.8 trillion of equity value on $200B of trailing revenue, both growing 65%+. Intel sits at the opposite corner: similar revenue to AMD with no growth, a fraction of the market cap, and burning cash on foundry. AMD's positioning — 30%+ growth at $34.6B of revenue and roughly $850B of market cap at the June 17 close — is where the bull case is being priced.

Why margins differ so much — peer economics

The single number after revenue that matters most in semis is gross margin — the cleanest read on pricing power, mix, and competitive position.

Three patterns are doing all the work here:

• NVIDIA's 75% gross margin reflects pricing power on AI accelerators where supply trails demand and software lock-in (CUDA) lets it set the price floor. As long as data-center buildouts outpace TSMC's leading-edge capacity, this number is structural, not temporary.
• AMD's 50% gross margin is the central tension of the AMD story. It is dramatically better than where AMD sat a decade ago (in the 30s) but well below NVIDIA. Closing the gap requires lifting AI accelerator mix at favourable price points — exactly what management is betting the MI450 ramp will accomplish. Management has publicly targeted over 35% non-GAAP operating margin long-term, with non-GAAP gross margin already at 55% in Q1 2026.
• Intel's 33% gross margin is the cautionary tale. An IDM that has lost the leading-edge race ends up with a high fixed-cost base, idle fab capacity, and product mix shifted toward older, cheaper chips. This is the gravitational pull AMD escaped by being fabless.

A structural climb from below 30% in 2016 to roughly 50% today, propelled by Zen, EPYC, Xilinx, and now AI accelerators. Closing further toward NVIDIA's 70s is the leg embedded in the current valuation.

Supply-chain pinch points — where this industry breaks

For an AMD investor, four supplier-side chokepoints matter more than any other line items on the income statement.

The headline risk here is CoWoS. To build an AI GPU like the Instinct MI300, AMD needs TSMC to (1) fabricate the GPU die at 5nm, (2) source HBM3 from SK Hynix or Samsung, and (3) bond them together with CoWoS-S packaging at TSMC. Every AI chip in the world is competing for the same CoWoS slots. Bears argue NVIDIA's priority access to CoWoS structurally caps Instinct revenue regardless of demand — this is the single most-cited bear thesis on AMD's AI ramp.

Regulatory and geopolitical overlay

Semiconductors moved from an obscure trade category to a top-tier national security concern in the past five years. Three regimes shape what AMD can sell and where:

Where this leaves AMD — synthesis

Knit together, the industry view sets up the rest of this report as follows.

---

Know the Business

AMD is a fabless design house whose product portfolio sits on top of two of the most economically rivalrous franchises in technology: x86 server CPUs and AI accelerators. Everything else — Ryzen client chips, semi-custom console SoCs, Xilinx FPGAs, Pensando networking — scaffolds those two engines. The company prints $34.6B of revenue, $6.7B of free cash flow, and at the June 17 close of $520.89 commands ~$849B of market value — priced as if the next leg of execution is largely in the bag.

This page maps the parts behind that price tag: the economic engine, the segment-by-segment moat, the customer concentration that comes with hyperscaler dependence, the OpenAI warrant that transfers ~9% of the company to a single customer, the structural margin climb from 27% a decade ago to 55% non-GAAP today, and the lens for underwriting the stock.

The economic engine — three segments doing three different things

AMD reorganised in Q1 FY2025: Client and Gaming merged into one reportable segment, alongside Data Center and Embedded. Each segment is a different competitive game with different cyclicality, customers, and margin profile.

Three things in this table do most of the work for valuation:

• Data Center is now the largest segment (48% of FY2025 revenue, on track to roughly 56% in Q1 FY2026). Its operating margin of 21.7% looks lower than you'd expect — that is because the 2025 number absorbed roughly $440 million of net inventory charges from the MI308 China export ban. Strip that out and segment operating margin runs north of 24%; Data Center segment margin was 28% in Q1 FY2026, which is closer to the structural number.
• Embedded is the highest-margin segment at 36% segment operating margin (39% in Q1 FY2026). This is the Xilinx franchise. FPGAs are sold into aerospace/defence, industrial, and comms infrastructure where designs lock in for a decade and pricing is sticky. The segment fell 3% in 2025 on cyclical industrial/comms softness — it is, in effect, AMD's countercyclical anchor and its highest-quality earnings stream.
• Client and Gaming roughly tripled its operating margin from 12.3% in FY2024 to 19.6% in FY2025 on a 51% revenue jump. This is the Ryzen share-take story plus a Sony/Microsoft console cycle restart and the new Valve Steam Machine win — but it is also the most cyclical of the three segments, with management already flagging H2 PC demand softness from memory-cost inflation.

The shape of the FY2024 → FY2025 step-up makes it clear how unevenly the AI cycle is being distributed across the company.

In FY2025, Client and Gaming added $4.9B of revenue and $1.7B of segment operating income — more incremental segment OI than Data Center, which absorbed the MI308 inventory hit. That split is largely transient; Q1 FY2026 has Data Center back as the dominant earnings engine. The mix moves quarter-to-quarter; a consolidated multiple that ignores the segment shape will misprice this business.

Why fabless? The model that built the margin curve

AMD is a fabless designer. It owns design, IP, software, and customer relationships. It does not own fabs — it buys leading-edge wafers from TSMC, packages through Taiwanese ATMP partners (SPIL, ASE, KYEC, Tongfu JVs), and licenses some legacy 12/14nm capacity from GlobalFoundries via the long-running wafer supply agreement. This single architectural choice — to not run fabs — explains roughly everything about the company's economics.

Two facts to hold. AMD's capex is roughly 3% of revenue, not 30%; Intel's is north of 30%. AMD converted 19% of revenue to FCF in 2025; Intel ran negative free cash flow on higher revenue. Second, fabless does not equal high gross margin — NVIDIA's 75% GM and AMD's 50% GM are both fabless. The 25-point gap is mix, software lock-in, and pricing power, and it is the distance between today's valuation and the bull case.

The cost of fabless is a single-source supplier risk concentrated in TSMC at the leading-edge node (3nm, 2nm), TSMC CoWoS-S advanced packaging (the binding constraint on Instinct production), and HBM3E memory from SK Hynix/Samsung/Micron. Every dollar of incremental AMD revenue from data-center products has to win allocation against NVIDIA, Apple, Broadcom, and others at the same set of bottlenecks. Read the industry primer for the full chokepoint map — for this page, hold the single fact that AMD does not control its own supply.

The fabless choice has produced the cleanest fact in AMD's last decade: a structural climb in gross margin from the high-20s to ~50%, with management guiding non-GAAP gross margin to a 55–58% range over the medium term.

The flat read between FY2024 and FY2025 (49.4% → 49.5%) understates the underlying trajectory — the FY2025 figure is the post-MI308 charge number. Excluding the $440 million of net inventory write-downs, the underlying figure is roughly 50.7%, and the Q1 FY2026 non-GAAP gross margin of 55% sits above the historical range. Management's stated long-term operating margin target is north of 35% on a non-GAAP basis, against ~25% delivered in Q1 FY2026 — i.e. there is another ~10 points of margin embedded in the plan that has yet to be earned.

Moats — segment-by-segment, with the mechanism behind the score

A moat in semiconductors is usually a combination of (i) IP/architecture leadership at a leading-edge node, (ii) software lock-in, (iii) customer switching cost, and (iv) supplier preference at scarce manufacturing capacity. AMD's franchise is a stack of partial moats with very different durability profiles.

Two interpretive points:

• The semi-custom monopoly is the moat nobody talks about. AMD has 100% of current-generation Sony and Microsoft console silicon. Console contracts run 7–10 years from design start to console end-of-life and require integrated CPU+GPU+software co-engineering that no other public semis firm can offer at AMD's price point. Gaming margins are well below corporate average — Wall Street therefore underweights it — but it provides predictable cash for a decade and effectively pre-funds R&D for the broader graphics roadmap. The Valve Steam Machine win added in 2025 is meaningful in the same way.
• ROCm is the bear case in one acronym. NVIDIA's CUDA software platform is the actual moat in AI accelerators — not silicon. AMD has been investing heavily in ROCm (its open counterpart), and the 2025 release notes show real progress: day-0 model support, agent-based coding integration, MLPerf wins at MI355X. But this is a 15-year head-start that AMD is trying to close. Bulls cite the multi-gigawatt OpenAI and Meta commits as evidence that ROCm is "good enough" for inference at scale; bears cite the inference-skewed deployment mix as confirmation that training still belongs to CUDA. The truth is probably in between, with AMD's wedge in inference widening and training share remaining heavily NVIDIA-leaning.

Customer concentration — hyperscalers, two consoles, and a warrant for OpenAI

AMD discloses that "a small number of customers will continue to account for a substantial part of AMD's revenue and receivables in the future." The 10-K does not put names on the concentration, but the disclosed deals, the hyperscaler-led EPYC and Instinct ramps, and the semi-custom business make the customer set legible:

The OpenAI warrant — the most important economic fact most readers miss

In October 2025, AMD signed a 6-gigawatt product purchase agreement with OpenAI for Instinct MI450 deployment. Concurrently, AMD issued OpenAI a warrant to purchase up to 160 million AMD shares at an exercise price of $0.01 per share, vesting in tranches tied to GPU purchase milestones and stock-price targets, exercisable through October 5, 2030. None of the warrant shares had vested at FY2025 close, so there is no GAAP impact yet — but the economics are highly significant, and a serious investor must price this in explicitly.

AMD committed roughly $34B of equity value to OpenAI in exchange for an undisclosed multi-year GPU revenue stream. Whether the math is accretive depends on Instinct gross margin and revenue scale — neither has been disclosed. The Meta 6 GW commit announced in Q1 FY2026 carries no comparable warrant, suggesting the OpenAI structure was the cost of anchoring a first hyperscaler commit, not a recurring tax.

Geography — half the business ships internationally, but invoiced in dollars

International sales were 67% of FY2025 revenue (66% FY2024). Substantially all sales are denominated in US dollars, so there is no FX-translation drag. The economic geography is dominated by Asia, where ATMP/OSAT partners assemble and ship to OEM/ODM customers and forward to global end markets.

The more material geographic exposure is in manufacturing: TSMC's leading-edge wafer capacity is overwhelmingly in Taiwan. Cross-strait risk is industry-wide, not AMD-specific. The MI308 China export-ban (~$440M net inventory charges in 2025, partially reversed in Q4) is the more proximate regulatory drag — ~1.3% of revenue, but indicative of how a single BIS action can flip a quarter.

The cash machine — and where the cash is going

AMD's cash generation has stepped up sharply. Operating cash flow more than doubled in FY2025 to $7.7B (from $3.0B), and free cash flow tripled to $6.7B. Q1 FY2026 alone generated a record $2.6B of FCF — 25% of revenue.

Where does the cash go? The pattern is a tell about how management is allocating capital — heavy R&D investment, growing capex to support the data-center ramp, persistent stock-based compensation that runs roughly equal to buybacks, and the major Xilinx and ZT Systems M&A spend.

Two patterns to call out:

• Stock-based comp now runs at roughly $1.6B/year, against buybacks of ~$1.9B/year. Net buyback effort is therefore only ~$300M after offsetting SBC dilution — a fraction of the headline. For a roughly $800-850B-cap firm, that is not a meaningful capital-return policy. The OpenAI warrant tightens this further: future buybacks will de facto be funding warrant dilution rather than per-share earnings growth.
• R&D ran at 23% of revenue in FY2025 ($8.1B). That is unusually high for a 50%-gross-margin business and reflects the fact AMD is funding three simultaneous roadmaps (server CPU, AI GPU, AI PC) against an entrenched leader on the GPU side. Management has flagged that R&D will continue to outgrow SG&A as the AI investment cycle continues, which means operating-margin expansion has to come through revenue leverage, not cost discipline.

Return on capital — book ROIC is misleading, tangible ROIC tells the truth

The reported FY2025 return on equity of 7.1% and ROIC of 6.6% would suggest a mediocre business. They are not the right numbers. AMD's balance sheet carries $25.1 billion of goodwill and $16.7 billion of acquired intangibles — almost entirely from the $48.8 billion Xilinx acquisition closed in February 2022. Strip those out, and the picture is very different.

Peer economics — where AMD sits on the spectrum

The cleanest test of whether AMD is a high-quality business is to set its margin and growth profile against direct peers. The table below uses the latest reported annual figures for each company; market caps and enterprise values are as of early June 2026.

Margin ladder in one chart — the order of merit is unambiguous:

Three things to take away:

• AMD sits in the middle of the fabless peer set on gross margin and at the bottom on operating margin. It is not a NVIDIA, but it is not an Intel either. The 50% gross margin is well below NVIDIA's 75% and Broadcom's 63%; this is the closeable gap if AI accelerator mix scales.
• Operating margin tells a sharper story: AMD's 10.7% GAAP op margin is hit hard by R&D intensity (23% of sales) and acquisition-related intangible amortisation. On a non-GAAP basis AMD ran 25% op margin in Q1 FY2026. The cleanest framing is that the gap between GAAP and non-GAAP op margin is the Xilinx amortisation tail — about 6 points of margin — that runs off mechanically over the next several years.
• EV/sales of ~22x at AMD is in the same neighbourhood as NVIDIA's 23x. This is the part of the page that pays attention: the market is already paying NVIDIA-like multiples on AMD's revenue, on the assumption that the next leg of AI accelerator share gain and margin expansion will arrive. If you do not believe that, the stock is not cheap.

The scatter below is a quick way to see relative positioning — bubble size is market cap.

The shape is the bull case for AMD: it is in the upper-right quadrant alongside NVDA and AVGO (the businesses earning premium multiples) and well above INTC and the slower-growth fabless peers. The bull case is that AMD migrates up — toward NVDA-like gross margin as Instinct mix scales — while staying in the high-growth band. The bear case is that gross margin compresses as MI450 ramps at below-corporate-average margin while R&D stays elevated, leaving AMD stuck below 50% gross margin for longer than the multiple implies.

Cyclicality and the AI override

Semiconductors are deeply cyclical historically (industry primer covers this in detail). AMD's segment mix means cyclicality enters the consolidated P&L unevenly:

The composite cyclicality of the consolidated company is now moderate, not high — because the AI-driven Data Center revenue (and Q1 FY2026 mix at 56% of company revenue) is large enough to absorb cyclical drawdowns in Client/Gaming/Embedded. But this is conditional on the AI cycle. If hyperscaler capex normalises, the Data Center segment becomes cyclical at the same time as everything else, and the consolidated business looks more like the 2022–23 dip (revenue down 4% in 2023). A serious investor sizes that risk before paying 22x sales.

How to value this business

There is no single multiple that captures AMD. The right way to frame this is a sum-of-the-parts where each segment is valued on its own economics, then haircut for warrant dilution. The lens varies because the underlying businesses are unalike.

The single chart to sit with before forming a view:

This is the central tension. AMD trades at NVDA-like EV/sales but with materially lower current gross margin and operating margin. The implicit assumption in the multiple is that AMD closes a large fraction of the gap toward NVDA's margin profile over the next 3–5 years. That assumption rests on (a) Instinct gross margin migrating toward corporate average and (b) ROCm becoming "good enough" to widen inference share. If either fails, the multiple compresses.

Synthesis — what an intelligent investor should walk away with

---

The Underwriting Question, Reframed for Five Years

Holding AMD for five-to-ten years is not a bet on AI demand — that is already in the multiple. It is a bet on a narrower proposition: that AMD compounds durable earnings across four franchises whose moats are unequal, while closing a software gap (ROCm vs CUDA) it does not own, against a supply-side bottleneck (TSMC CoWoS, HBM) it does not control, and while absorbing ~9% structural dilution from a customer-equity warrant most consensus models still ignore. Only the AI accelerator franchise must scale on schedule for the current ~$850B market cap (June 17 close) to make sense. The other three (Embedded, console semi-custom, server CPU) are the floor: ~$6.7B of FCF already in hand, the reason this is an underwriteable long-term holding rather than a momentum trade. The frame below separates the signals that prove or break the five-year case from the noise of any given quarter.

What Has To Be True — The Five-Year Frame

The 5–10 year thesis rests on five propositions, each independently testable. Three have decade-long head starts; two are unproven. The table is the operative thesis statement of the page — every section that follows interrogates one of these.

The asymmetry is the entire investment case: three of five long-term propositions are proven, one is unproven and carries the multiple, one is the swing factor on per-share compounding. A 5-year holding works if P2 lands within ~70% of management's framing — and it can land below management's framing and still produce a defensible return because P1/P3/P4 compound underneath. It breaks if P2 and P5 fail together: Instinct fails to earn corporate-average margin and warrant vesting absorbs the buyback pace.

The Four Engines — Different Cycles, Different Moats, Different Lenses

A consolidated AMD multiple obscures more than it reveals. The right frame is a sum-of-engines, each underwritten on its own economics over a 5–10 year horizon. The valuation lens varies because the underlying franchises are unalike.

The chart below is the central insight: AMD's earnings quality is concentrated in the engines that drive the least of the market cap, and the engine that drives the most of the market cap is the least-proven. That asymmetry is the source of both the bull case (re-rating if Instinct earns rent) and the bear case (multiple compression if Embedded margin is the only thing carrying through-cycle quality).

The EPYC Compounder — The Most-Underweighted Proven Moat

If anything in AMD deserves a Buffett-style "inevitability" label over a 5–10 year horizon, it is the server CPU franchise. Eight consecutive years of revenue-share gains across two semi cycles is not a tailwind; it is a mechanism — hyperscaler requalification cost, node-process lead via TSMC, and an architectural roadmap (Zen 5 → Zen 6 → Zen 7) that has been delivered on schedule for nearly a decade.

Two things in this chart matter for the 5-year frame:

• The revenue line is well above the unit line and the gap is widening. That is the ASP premium AMD earns over Intel — running near 42% per CPU. It is the cleanest piece of evidence that EPYC is not a discount-driven share gain but a pricing-power moat. ASP premium has expanded every year of the series.
• The slope did not break in either cyclical drawdown. 2022–23 was a brutal PC and server downcycle for the industry; AMD took share through it. That is the discriminating evidence between "moat" and "favourable cycle".

Management's revised server CPU TAM is $120B by 2030 (up from ~$60B). Mid-case, 50% share = $60B of EPYC revenue by 2030 — roughly 4x today's franchise — at the segment's historical 25-30% operating margin. That alone is ~$15-18B of segment OI. Net of intangible amortisation run-off and at a conservative 15x multiple, the EPYC franchise standalone is worth roughly $250–300B of market value by 2030 — i.e. covers ~35% of today's market cap from a single engine that already exists and is compounding.

The Instinct Wedge — Where The Thesis Is Made Or Unmade

Instinct is the engine that decides the next $300B of equity value. It is also the engine where the moat is least proven and the supply chain is least controlled. The bull case is that ROCm closes enough of the CUDA gap to lift gross margin into the corporate range while volume scales 5-10x; the bear case is that AMD is competing as a price-taker on hardware against NVIDIA's software-priced silicon, on a supply chain where NVIDIA gets allocation priority, and against a custom-ASIC alternative (Broadcom + Marvell) that takes the highest-margin AI dollars before merchant GPU sees them.

The three scenarios anchor to specific mechanics. The bull path ($70B by 2030) requires ROCm to reach framework default-path status by FY2027 and CoWoS allocation to rebalance toward AMD as TSMC doubles capacity. The base path ($35B by 2030) is roughly management's "tens of billions in 2027" trajectory carried forward at 30% CAGR through the decade — credible if MI450/MI500 ramp on schedule but with Instinct GM below corporate average through 2027. The bear path plateaus at $12-14B because custom-ASIC compression and CUDA lock-in cap merchant-GPU TAM at AMD's share. Note the three paths diverge in 2027, which is exactly when the OpenAI deployment commits start to print and when consensus will mark its FY2028 model. 2027 is the year the multiple is set for the rest of the decade.

The Hidden Compounder — Why Embedded Is The Most Underpriced Part Of AMD

The Embedded segment generates 10% of revenue, 36% of segment OI (39% in Q1 FY26 — highest in the company), and almost none of the equity-story attention. It is the cleanest example in the report of an annuity-like franchise hiding inside a portfolio whose blended margin denies it visibility.

Read this chart with the segment lens: Embedded revenue fell 33% peak-to-trough through the worst industrial/comms downcycle in a decade and segment operating margin still held above 36%. Lattice — the pure-play FPGA comparable — runs $523M of revenue against AMD's Xilinx franchise's $3.5B base, with a smaller R&D pool and no ability to fund Versal-class adaptive SoC roadmap. The competitive structure of FPGA is one of the cleanest duopoly-monopoly setups in semiconductors — Vivado/Vitis customer lock-in, 10+ year aerospace/defence design cycles, qualification-cost barriers, and no credible high-volume ASIC migration path for the design wins that drive the segment.

A 5-year frame for Embedded that recovers to FY2022's $4.5B revenue at 38% segment OI = $1.7B of segment OI growing low-double-digits as design-win backlog (~$50B cumulative) converts. At a conservative 15x multiple — well below FPGA-pure-play Lattice's >30x — Embedded is worth $25-35B standalone. That is roughly twice the goodwill impairment risk that bears flag on the Xilinx purchase. The hidden compounder is hidden because Wall Street allocates the entire $42B Xilinx purchase price to the consolidated DCF and forgets to ask what the segment is worth standalone.

Capital Allocation — The OpenAI Warrant Reframes The Math

For a 5-10 year holder, the long-term per-share compounding equation reduces to a simple identity: FCF growth must outrun (SBC dilution + warrant dilution + acquisition dilution) for the equity to compound for the buyer. AMD has the cash generation; the question is whether the warrant absorbs it.

Two facts the consolidated FCF number obscures:

• Net buyback effort after SBC is ~$300M against a $760B market cap. That is roughly 0.04% of equity retirement annually. Buybacks are not a per-share compounding lever at this scale; they are an SBC-offset mechanism. For shareholders to earn EPS growth above revenue growth, the share count has to fall, not stay flat.
• The OpenAI warrant is the largest single equity issuance event in AMD's post-Xilinx history. At max vest (160M shares), the fully-diluted count moves to ~1,780M — a one-time 9.6% dilution that the FY25-26 buyback pace cannot absorb. The ASC 606 contra-revenue mechanic compounds the issue: as warrant vesting becomes probable, AMD records the fair value of the vested tranche as a reduction of Instinct revenue, compressing reported gross margin precisely when the AI ramp is supposed to inflect. Consensus FY27 EPS models that use 1,650M shares understate dilution by 6-8%.

The redemption is that the Meta 6GW commit (announced Q1 FY26) does not appear to carry a similar warrant. That is the cleanest evidence that the OpenAI warrant was a one-time cost of entry to anchor the first hyperscaler AI commit — not a recurring tax on every future deal. If a third anchor signs at standard pricing, the warrant is a known overhang to be modelled, not a precedent. If the third anchor also requires equity, the franchise is structurally weaker than the bull case requires.

Margin Architecture — Where The 35% Non-GAAP Op Margin Comes From

Management's stated long-term target is >35% non-GAAP operating margin against the 25% delivered in Q1 FY2026. That ~10-point gap is the single most important structural assumption inside the current multiple. The build is mechanical in three parts and discretionary in one.

The bridge resolves into a clean test. Roughly 11 points of margin expansion are mechanical — Xilinx amortization runs off, Embedded recovers, EPYC scales, R&D leverages on DC mix. Roughly 4 points are discretionary — they require Instinct to migrate toward corporate-average margin against CFO-confirmed near-term compression. If the mechanical 11 points arrive and the discretionary 4 points don't, AMD lands at ~31% non-GAAP operating margin against a 35%+ target. That is still a doubling of operating leverage over the next 5 years, and at $76B of FY27 consensus revenue produces ~$24B of non-GAAP operating income — a real number, just below the mark the multiple is paying for.

The historical curve is the most underweighted piece of evidence for the long-term case: AMD has already delivered the hardest part of the margin journey — from negative operating margin in 2014 to 25% non-GAAP today. The next 10 points are smaller than the 27 points already earned, and a larger share of them is mechanical rather than execution-dependent.

The Reinvestment Runway — Why R&D Intensity Is Both Strength and Risk

AMD spent $8.1B on R&D in FY2025 — 23% of revenue — to fund three simultaneous roadmaps (server CPU, AI GPU, AI PC), against an entrenched leader on the GPU side. That intensity is the highest among large fabless peers and it is a strategic choice: management has flagged that R&D will continue to outgrow SG&A as the AI investment cycle continues. For a 5–10 year holder, this is the cleanest reinvestment-runway test in the company.

Two things this chart says. First, AMD is reinvesting at Intel-like intensity (Intel's R&D-to-revenue is actually similar — 24.6%) but earning fabless-peer cash flow. That is the structural mismatch the multiple is paying for: if AMD can drop the R&D intensity from 23% to a fabless-peer-typical 12-15% as revenue scales 2-3x, every freed point of revenue flows directly to operating margin. Second, AMD is not Intel because the R&D productivity is producing share gains, not architectural catch-up — 8 years of EPYC compounding share is the receipt that AMD's R&D dollar buys results that Intel's R&D dollar does not. The risk is that the next 5 years of R&D goes into ROCm and Instinct against CUDA — the area where AMD's R&D productivity is not yet proven.

The reinvestment runway is therefore both the bull and bear case in one number. Run R&D at 23% of revenue for 5 more years and AMD has an unbeatable engineering moat; run R&D at 23% and still lose CUDA, and the buyer paid for software parity AMD never earned.

The Five Ways This Breaks — Multi-Year Failure Modes

A long-term holding doesn't get broken by quarterly miss-and-beat noise. It gets broken by structural failures that compound across multiple periods. The five failure modes below are the ones a PM should size against — each is independently testable, observable at known cadence, and tied to a specific thesis proposition.

Multi-Year Watch Signals — The PM's Long-Term Dashboard

The dashboard below is the discipline this page enforces. Each signal is not a quarterly catalyst — it is a multi-year direction-of-travel reading that should change conviction over rolling 12-24 month windows. This is what a 5-10 year holder should be tracking; everything else (sell-side note flow, single-quarter beats, daily borrow data) is noise.

Scenario Architecture — What Compounds, What Doesn't

The scenario table below is the discipline I would force a junior PM through: for each five-year case (bear, base, bull), name the evidence pattern that produces it, the implied per-share economics, and the capital allocation reality that makes it happen or not. The cases are deliberately asymmetric in probability — base is the largest weight, but the bear case is non-trivial because the variable that decides it (Instinct GM) is the one CFO has explicitly guided against.

Probability-weighted fair value 2030 ≈ $360. Against today's ~$520 reference price, that math is uncomfortable: even at 50% weight on a $432 base case and 20% weight on a $940 bull case, the expected value sits ~30% below spot. The underwriting question rests on that asymmetry — the price already pays for the bull case to materialise.

The bear case understates a structural feature, though: the EPYC + Embedded + Console floor is durable enough that the equity does not zero — it re-rates to a slower-growth profile while the franchises themselves stay intact. AMD is neither Intel 2014 (terminal share loss) nor Snowflake 2024 (one-product franchise) — it is a portfolio of moats with one not-yet-proven leg. The cleanest 5-year setup to underwrite this stock is not at current prices; it is into the next 30-40% drawdown, when Instinct execution is in doubt and the floor franchises trade as cyclical drag.

Where Conviction Should Sit — And Where It Should Not

The final synthesis is a clear separation of what a 5-10 year holder should believe from what they should price as optionality.

The honest read: four out of eight propositions should be believed with high conviction; three should be priced as optionality; one (per-share compounding) should be believed only with discipline on the buyback pace. That gives the buyer a defensible long-only thesis on the proven franchises while allowing the unproven legs to drive upside without underwriting them at full freight.

The PM's Synthesis — One Paragraph

---

The Competition

Three questions an AMD investor needs a view on: who can hurt AMD, who AMD can beat, and what evidence proves the difference — segment by segment, name by name.

The verdict, in one paragraph

The peer set — and why these five

No single public company is a clean comp for AMD. The right peer set is the smallest collection that covers every segment AMD competes in. Starting from AMD's own FY2024 10-K (Intel, NVIDIA, Qualcomm, Marvell, Broadcom, Lattice across CPU, GPU, FPGA, ASSP) narrows to five primary names plus one supplemental.

The four names the staged work rejected — and why — are worth stating explicitly so the peer set isn't taken at face value:

• Texas Instruments (TXN) — analog/embedded only; no overlap with leading-edge compute, GPU, or AI silicon.
• Micron (MU) — memory peer, not logic. Memory cycle, not compute cycle.
• TSMC (TSM) — AMD's foundry, not a competitor; belongs in supply-chain, not peers.
• Apple (AAPL) — captive vertically-integrated SoC for its own devices; not a merchant-semis peer.

Lattice is kept supplemental rather than primary because its $19B market cap, $0.5B revenue, and pure-FPGA mix are too narrow to anchor a five-name set — but for the Xilinx Embedded franchise it remains the cleanest standalone comparable.

Peer comparison — scale, growth, margin, value

The most useful single table on the page. Latest-fiscal-year revenue and operating-margin figures; market cap and enterprise value as of early-to-mid June 2026 per the staged Yahoo Finance Parallel Task. AMD is included as the comparison anchor.

The peer landscape on two dimensions — growth on the x-axis, gross margin on the y-axis, market cap as bubble size — is where the AMD setup gets visible:

AMD sits in the upper-right quadrant alongside NVDA and AVGO — the premium-multiple names — but lower on gross margin and materially lower on operating margin. INTC is the cautionary tale: same revenue scale, zero growth, half the gross margin, negative operating margin, a fraction of the equity value. Bulls expect AMD to migrate up this chart as Instinct mix builds; bears expect AMD to stay at 50% gross margin while ASIC competitors take the high-margin custom AI dollars.

The arena map — segment-by-segment scorecard

AMD's moat strength is highly uneven across segments. The scorecard below names, for each arena, the market structure, AMD's current position, the rival that pressures AMD, the evidence, and a directional moat call.

The same scorecard as a heatmap — green is AMD wins, red is AMD loses:

The pattern is the central fact: AMD wins decisively where it has been winning for years (server CPU, console, FPGA) and loses decisively where the software stack and capital model are the moat (CUDA, custom ASIC). The middle rows — AI accelerators, discrete consumer GPU, AI PC, networking — are where the next three years of share movement decide the equity story.

Where AMD wins — the four concrete advantages

Each row is tied to specific evidence in the filings or web research.

Server CPU is AMD's most-proven competitive advantage, and the line is still expanding.

Where competitors win — four concrete weaknesses

Same standard — name the competitor, name the evidence.

The gross-margin ladder makes the second of these weaknesses the most visible — the fabless-with-software-moat names (NVDA, AVGO) sit a clean 15-30 points above AMD:

Threat assessment — the top six

The threat is not "competition is intense." The threats are specific, named, and timed.

Moat watchpoints — five signals that change the call

Forward signals — measurable, disclosed at known cadences, and tied to a competitive judgment the price depends on.

Synthesis — the competitive call in three sentences

---

Current Setup & Catalysts

AMD closed $520.89 on June 17, 2026 — +133% YTD, +24% in the last month, 4.8% below the $547.26 all-time high, 30-day realized vol 91% (top decile, ten-year). The market is pricing the OpenAI 6 GW + Meta 6 GW commits as contracted FY27 revenue while leaving the structural drag uncredited — CFO-confirmed sub-corporate gross margin on MI450, ASC 606 contra-revenue when the first 1 GW deployment becomes probable in 2H 2026, and the TSMC CoWoS allocation ceiling AMD does not control. This page bridges the 5-10 year thesis (which turns on whether ROCm closes the CUDA gap by mid-FY2027) and the three near-term hinge events — Advancing AI on July 23, the Q2 print on August 4, and the Q3/Q4 FY26 Helios + MI450 ramp — that decide which way the FY27 consensus number ($13.10) is revised.

The Variant View — Where We Sit vs Consensus

The page is organized around edge, not around the calendar. Before any catalyst is ranked, the question every reader needs answered: where does the analyst differ from the Street, and by how much?

What Changed in the Last 3-6 Months — The Setup That Brought Us Here

Five things have shifted between February and June 2026. The first three are the reason the stock is at $521; the last two are why it should not be priced for perfection from here.

The arc since February: a soft FY26 guide and -17.3% reaction (Feb 4) was overwhelmed by a +14.9% Q1 print on May 5, followed by a parabolic rally through May (+46% from $354 → $516). The bull narrative has now had its receipts; the bear narrative is still waiting for its first scheduled event to land — most likely the Q3 print in November, when MI450 mix first dilutes reported GM.

Historical Earnings Price-Reaction Base Rate — Anchor for "How Much Does This Move?"

A "high impact" claim is only credible if it is anchored to how the stock has actually moved on past prints. AMD's earnings base rate over the last eight quarters tells a clear story: modest EPS surprises but outsized stock reactions — the median absolute one-day move on an earnings or anchor-event day has been roughly 6.5%, with extreme outliers in both directions when the AI narrative is in play.

Three patterns from the base rate that govern every High-impact magnitude claim in the timeline below:

• The headline EPS surprise is uncorrelated with the move. Q4 FY25 beat by +16% and the stock fell -17.3%; Q3 FY25 beat by +2.5% and rallied +5.8%. The reaction lives on guidance, segment mix, and gross-margin trajectory — not on the print itself.
• Asymmetry is in the negative tail. Five of eight earnings days moved up, three down, but the three downs averaged -12.0% versus the five ups averaging +6.6%. With realized vol now 91% and one ATR equal to a 4.6% swing, an earnings miss in this regime is sized at -10 to -15%, and an OpenAI-scale deal announcement is sized at +15-25%.
• Non-earnings narrative events can be the biggest movers. The OpenAI announcement (Oct 6, 2025) was the largest single-session move in five years (+23.7% on 4.3x average volume). The Advancing AI 2026 keynote on July 23 is the same archetype — calibrate the upside scenario accordingly.

What the Market Is Watching Now — The Live Debate

Six questions are actively contested in the file. For each: what the market cares about, what would confirm the current bull-led view, and what would challenge it.

Ranked Catalyst Timeline — By Decision Value, Not Date

The required artifact. 8 catalysts inside the next 6 months, ranked by what would most update the underwriting debate, not by chronological order. Every High-impact row carries a quantified magnitude (estimate delta and expected stock move), a skew read, and a positioning line.

Catalyst Calendar at a Glance

The cluster is unusually dense for a roughly $800-850B-cap: three hard-dated High-impact catalysts inside 90 days, plus continuous hyperscaler capex tape. The first 50 days carry the bull-confirming events (Advancing AI + Q2 print + potential Meta/MSFT reaffirmation). The 90-180 day window houses the asymmetric-down setup (first MI450 mix quarter + first ASC 606 disclosure).

Impact / Decision View — Which Events Resolve the Debate?

The table above ranks by decision value; this view separates the events that resolve the underwriting debate from those that add information. Not every catalyst is a thesis update.

Next 90 Days — The Focused Watchlist

The 90-day calendar is the densest AMD has carried since the OpenAI announcement. The PM read: the bull-confirming window closes August 4; the asymmetric-down window opens November 4. The intervening 90 days is when the variant view either lands or gets called.

What Would Change the View — The Three Signals to Watch

If only three things change over the next ~6 months, these are the three that would force a real thesis update. Each is tied back to a specific upstream conclusion.

The PM's One-Paragraph Read

---

Bull and Bear

Verdict: Watchlist — the bear's valuation-vs-margin tension is decisive, and the data that resolves it is two to three prints away. At ~22x EV/sales, AMD trades inside NVIDIA's neighborhood, and the CFO has explicitly said MI450 ramps below corporate-average gross margin for the next one-to-two years. The EPYC moat and tripled free cash flow are real and underpriced — but neither alone carries a 22x EV/sales multiple if Data Center gross margin compresses on the AI mix shift. The question is whether AMD's AI-franchise multiple is being paid for revenue, or for margin that the operator has guided against on the consensus path. What changes the call: two consecutive quarters of Data Center segment gross margin holding above 55% while Instinct revenue scales, or a single print landing in the bear's sub-22% Data Center operating margin scenario.

Bull Case

Bull scenario target is $700 on roughly 47x a Bull FY27 EPS of $15 (slight upside to $13.10 consensus if Data Center non-GAAP operating margin sustains at 25%+), cross-checked at ~17x EV/sales on $76B FY27 revenue. Time horizon is 12–18 months through the Q4 FY26 and Q1 FY27 prints that anchor consensus on FY27 EPS, with MI450 volume milestones in H2 FY26 the primary catalyst. Bull's disconfirming signal: Mercury Research server-CPU revenue share rolling back below 42% in any quarter before Intel 18A Panther Lake is in volume.

Bear Case

Bear scenario target is $270 (roughly 47% below $520.89) on forward P/E compression to peer-median ex-NVDA ~30x applied to a normalized FY27 EPS of $9 (vs $13.10 consensus). The haircut combines ~9% share dilution from the OpenAI warrant plus ASC 606 contra-revenue drag, MI450 gross margin sub-corporate-average through 2027, and the FY26 effective tax rate normalizing to 13% from FY25's −2.5%. Time horizon is 12–18 months for Q2–Q4 FY26 prints to validate margin compression; the primary trigger is a Q3 or Q4 FY26 print with total GAAP gross margin sliding below 52% and Data Center operating margin below 22% as MI450 mix builds. Bear's cover signal: a single quarter printing Instinct revenue above $10B annualized at greater than 50% segment gross margin, paired with PyTorch or TensorFlow shipping default-path ROCm support at parity with CUDA.

The Real Debate

Verdict

Watchlist. The bear carries more weight at the current price because the bull case rests on Data Center gross margin expansion that the company's own CFO has guided against for the next 1–2 years — and the multiple already prices the bull outcome. The decisive tension is Debate #2: at 22x EV/sales, AMD is paid for margin trajectory CFO commentary contradicts, and the bull's mechanical Xilinx-amortization argument does not offset Instinct mix dilution if MI450 ramps below corporate average through 2027. The bull could still be right — the EPYC moat is the most-tested franchise in the report (eight years, two cycles, no rebut from Intel 18A in volume yet) and the Meta 6 GW commit without a warrant is real evidence the AI anchor flywheel does not require recurring equity payment. The thesis-breaker is durable — two consecutive quarters of Data Center segment gross margin holding above 55% while Instinct revenue scales meaningfully would prove the AI franchise earns rent rather than volume. The nearer-term marker is the next 10-Q's ASC 606 disclosure on whether OpenAI warrant vesting has been judged probable — that determines whether contra-revenue lands in reported numbers before the margin debate resolves. Anything else (Mercury share data, single-quarter beats, sell-side rating changes like the June 8 Northland downgrade) is noise relative to those two markers.

---

Moat — What Protects AMD, And What Doesn't

The verdict in one paragraph

How to read this page

The Industry and Business pages covered what AMD does; Competition covered who can hurt it. This page asks the question that matters for sizing: what specifically prevents a well-funded competitor from taking AMD's customers, margin, and share — segment by segment — and how do we know it works? Evidence over adjectives. Mechanism over labels.

Mechanisms in play vs. not in play

Semi moats are usually some combination of architecture/IP leadership, software lock-in, customer requalification cost, and preferential access to scarce manufacturing capacity. The table maps each candidate source against AMD — present, partial, or absent — and where it actually operates, because a moat that protects 10% of revenue is not the same as a moat that protects 50%.

Two readings collapse the table.

• The protective mechanisms are concentrated in the smaller segments. The clearest switching costs and the cleanest software lock-in sit in FPGA and console semi-custom (~22% of revenue combined), not in Instinct (which carries the multiple). The arenas where AMD has the most evidence of durable advantage are not the arenas that drive the next ~$200B of equity value.
• AMD is on the wrong side of the only mechanism that matters at scale right now. CUDA is the binding moat in AI accelerators, and NVIDIA owns it; CoWoS is the binding supply constraint, and NVIDIA has preferential access to it. AMD is competing into a duopoly where the rival controls both the demand-side moat (developer ecosystem) and the supply-side bottleneck.

Segment moat scorecard — rating, mechanism, evidence, stress test

The operative table. Each row scores a franchise on a five-step scale, names the mechanism, cites the evidence, and stress-tests the moat against a specific shock in the next 24–36 months.

The visual cut of the same scorecard — where rating is high (5) and where it isn't (1) — makes the asymmetry obvious:

The proof — does the moat show up in the numbers?

A moat that does not show up in returns, pricing, share, retention, or margin spread is a story, not a moat. Three readings test whether AMD's claimed advantages are working.

1. Share momentum — the cleanest evidence the EPYC moat is real

Server CPU unit share is the single most-cited piece of AMD-specific moat evidence, and it is real. The eight-year compounding share gain has continued through two semi cycles and has accelerated in the AI era — a pattern consistent with structural advantage, not cycle-aided tailwind.

The 32.6% Q1 FY26 unit share against 46.2% revenue share also reveals an ASP premium — AMD's average server CPU sells for ~42% more than Intel's by unit, a quantitative proof that EPYC pricing power is intact even as share has compounded. Industry reporting in May 2026 noted EPYC Turin production "capacity sold out" with rising lead times — a supply-allocation signal that confirms pricing is not being earned by discount.

2. Margin structure — fabless economics, but not yet pricing-power economics

A fabless competitor with a real moat should sit near the top of the gross-margin ladder; AMD sits in the middle. The ladder is the proof of where pricing power lives across the peer set.

The reading: AMD's 50% gross margin is closer to the floor of the fabless peer set than the ceiling. The 25-point gap to NVIDIA is the price the market is paying AMD not to be a moat-bearing AI franchise. Even small-cap Lattice — a 60% gross-margin pure-FPGA peer — earns higher gross margin than AMD's consolidated business, because Lattice has no Instinct or semi-custom drag. AMD's moat franchises are real but they live inside a portfolio whose blended margin denies them visibility. That is why the Embedded segment is, as the Business page argues, the most underweighted franchise on the page.

3. Returns on tangible capital — book ROIC misleads; tangible ROIC tells the truth

The first instinct from FY2025's 6.6% reported ROIC is that AMD is not a moat-bearing business. That number is an artifact of the $42B Xilinx goodwill+intangible base, which the operator did not pay for in cash AMD ever has to earn back on. On tangible capital — what an operator running this franchise from scratch would need to fund — the return is materially better.

Tangible-capital ROIC of ~24% is consistent with a narrow-moat business that is earning a return above cost of capital on the operating assets it has. The pre-Xilinx ROE of 47% is the underlying earnings power of the standalone semiconductor design franchise; the Xilinx acquisition diluted the denominator and lowered the reported number. Underwrite the moat on tangible ROIC; underwrite legal claims on book ROIC. The two diverge by ~18 points.

4. Stress-test history — what survived

A moat that has not been tested through a downcycle is a hypothesis. AMD's franchises have differing track records here.

Five proven moat franchises and three unproven or non-participating ones is an honest summary. The proven franchises happen to be the legacy and incremental businesses; the unproven ones happen to be the AI franchise. The valuation pays for the unproven side.

Moat vs. execution — separating what's protected from what's earned

Three of AMD's strongest commercial facts are not moats — they are execution, and they could be reversed by execution at a rival. Separating these out matters because investors routinely conflate the two.

This is the discipline the page is designed to enforce: every line item that is not a mechanism preventing competitor action is something else — execution, industry structure, contract, or balance-sheet optionality — none of which are moats and none of which deserve a moat multiple.

Customer concentration — moat or vulnerability?

A moat that depends on a small number of customers is a contract, not a moat. AMD's disclosed customer concentration is high and getting higher, and the OpenAI warrant has reframed what "customer" means.

Supply moat — AMD does not have one; NVIDIA does

A moat that runs upstream — preferential access to scarce manufacturing capacity — is one of the most durable in semiconductors, because supply allocation runs 12+ months ahead and is sticky to incumbents. AMD does not have this moat. It actually has the opposite: it is a price-taker on the binding constraint.

The honest reading: NVIDIA has a supply-side moat that AMD does not. That is one of the structural reasons NVIDIA earns 75% gross margin and AMD earns 50%. The Industry primer made this point at the chokepoint level; the Moat page makes it at the competitive-dynamics level. A moat-quality investor must price the asymmetry, not assume capacity neutrality.

What would weaken the moat — five trigger-level watchpoints

Forward signals. Each is measurable, disclosed at a known cadence, and tied to a specific moat-quality judgment the stock price depends on. Trigger levels are deliberately concrete — if the metric prints below the level shown, that part of the moat is materially weaker than the current rating implies.

What would strengthen the rating to "wide"

For symmetry — three discrete signals would move the verdict from narrow to wide, and they are roughly the inverse of the trigger watchpoints above.

Synthesis — the three sentences

---

Financial Shenanigans

Verdict — Watch (32/100). The reported numbers look like a faithful representation of economic reality: no restatement, no SEC investigation, no auditor flag, clean unqualified E&Y opinion. But FY2025 GAAP earnings carry unusually high one-time content — an $853M release of uncertain tax positions flipped the effective tax rate negative, a $360M reversal of a prior MI308 inventory write-down padded Q4 gross margin, and a $577M "other income" line is dominated by mark-to-market gains on long-term investments. The 13-test scan returns ten green and three yellow categories — no red — but underwriting should haircut FY2025 earnings quality before sizing.

Headline forensic KPIs

The cash-conversion picture is the strongest single piece of clean evidence: CFO has run 1.8× net income on a 3-year stack and receivables grew 2% against 34% revenue growth, so headline strength is not being built on credit terms. The principal yellow flags are not in the operations — they sit in tax, "other income," and the inventory-reserve reversal, all of which boost FY2025 GAAP earnings in ways unlikely to repeat in FY2026.

The 13-shenanigan scorecard

Zero red, three yellow, ten green. The yellows cluster around a single theme: FY2025 headline GAAP earnings benefit from low-quality items (tax-position release, reserve reversal, mark-to-market gains) and from a one-time CFO boost from the seven-month ZT Manufacturing round-trip. None constitute manipulation; all reduce the run-rate signal in the reported numbers.

The three things flattering FY2025 earnings

Three discrete items added an estimated $1.5–2.0B of after-tax benefit to GAAP earnings that are unlikely to repeat. Stripped out, FY2025 net income converges toward the underlying run-rate rather than the reported $4.27B.

The $853M tax benefit is the single largest item: in April 2025, the IRS granted reasonable-cause relief on dual consolidated losses, allowing AMD to release uncertain-tax-position reserves. The management discussion is explicit that this is a one-time effect — the FY2026 outlook in the Q4 call assumes a 13% non-GAAP effective tax rate, vs the FY2025 GAAP rate of negative 2.5%. The Q2 MI308 inventory was written down $800M; Q4 brought back $360M as licenses came through. AMD's own critical-accounting language acknowledges this asymmetry: "if in any period we are able to sell inventories that had been written down… related revenue would be recorded with a lower or no offsetting charge to cost of sales resulting in a net benefit to our gross margin." The $577M "Other income" line was up $396M YoY on unrealized gains in the private-investment portfolio — a real GAAP item but one with no economic linkage to AI/CPU operating performance and zero predictability.

Cash flow quality — strong but the headline overstates it

CFO of $7.71B in FY2025 looks exceptional, and most of it is real, but two adjustments matter. First, $1.22B of that CFO comes from discontinued operations — ZT Manufacturing was held inside AMD for roughly seven months before being sold to Sanmina, and its operating cash inflow during that window flows through consolidated CFO. Second, the cash invested to buy ZT ($2.0B net of cash acquired) hits investing activities, while the proceeds from the sale ($1.4B net) hit investing activities for discontinued operations. The cleanest economic measure for FY2025 is continuing-operations CFO ($6.49B) minus continuing capex ($1.01B) = $5.48B underlying FCF, and FCF after all acquisitions of $4.94B.

The walk matters because the headline gap between FY2024 FCF ($2.41B) and FY2025 FCF ($6.70B) implies 178% growth, while the underlying acquisition-adjusted figure ($1.86B → $4.94B) is 166% growth — still excellent, but should be the benchmark for FY2026 modeling, not the headline.

Receivables, inventory, and the working-capital signal

Receivables growth running well below revenue growth is the single most positive forensic data point in the FY2025 file. DSO compressed from 75 days at FY2024 year-end to 66 days at FY2025 year-end, indicating customers paid faster, not slower — the opposite of channel-stuffing pattern. Inventory days, however, tell a different story: DIO is 142 days, the highest in our 10-year series, driven by an intentional build to support MI350 ramp and advanced-node CPUs. That is operationally rational but creates write-down risk if AI demand softens or if export controls bite again.

The FY2024 spike in AR growth (43% vs revenue +14%) was the single ugliest period — and it was concentrated in Q4 FY2024, where the management discussion says "$1.9 billion increase in accounts receivable driven primarily by higher revenue in the last month of 2024." The pattern reversed in FY2025, which is what a non-manipulator looks like: when collections catch up, they catch up. If management were stretching, FY2025 AR would have grown with revenue or worse.

Goodwill, intangibles, and the Xilinx legacy

The post-Xilinx balance sheet (Feb 2022 close, $48.8B all-stock deal) carries $25.1B of goodwill and $16.7B of net intangibles. That is 54% of total assets in soft form — material exposure to a future impairment trigger if the Embedded segment (Xilinx's home) continues to underperform. The Embedded segment posted a 3% revenue decline and a $178M operating-income drop in FY2025.

AMD's impairment test is "qualitative" rather than quantitative — the company concluded the carrying value of each reporting unit does not exceed fair value, but did not publish headroom calculations. This is standard practice and PCAOB-compliant. It is also where the next material write-down, if one ever comes, will surface. Embedded operating income of $1.24B against an estimated allocated soft-asset base of $10B+ implies very modest cushion. A 20% decline in Embedded long-term cash-flow expectations could trigger a quantitative test.

Non-GAAP hygiene — material gap, recurring exclusions

AMD reports both GAAP and non-GAAP results. The non-GAAP gap is unusually wide and most of it is not genuinely non-recurring.

The $2.3B amortization of acquired intangibles is real economic cost — AMD paid $48.8B for Xilinx and the intangibles amortization is part of the recovery of that price. Excluding it makes adjusted earnings look stronger than the deal economics support. The $1.64B SBC is a recurring real cost — the equity granted in FY2025 will dilute shareholders unless offset by ongoing buybacks, and AMD's net dilution (SBC $1.64B vs buybacks $1.92B) is roughly balanced in FY2025 but was negative in earlier years. The published non-GAAP EPS of $4.17 should not be the anchor for valuation work; building a "core" EPS that adds back only the truly one-time items (DCL release, MI308 reversal, mark-to-market) is more useful and lands roughly between GAAP $2.67 and non-GAAP $4.17 — call it $3.20–3.50 of underlying run-rate FY2025 EPS.

The OpenAI warrant — pending dilution and revenue-recognition complexity

In October 2025, AMD issued OpenAI a warrant for up to 160 million shares (~10% of share count) at a $0.01 strike, vesting against deployment milestones for the MI450 series and AMD stock-price targets. None vested in FY2025; AMD says the warrant had no impact on FY2025 financials. Two forensic considerations follow.

First, ASC 606 generally requires share-based consideration paid to a customer to be recorded as a reduction of revenue once vesting becomes probable. As deployment milestones approach, AMD will likely recognize a non-trivial revenue reduction (or contra-revenue charge) tied to the vested portion. The accounting will be GAAP-compliant but optically painful — non-GAAP could exclude it, widening the GAAP vs non-GAAP gap further. Second, the 160M-share maximum dilution against a ~1.63B float is roughly 9.8% — a meaningful future GAAP EPS drag if the program scales fully. Nothing here is shenanigan; it is, however, a known future earnings-quality headwind worth pricing today.

Stock-based compensation cadence

SBC at $1.64B in FY2025 is 4.7% of revenue and 21% of CFO. Buybacks ($1.92B) just offset SBC dilution in FY2025, but the trend on net dilution has been mixed across the post-Xilinx period.

Net share repurchase is roughly neutral relative to SBC at recent run-rate, but if the OpenAI warrant begins to vest while buybacks stay around $1.5–2.0B/year, net dilution will turn negative. Underwriting should model this explicitly.

Breeding ground — governance and incentive read

The breeding-ground signals are mixed-to-favorable and dampen rather than amplify the yellow flags above.

The structural conditions for accounting manipulation are mostly absent: independent audit committee with strong financial credentials, no related-party transactions, no auditor issues, no reverse-merger or shell-company history. The two yellow tints — combined Chair/CEO and equity-heavy comp tied to non-GAAP — are common features of every other US semiconductor mega-cap, not anomalies. The breeding ground here does not raise the probability that the three yellow scorecard items mask something more serious.

Peer-relative read

AMD's cash-conversion ratios sit in the healthy middle of the peer set — better than NVDA (which holds heavy operating profit close to net income), worse than MRVL/INTC (whose CFO/NI ratios are flattered by depressed net income). Soft-asset weight at AMD (54%) is lower than AVGO (78%) and MRVL (69%) but materially higher than NVDA (~4%) — reflecting the Xilinx acquisition. This is a sector-normal profile, not an outlier.

What to underwrite next

The work to track over the next four quarters, in order of materiality to the forensic verdict:

What would downgrade the grade (toward "Elevated"): any FY2026 reversal of a previously released tax position, a quantitative goodwill impairment on the Embedded reporting unit, a new restricted-customer write-down without disclosure of the offsetting reserve mechanics, or insider sales by Su or Hu accelerating into a positive guidance event.

What would upgrade the grade (toward "Clean"): FY2026 effective tax rate landing at the 13% guide without further one-time items, inventory days normalizing toward 110–120, no contra-revenue surprise on the OpenAI warrant, and continuing-operations CFO/NI staying above 1.5×.

Bottom line for the PM. The accounting risk is not a thesis breaker. It is also not a footnote. The right place for it is in earnings normalization — model FY2025 underlying earnings at roughly $3.20–3.50 of "core" EPS rather than the reported GAAP $2.67 or the reported non-GAAP $4.17, and benchmark FY2026 against that base. The risk does not warrant a position-size cap, but it does warrant skepticism of the headline FY2025 growth rate when projecting FY2026 — much of the upside in the GAAP line was tax, not operations.

---

People & Governance

Verdict (B). AMD is a widely-held, professionally-managed company with a technically deep board, robust pay-for-performance structure, and clean related-party record — but two real frictions: (1) Dr. Lisa Su has been a one-way insider seller (78 disposals worth $184M across 2025–2026, zero open-market purchases), and (2) Chair and CEO roles are combined. The Lead Independent Director (Nora Denzel) and a 7-of-8 independent board mitigate but do not eliminate the entrenchment risk a 12-year combined-role CEO carries at all-time highs.

Who runs AMD

The bench around Su is unusually deep for a chip company of this scale. Papermaster's CTO seat (since 2011) has spanned the entire Zen architecture run; Norrod owns the data-center business line that delivered the MI300/MI350 ramp; Hu brought a clean Marvell-to-AMD CFO transition in 2023; Grasby got a one-time promotion/retention award in February 2025, signaling the board is paying to keep the commercial engine intact. AMD also added Phil Guido as Chief Commercial Officer in 2024 (the proxy notes a single late Form 4 filing on his initial open-market AMD purchase — administrative, not substantive).

The single-point-of-failure risk is real. AMD has been a Lisa Su story since 2014 — the stock has compounded roughly 100×. Su is 56 and shows no signs of leaving, but with a CEO this central to the equity story, succession depth is what protects shareholders. Norrod and Papermaster are credible internal heirs; both are 60+, which means the next COO/CEO pipeline is also worth tracking.

Compensation — heavy equity, performance-tied, but very large

The structure is conventional best-in-class: cash salary under 3% of total, the rest in PRSU (3-year TSR vs S and P 500 percentile), time-vesting RSU, and options. The 2025 Executive Incentive Plan paid out at 121% of target on financial milestones; the 2022-vintage PRSU paid at 96.84% of target because AMD's three-year TSR (53.55%) trailed the index (55.13%) by a hair — a real failure of the metric to pay 100%, which is reassuring. The board also retooled the 2025 PRSU design to a percentile rank within the index (a cleaner relative measure) and held the change-of-control terms at 2× cash with double trigger.

Where to push back: the absolute size. $55M to the CEO of a company whose three-year TSR essentially matched the index is hard to reconcile with "rigorous" pay-for-performance — that grant was earned on the future outcome, but the current package is benchmarked to peers (NVDA, Broadcom, Qualcomm) whose CEOs are themselves at extreme pay levels. Stockholders have signaled support (Say-on-Pay passed comfortably in 2025 per the proxy's engagement summary), so this is a tolerance issue, not a governance failure.

The alignment question — insider behavior matters more than ownership %

This is where the page gets uncomfortable.

181 Section 16 transactions across the past 18 months. Zero open-market purchases by anyone. Su accounts for ~75% of disposed value, with 9 gift transactions (charity / family trust) on top of 78 outright sales. Almost all sales appear to be 10b5-1 plan executions (predictable, pre-scheduled, defensible), but the pattern — a CEO whose entire compensation comes in stock, who sells everything that vests and never buys with the cash proceeds — is what it is. There is no insider conviction signal here at all-time-high stock prices.

Context that softens the picture:

• AMD's equity-heavy comp design makes selling rational — virtually 100% of after-tax compensation arrives as stock, so executives must sell to convert to cash to live.
• 10b5-1 plans are standard practice and remove timing-discretion liability.
• Su still owns ~4.3M shares (~$1.9B at recent prices); she has not sold to the point of breaching the 6× base-salary ownership requirement.
• AMD has an anti-hedging policy and prohibits pledging without committee approval (none has been granted).

Context that hardens the picture:

• The volume accelerated in 2026 (100 sales / $216M) vs. 2025 (13 / $29M) — a clear 7× step-up.
• Other NEOs (Papermaster, Grasby, Norrod) also sold materially with no buys.
• This is the one governance data point most likely to show up in a critical sell-side note.

The Board — independent, technically credible, but old-AMD heavy

The board has two genuinely strong technical voices — Su herself and Abhi Talwalkar (ex-LSI CEO, current Lam Research Chair) — both of whom can hold management's feet to the fire on roadmap and process choices. Marren's Temasek/TPG capital-allocation lens is the right voice for M and A discipline as AMD digests Xilinx, ZT Systems, and Pensando. Householder's audit chairmanship is unimpeachable (ex-PwC partner, ex-Sempra CFO/COO). McClure (joined Jan 2026, ex-Accenture CFO of 37-year tenure) added a second senior-finance voice; this was a thoughtful refresh, not a token addition.

Two structural critiques:

1. Tenure overhang. Denzel and Householder have been on the board since 2014, the same year Su took over. The board that hired her is the board that still oversees her. The 2026 retirement of Jon Olson (a 2008 director) starts to refresh this; one more long-tenured rotation would be welcome.
2. No standing AI ethics / national-security committee. AMD is now the #2 frontier-AI compute supplier with material China-export exposure (BIS controls, MI300/MI308 customer mix). The Innovation & Technology Committee covers the roadmap; the Audit committee covers regulatory compliance; nothing on the board sits between the two on AI-policy strategy. A capable board of this caliber should formalize that seam.

Ownership — widely-held, no control block

There is no founder block, no promoter, no controlling shareholder. The top-5 holders are entirely passive index funds. Net effect: AMD's accountability mechanism is the market plus index-fund stewardship plus an activist threat that has not (yet) materialized. That is a perfectly normal large-cap structure; it is not a structure that produces founder-level long-term thinking unless the CEO supplies it personally — which, to give Su credit, she has.

Insiders as a group hold under 0.5% of float. Su's own ~4.3M shares are worth roughly $1.9B at recent prices — a vast personal stake in absolute terms, a small one in voting terms. The 6× base-salary ownership guideline ($7.75M minimum) is met >200× over.

Governance hygiene — clean on the formalities

Formally, the housekeeping is excellent. The one late Form 4 (Guido) was disclosed and explained; the proxy notes no related-party transactions met the $120K SEC threshold in FY25; clawback and severance terms are at or above best-practice. There is nothing here that would draw an ISS adverse recommendation.

What would change my grade

Bottom line. AMD's people story is strong on capability and clean on formal governance, weaker on two things that matter most to outside shareholders during a euphoric tape: a CEO selling rather than buying, and the same person serving as Chair and CEO. Neither is a vote-against signal. Both are reasons to watch insider behavior more carefully than a standard semi governance template would suggest.

---

How the Story Changed

AMD's history is an eleven-year arc with the same CEO. Lisa Su took the seat in October 2014 with the company on its hospital bed — three losing CEOs in six years, $2.5B in debt, revenue collapsing, stock near $3. Eleven years later AMD is a $34.6B-revenue, ~50%-gross-margin AI infrastructure company shipping a 6-gigawatt deal with OpenAI and a 6-gigawatt deal with Meta. Nothing on this page was inherited — every revenue, margin, and credibility point was built under the current team. The narrative shift since FY2021 matters: management has executed every near-term commitment of consequence but has steadily replaced precise dollar guides with multi-year aspirational framing as the stakes have grown.

The Decade in One Chart

Three discrete chapters live inside one CEO tenure: a near-death triage (2014–2016), the Zen-led rebuild (2017–2021), and the post-Xilinx pivot into AI infrastructure (2022 onward — accelerating sharply from late 2023). The current strategic chapter — Instinct/Helios as a top-line AI franchise — really starts the day MI300 ships in volume in late 2023 and is fully formed by the OpenAI announcement in October 2025.

Chapter 1 — The Hospital Bed (2014–2016)

The business Lisa Su inherited was failing on every dimension. Revenue fell from $5.5B in FY2014 to $3.99B in FY2015. Gross margin collapsed to 23–27%. Operating losses totaled roughly $1B across 2014–2016. The PC business was being gutted by Intel; the server share was rounding to zero; the foundry (GlobalFoundries) had been spun off in 2008; and the company carried $2.5B of legacy debt against a stock the market had nearly written off.

This chapter sets the bar for everything that follows: the business was unambiguously not high-quality when current leadership arrived. Every share gain, every margin point, every AI dollar today was built — not inherited. That fact governs how the rest of the page should be read, and it governs the capital-allocation thesis the Long-Term view will build on.

Chapter 2 — Zen Earns Trust Back (2017–2021)

The Zen architecture, taped out under Mark Papermaster and Lisa Su's leadership, broke cover with Ryzen (consumer CPU) in March 2017 and EPYC (server CPU) in June 2017. The financials inflect immediately: FY2017 returns to operating profit, FY2018 doubles operating income, FY2019 launches Zen 2 on TSMC's 7nm — overtaking Intel's process node for the first time in AMD's history — and FY2021 closes with revenue up +199% over FY2014 and operating income at $3.65B.

The credibility built in this period is the foundation everything since rests on. Su's original turnaround promise — a competitive x86 architecture, a path back to server share, and a return to gross margins above 45% — was delivered in full by 2021. No serial-promiser ever recovered from a five-year run that good. The bar for what investors will take on faith today was earned here.

Chapter 3 — Xilinx, Pensando, and the Wait for AI (2022–2024)

The post-Zen chapter is harder to read cleanly because two stories run at once: a successful diversification on paper (Xilinx closed Feb 2022 for $49B in stock, Pensando May 2022 for $1.9B in cash, Silo AI Aug 2024) and a brutal cyclical drawdown in three of four segments. Embedded — the Xilinx franchise — fell roughly 33% in FY2024 versus FY2022; Gaming peaked in FY2022 and was down 58% by FY2024; Client wobbled with the PC inventory unwind in late 2022. Operating income compressed from $3.65B (FY2021) to $0.40B (FY2023) before the AI franchise lit up.

What carried the chapter was MI300. The hyperscaler design wins were already disclosed in the FY2023 10-K — "large hyperscaler customers committed to deploy" the MI300 — and by Q2 FY2024 Data Center GPU revenue cleared $1B in a single quarter. Management raised the FY2024 Data Center GPU dollar guide three times in three quarters: $4B → $4.5B → $5B+, and delivered every raise.

Less clean: the embedded recovery promise. In Q2 FY2024, Lisa Su called Q1 "the bottom" and promised "gradual recovery" in the second half. Embedded did not turn positive year-over-year until Q4 FY2025 — six quarters later than the implied turn. Investors gave the embedded miss a pass because the AI franchise was hauling everything. But it is the cleanest example in the dataset of management optimism unpunished only because something bigger went right.

Chapter 4 — The AI Infrastructure Pivot (2025– )

The story that is now central to AMD did not exist in this form even a year ago. Three things happened in 2025 that re-cast the company:

1. The dollar guide was retired. Starting with the Q4 FY2024 call (Feb 2025), management stopped giving a specific Data Center GPU revenue number for the year. The new framing is "from more than $5 billion in 2024 to tens of billions of dollars of annual revenue over the coming years." Analysts pushed back hard — Vivek Arya called it out directly. Management held the line and replaced the dollar with a 60%+ data center CAGR and a $500B 2028 accelerator TAM.

2. The MI308/China shock — and the reserve reversal. A new April 2025 export control restriction killed MI308 sales to China, triggering an $800M inventory and related charge in Q2 FY2025 (gross margin fell to 43% headline). Then in Q4 FY2025, licenses were partially granted; $390M of MI308 shipped and AMD released $306M of the inventory reserve as a margin tailwind. Net of the reserve release and China revenue, Q4 FY2025 gross margin would have been ~55%; reported was 57%. The handling was technically transparent but commercially opportunistic — a credibility-neutral episode.

3. OpenAI and Meta. In October 2025 AMD announced a 6-gigawatt Instinct deployment with OpenAI ("well over $100 billion in revenue over the next few years"), backed by an OpenAI warrant for up to 160M shares at $0.01 strike vesting on purchase milestones and AMD stock-price targets. Then in the Q1 FY2026 call AMD announced a second 6-gigawatt Meta deal spanning multiple Instinct generations. The "tens of billions" gets a year attached: 2027. The CPU TAM is revised from $60B to $120B by 2030.

Narrative Drift — What Management Stopped Saying

The strongest evidence of how the story has shifted is what's no longer in the disclosure language. Reading the 10-K risk factors across FY2021–FY2025 in sequence, the lead risk silently changes from "Intel" to "Nvidia"; COVID disappears entirely after FY2023; CHIPS Act competitive disadvantage appears in FY2023 and is gone by FY2025; and an entirely new cluster of risks — China export controls, AI customer concentration, AI customer ability to secure capital and energy for data-center buildout — moves to the top of the deck.

The heatmap reads like a chip-stack diagram of the company's evolving worldview. The Xilinx integration story has been quietly absorbed (the phrase "Xilinx synergies" effectively disappeared from earnings calls after Q1 FY2025; the products are now pitched on their own merits as Versal / Alveo). The Pensando "DPU synergy" pitch is similarly gone. The Silo AI acquisition (Q3 FY2024) is barely named after one quarter. Meanwhile Helios — a brand that did not exist in any disclosure before Q2 FY2025 — is now the dominant data center framing in five quarters flat. That speed of brand consolidation is unusual and is a positive credibility signal: management names things and ships, rather than naming things and forgetting them.

What also changed: management has stopped using the word "discipline" and started using the word "opportunity" in capital-allocation language. The buyback authorization was expanded to $14 billion in late 2025 and the pace of repurchase doubled in FY2025 — coupled with the OpenAI warrant and $12.2B of unconditional purchase commitments (multi-year cloud capacity), capital deployment has shifted decisively from defensive to offensive.

The Promise / Delivery Track Record

The receipts that matter. Each row is a specific, valuation-relevant promise — large enough that a buy-side reader can verify it.

The headline ratio on decided promises is 6 of 10 cleanly kept, but the texture matters more than the count. Every miss has a common pattern: management was optimistic about a recovery timing (embedded soft landing, H1 2025 DC, China MI308) and conservative about a core product ramp (Zen, EPYC share, MI300 GPU dollars, MI350 schedule). The product ramps have been delivered every time. The cyclical recoveries have run late, every time. Bet on what AMD builds; don't bet on what AMD says about end-market timing.

Credibility Verdict

Score: 8 / 10. The verdict rests on three asymmetric pieces of evidence:

• The base rate is exceptional. Lisa Su delivered the most successful CPU architecture comeback in semiconductor history. Revenue ~6x, gross margin ~45 percentage points off the trough, debt-to-EBITDA collapse from distressed to investment grade. This is not a young narrative looking for trust — it is an eleven-year compounder of credibility.

• The retail-grade misses are honest. The embedded "gradual recovery" promise was kept reframing itself for six quarters with no excuses; the MI308 China handling included a transparent $800M write-down before any later reserve release; the H1 2025 DC "consistency" promise was acknowledged as broken by the export-control story rather than spun away.

• The risk is forward, not backward. The 2027 "tens of billions," the >60% DC CAGR, the $120B 2030 CPU TAM, and the OpenAI/Meta gigawatt deals are the largest concentrated set of forward commitments in the company's history. None can be checked yet. The dollar-guide retirement after Q4 FY2024 was a deliberate choice to trade quarterly accountability for narrative flexibility — a signal that management is now in capture mode on the upside and does not want to be boxed in by a number. Worth watching.

What to Believe vs. Discount

Believe:

• The company can build product. Annual cadence promises (MI300 → MI325 → MI350 → MI355X → MI400 → MI500) have been hit or pulled forward, never slipped. Helios is real because ZT Systems is real and the rack ships in 2H 2026.
• The OpenAI and Meta gigawatt commitments are anchored in actual purchase contracts and (in OpenAI's case) a vesting warrant — not slideware.
• Gross margins above 50% are now structural, not cyclical: mix is moving toward Data Center every quarter.

Discount:

• The "tens of billions in 2027" framing is aspirational until the first MI450 gigawatt deploys in 2H 2026 and the first OpenAI revenue prints. Until then it is a target, not a number.
• Cyclical recovery timing (embedded, gaming, semi-custom). The Q4 FY2025 call already warned of "seventh year of console cycle" and double-digit semi-custom decline expected in 2026. Embedded design wins (now $50B+ since the acquisition) are real but conversion has consistently lagged.
• The MI308 China line — small but volatile and increasingly political; MI325 China licenses are now in the application pipeline and the outcome is unknowable.
• "AI customer power / capital risk" is now a top-five risk factor in the 10-K. That is unusual language for a supplier — it is an implicit acknowledgement that hyperscaler buying power and AI-customer financing capacity could become the binding constraint, not GPU supply.

The story today is simpler than the diversification chapter of 2022–2023 and more stretched than the Zen chapter of 2017–2021. It is essentially one bet — that AMD is the credible second source in AI infrastructure, with a CPU franchise underneath that funds the bet. The credibility built over the prior decade is the reason a $300B+ market capitalization will take that bet at face value. The credibility built over the prior decade is also exactly what would be at risk if 2027 disappoints. That asymmetry is the central feature of the AMD story now.

---

Financials — What the Numbers Say

AMD is no longer a turnaround story. After two decades of fab divestitures, restructurings, and tax-loss carryforwards, AMD ended FY2025 with $34.6B in revenue, $6.7B in free cash flow, $10.6B of cash, and just $3.2B of debt — a fabless design house whose balance sheet now resembles a software company more than a chipmaker. The single number that defines the investment debate is the gap between GAAP operating margin (10.7% in FY2025) and the AI-accelerator operating margin the market is paying for in FY2026–27 consensus — every multiple on this page depends on closing that gap.

The Thirty-Second Verdict

Reader glossary — "GAAP" is the U.S. accounting standard; reported earnings under GAAP include large non-cash charges from the Xilinx purchase amortization. "FCF" (free cash flow) is operating cash flow minus capex — the cash actually available to shareholders. "SBC" is stock-based compensation, a non-cash expense that dilutes shareholders. "Net cash" is cash minus all debt — a positive number means the company has more cash than debt.

How To Read This Business

AMD is a fabless semiconductor designer — it designs chips and outsources fabrication to TSMC. Four reportable segments split the revenue: Data Center (EPYC server CPUs, Instinct MI-series AI accelerators, Pensando DPUs), Client (Ryzen PC CPUs), Gaming (semi-custom chips for PlayStation and Xbox, plus Radeon discrete GPUs), and Embedded (the legacy Xilinx FPGA business, plus auto/industrial). Roughly half of FY2025 revenue and the overwhelming majority of profit-margin upside come from Data Center, which is the only segment growing at AI-cycle speed. The financials below are read through that lens — the rest of the business funds the AI push.

Three accounting realities follow from that structure and you must internalize them before the multi-year table makes sense:

(1) A massive intangible amortization gap. The Xilinx acquisition closed February 2022 in an all-stock deal valued ~$49B, adding $24.2B of goodwill and $24.1B of intangible assets. Amortization of those intangibles runs roughly $3B per year — depressing GAAP operating income but not cash flow. This is why FY2023 net income of $854M coexisted with $1.7B of operating cash flow.

(2) The Xilinx deal doubled the share count. Diluted shares jumped from 1.23B (FY2021) to 1.57B (FY2022) — a ~28% one-shot dilution against a smaller, growing-faster pre-deal AMD. Per-share metrics before and after 2022 are not directly comparable.

(3) Q2 FY2025 has a one-quarter scar. GAAP gross margin fell to 39.8% and operating margin to negative 1.7% on an ~$800M inventory and related charge tied to MI308 export-control restrictions on China. Margins snapped back to 51.7% and 13.7% in Q3 FY2025 and to 54.3% and 17.1% in Q4 FY2025. Treat Q2 FY2025 as an outlier, not a trend.

The Recent Quarterly Inflection

Revenue stepped from $5.4B in Q2 FY2023 to $10.3B in Q4 FY2025 — a near-doubling in roughly two years, accelerating sharply from Q3 FY2024 as MI300 accelerator shipments and EPYC Turin server CPUs took hold. The exit-rate margin matters more than the quarterly average: Q4 FY2025 operating margin of 17.1% on gross margin of 54.3% is the cleanest read on through-cycle Data Center profitability and is the number every forward valuation is implicitly extrapolating.

The Year-Wise Statements

Three eras are visible in the table. First, the survival era through FY2017 — losses, fragile balance sheet (negative equity in FY2015), and meaningful debt against minimal cash flow. Second, the Zen-driven recovery FY2018–FY2021 — revenue grew from $6.5B to $16.4B, gross margin expanded from 38% to 48%, operating margin reached 22%, and FCF reached $3.2B with returns on capital of 47–67%. Third, the Xilinx era FY2022 onward — the balance sheet ballooned with $54.8B of equity overnight, returns optically collapsed (ROE fell from 47% to 4% mathematically as the equity denominator grew), and GAAP earnings were depressed by amortization. FY2025 marks the first year the post-Xilinx operating model produces both scale (revenue past $34B) and meaningful cash flow ($6.7B FCF) at once — without it, the current multiple has no anchor.

Earnings Quality — Where the Cash Actually Comes From

Two takeaways. (1) Cash beats earnings, by a wide margin, in every Xilinx-era year. FY2022 net income was $1.3B but operating cash flow was $3.6B — the $4.3B of depreciation and amortization (most of it Xilinx intangibles) is the bridge. (2) The conversion is genuine, not a working-capital trick. Across FY2022–FY2025, cumulative net income was $8.2B, cumulative FCF was $13.3B — a 1.6x conversion ratio. Operating cash flow exceeded reported earnings by $4–6B per year because the non-cash D and A was real economic cost paid once, in Xilinx-deal stock, four years ago.

Diluted share count actually went up from 1.62B to 1.62B (essentially flat) between FY2023 and FY2025 despite $4.9B of cumulative buybacks — the buybacks are offsetting SBC and vesting equity, not retiring net shares. This is a real capital cost the headline FCF figure does not show. Net share retirement is at most a minor tailwind, not a per-share growth driver, at current SBC levels.

Balance Sheet — A Weapon, Not A Constraint

AMD exited FY2025 with $10.6B of cash against $3.2B of total debt — a $7.3B net cash position. Debt-to-equity is 5.1%, the current ratio is 2.85x, and interest coverage on the residual debt is effectively meaningless because operating income is roughly 100x interest expense. External research describes the synthetic credit rating as investment-grade-strong; the practical read is that AMD could finance any reasonable strategic acquisition out of cash or with negligible refinancing risk. The balance sheet is a strategic weapon, not a constraint — and that distinction matters because the closest competitor on AI accelerators is Nvidia and the closest custom-silicon competitor is Broadcom, both of whom carry less optionality on incremental M and A.

The flip side: most of the equity is goodwill and intangibles. Of $63.0B of book equity, $25.1B is goodwill and $16.7B is identifiable intangibles — together 66% of equity. If the AI thesis impairs, an Xilinx-style writedown is a balance-sheet risk that does not affect cash but does compress reported book value and could trigger covenant scrutiny on any future debt.

Returns on Capital — The Mathematical Trap

Why ROIC dropped from 67% to 4.6% in one year. It did not. FY2022 returns optically collapsed because AMD issued $49B of stock for Xilinx — invested capital exploded overnight while the new operating income contribution had not yet materialized. The accurate read of post-Xilinx returns starts at FY2023 (a trough year hit by both a PC downturn and full-year Xilinx amortization) and trends upward: 1.2% → 2.7% → 6.6%. The pre-Xilinx Zen-era returns of 47–67% ROE are achievable on the standalone EPYC/Ryzen/Radeon business; the new question is what returns the combined business produces at AI-accelerator-scale revenue.

The honest read of FY2025 returns: a 6.6% ROIC with the largest intangible base in the company's history, in the first year the AI ramp is meaningfully in the numbers. That number triples if Data Center operating margin reaches Nvidia-like levels. It plateaus if MI-series accelerators stay at competitive-but-not-dominant share. The valuation requires the first outcome.

Capital Allocation — Where The Cash Has Gone

Three observations:

Capex stays under 3% of revenue. FY2025 capex was $1.0B against $34.6B of revenue — the fabless advantage. AMD does not need to build fabs; TSMC does that, and AMD pays for capacity inside cost of revenue. This is the structural reason FCF margin (19.3% in FY2025) is so high compared to Intel's.

M and A is back to being a real line item. $1.76B in FY2025 acquisitions (ZT Systems for rack-scale AI infrastructure, plus smaller AI software deals) marks a return to inorganic strategy after the Xilinx digestion. Watch this — incremental ROIC on these deals is the cleanest read on capital-allocation discipline.

Buybacks have done nothing to shrink the share count. $9.4B of cumulative buybacks since FY2021 against essentially flat diluted shares. The capital is being returned, but it is being used to absorb SBC, not to compound per-share value.

No dividend is paid and management has shown no inclination to start one — appropriate for a growth-stage AI compute leader, but a structural ceiling on the multiple any income-oriented investor will assign.

Peer Comparison — How AMD Stacks Up

The peer table tells a precise story:

AMD sits between two business models the market values very differently. Nvidia is the price-maker — 62% operating margin, 76% ROE, the only peer where forward P/E (25x) actually looks cheap on its own earnings power. Broadcom's combined custom-silicon plus VMware software franchise prints 40% operating margins. AMD's 10.7% GAAP operating margin (~13% Q4 exit rate) and 7.1% ROE are far closer to a competitive merchant semi than to a category leader, yet the market is paying 21.7x EV/Sales — closer to Broadcom's 34.9x and well above Qualcomm's 5.2x. The implicit forecast is margin expansion and sustained 40%+ revenue growth.

The balance sheet positioning is the cleanest peer differentiator. AMD's $7.3B net cash is best-in-peer-group apart from Nvidia's. Intel's $32B of net debt and Broadcom's $49B (post-VMware) constrain their capital-allocation flexibility; AMD's does not. This is the financial backstop for the AI thesis if execution stumbles.

Forward P/E says the market thinks AMD looks more like Nvidia than Marvell. AMD at 68.8x next-twelve-months earnings versus Nvidia at 25x and Marvell at 70x reveals the market is paying for an in-progress reacceleration. If FY2027 EPS hits the $13.10 consensus, the multiple compresses to 39x — still expensive, but a defensible number for a profitable AI compute platform with no leverage.

Valuation — What the Stock Already Assumes

Three reference points define the valuation debate. Trailing FY2025 metrics — 21.7x EV/Sales, 112x EV/EBITDA, 175x trailing P/E — are indefensible in isolation. Forward consensus is what the market is actually paying for: FY2026 P/E of 69x and FY2027 of 39x, against revenue growth of 43% and 54% respectively. Compared with AMD's own history, the current trailing P/E of ~190x looks extreme, but the multiple is right in line with the company's own recent past (P/E was 278x in FY2023 and 119x in FY2024) — what is different now is that the cash flow is finally there to validate it.

External research notes the EV/EBITDA of 102x trades at roughly 129% above its 10-year median (~45x) and 261% above the semiconductor industry median (~28x). The valuation does not break unless Data Center growth or operating margin disappoints — but it does not work unless they continue to surprise to the upside.

What The Financials Confirm and Contradict

Confirmed: AMD has become a genuinely high-quality fabless semi — the FY2025 financials show 49.5% gross margin, 19.3% FCF margin, $7.3B net cash, and a credible step-up to 17%+ exit-rate operating margin. The cash quality is real (D and A is non-cash Xilinx amortization, not deferred capex), the balance sheet is a strategic asset, and the capital intensity is structurally low.

Contradicted: The narrative that AMD is "underpaid for its AI exposure" does not survive the trailing or even forward multiples. EV/Sales of 21.7x and a forward P/E of 69x do not price in any execution risk. ROIC at 6.6% remains lower than the cost of equity any reasonable investor would assign — which means the current operating business does not yet earn an excess return on the capital that built it.

The one or two swing factors are simple to name and observable each quarter. (1) Data Center segment revenue growth — the closest single number that tells you whether the AI thesis is on track. (2) Total company GAAP operating margin — the cleanest read on whether the AI revenue is incremental margin or replacement-level margin. The first matters because consensus assumes 43%+ overall growth; the second matters because every dollar of revenue at 25%+ operating margin compresses the forward multiple at the rate the bull case requires.

The first financial metric to watch is quarterly GAAP operating margin trajectory. Q4 FY2025 hit 17.1% and Q1 FY2026 came in at 14.4%. If FY2026 prints at 17–18% on a full-year basis, the forward earnings ramp lands and the multiple supports 40x FY2027 EPS. If margin stalls at 12–14%, FY2026 EPS misses the $7.37 consensus and the forward multiple compresses.

---

What the Internet Knows About AMD

Bottom line up top. The web extends the filings rather than refuting them. Three facts not yet absorbed by consensus: (1) the OpenAI deal carries a 160-million-share penny warrant (~10% of AMD) with vesting gated on stock-price hurdles escalating to $600/share, and ASC 606 will likely treat the fair-value grant as contra-revenue once vesting becomes probable — compressing reported Instinct ASPs as soon as MI450 begins shipping in 2H 2026. (2) Management confirmed on the Q1 2026 call that MI450 ramps below corporate-average gross margin — the hyperscaler wins that drive the bull case are also the margin headwind for 2026–27. (3) Consensus 12-month price target (~$486) now sits below the ~$507 print, even after a parade of upgrades; the stock has run past the Street, the marginal upgrade is harder to find, and insiders sold ~$158M in 90 days with zero open-market buying. The forensic file is clean (no SEC matter, no restatement, E&Y unbroken since 1970, 10b5-1 mechanics on Lisa Su). This is a priced-for-perfection setup: the binary swing factor is MI450/Helios execution in 2H 2026 and the underappreciated drag is the customer-equity accounting structure.

---

The Ten Findings That Matter

1. OpenAI's 160-million-share penny warrant — the accounting overhang consensus has not priced

Per AMD's 8-K filed Oct 6, 2025, OpenAI received a warrant for 160 million shares at a $0.01 strike — close to 10% of AMD on a fully-diluted basis — with tranched vesting tied to both (a) 6 GW of Instinct deployment milestones (first 1 GW shipping 2H 2026) and (b) AMD stock-price hurdles that escalate to $600/share for the final tranche, expiring Oct 5, 2030. Britannica corroborates the deal; the OpenAI press release confirms the strategic-partnership structure.

The piece sell-side write-ups under-cover: under ASC 606 (and the FASB's ASU 2025-04 clarification surfaced in the forensic file), share-based consideration to a customer is treated as noncash consideration payable to a customer — a reduction in transaction price, i.e. contra-revenue — once vesting becomes probable. Translation: as soon as OpenAI's first 1 GW deployment is on a probable path, AMD begins recognizing fair value of the relevant tranche as a reduction of Instinct revenue. This is on top of the share dilution.

• Source: https://ir.amd.com/financial-information/sec-filings/content/0001193125-25-230895/d28189d8k.htm ; https://openai.com/index/openai-amd-strategic-partnership/
• So-what: Bull models project Instinct revenue at sticker. The contra-revenue cuts ASPs and reported gross margin precisely when the AI ramp is supposed to inflect — a known unknown that will dribble into reported margins through 2026–27. Combined with the $600 share-price gate, the warrant simultaneously caps near-term dilution (only fully painful if the stock rips) and pulls down margin (whenever the deal works).
• Priced in? Partially. The market has the deal but not the accounting mechanics; the sell-side commentary that does mention it (Simply Wall St) flagged it as the reason analyst fair-value-per-share estimates rose only modestly ($213.89 → $225.00) despite raising revenue.
• Importance: Red flag (structural).

2. Meta's 6 GW deal doubles the bet — at management-confirmed sub-corporate gross margin

Announced Feb 24, 2026, Meta committed to 6 GW of custom MI450-architecture GPUs plus 6th-generation EPYC "Venice/Verano" CPUs deployed on AMD's Helios rack platform — first 1 GW shipping 2H 2026, alongside OpenAI's first 1 GW. On the Q1 2026 call (May 5, 2026), CFO Jean Hu was explicit that MI450 will be below corporate-average gross margin during ramp; a Seeking Alpha analyst, summarizing the call commentary, projects no meaningful margin expansion for the next 1–2 years. The "dollar-per-gigawatt expands with each generation" language management uses is the offset, but it is a future claim, not 2026.

One unverified secondary source (abit.ee) claims Meta also received a ~160M-share performance warrant similar to OpenAI's; the AMD IR release does not confirm this. Worth watching but not bankable yet.

• Source: https://ir.amd.com/news-events/press-releases/detail/1279/amd-and-meta-announce-expanded-strategic-partnership-to-deploy-6-gigawatts-of-amd-gpus
• So-what: The two biggest GPU deals in AMD's history both dilute gross margin on the way up. The "tens of billions" of revenue scale is real; the margin compression is also real. Combined with OpenAI's contra-revenue path, gross margin trajectory is a 2026–27 headwind that the >35% non-GAAP operating-margin target implicitly papers over.
• Priced in? No — bulls model revenue beats; the GM line is where the surprises will land.
• Importance: Yellow/red flag.

3. Insiders sold ~$158M in 90 days with zero open-market buying — at the very top of the chart

Across the executive team, the 90-day insider window through mid-June 2026 shows 372,032 shares sold for $158.66M, against zero open-market buys from any named executive officer (only EVP Phil Guido has bought, ~$500K over 24 months). Specific transactions:

• Lisa Su, CEO — 125,000 shares on Jun 10, 2026 at $460.69 = $57.6M; identical 125,000-share clips on Dec 11, 2025 ($26.9M @ $215.14), Feb 11, 2026 ($26.8M @ $214.36), and May 13, 2026 ($432.90–$457.82). All under a 10b5-1 plan adopted Sept 9, 2025.
• Jean Hu, CFO — Yahoo insider data shows 4,023,581 shares disposed Feb 17, 2026 at $196.78–$205.12 (~$791M–$825M gross; likely includes net-of-vest activity — warrants confirmation but the size dwarfs CEO selling).
• Mark Papermaster, CTO — five tranches totaling ~$29M over 12 months; largest 31,320 shares @ $350 on Apr 24, 2026 ($11M).
• Forrest Norrod, EVP Data Center — 19,487 shares @ $431.40 on May 20, 2026 ($8.4M, ~5.7% of position) plus prior tranches.
• Ava Hahn, General Counsel — 56,814 shares sold Feb 18, 2026 ($11.3M), clustered against Hu's sale 24 hours prior.

Note also that Lisa Su's 2025 total comp jumped to ~$55.2M (Simply Wall St), roughly +78% YoY — a doubling that the May 2026 say-on-pay vote ratified without dissent.

• Source: https://www.marketbeat.com/stocks/NASDAQ/AMD/insider-trades/ ; https://www.stocktitan.net/sec-filings/AMD/form-4-advanced-micro-devices-inc-insider-trading-activity-afaaa69b9622.html ; https://aflcio.org/paywatch/AMD
• So-what: Mechanical 10b5-1 trades insulate against legal risk but don't mute the signal: every NEO is converting equity to cash at multi-year highs; no NEO is putting cash back in. Combined with comp doubling, the insider message is "valuation is full." Not a sell trigger on its own — it is a "don't add at these levels" tell.
• Priced in? Not really — insider conviction is rarely modeled directly.
• Importance: Yellow flag.

4. TSMC CoWoS is the binding constraint — and NVIDIA owns more than half of it

This is where the bull thesis meets the supply curve. Across the moat and competition files: NVIDIA has booked 800,000–850,000 wafers of TSMC CoWoS for 2026, more than half the global capacity; Broadcom and AMD share the remaining 40–50%. TSMC is scaling CoWoS from ~65–75K wafers/month in 2025 toward 120–130K wpm by end-2026, but lines are "fully booked" against ~1M wafers of total 2026 demand. Trefis rates AMD's advanced-packaging infrastructure "worse vs NVIDIA" with "high probability, high impact" — the bear thesis the bulls don't address. JPM commentary cited in the moat file: TSMC 3nm hits a capacity ceiling before 2026, creating a two-year supply gap.

AMD's response — Simply Wall St flagged a bypass attempt via panel-level EFB packaging — is a margin lever but doesn't change the 2026 ceiling.

• Source: https://wccftech.com/nvidia-alone-has-tsmc-advanced-packaging-lines-booked-for-several-years-ahead/ ; https://siliconanalysts.com/analysis/foundry-allocation-status-q1-2026
• So-what: Even if MI450 demand triples vs current "tens of billions" guidance, AMD cannot ship more than its CoWoS allocation permits. The MI450 numbers consensus is now extrapolating are gated by an externality AMD does not control. Order book matters less than allocation.
• Priced in? No — bulls extrapolate from order book; bears under-quantify the cap.
• Importance: Red flag (structural).

5. China: $800M write-down, $360M reversal, 15% revenue-share precedent — and DIO still bloated

The most documented forensic event in the file. Q2 2025: $800M MI308 inventory and related charge under US export controls compressed gross margin from 53% to 43% that quarter; Q4 2025 reversed $360M after partial licenses (net hit $440M). Critically, Lisa Su publicly stated AMD is prepared to pay a 15% revenue-share tax on China AI chip shipments — an unprecedented quasi-tariff that, if codified, becomes a permanent margin drag. Jan 13, 2026, BIS final rule moved MI325X (and NVDA H200) to case-by-case review with a 25% tariff plus required US lab testing for China exports.

Inventory remains the open question: Days Inventory Outstanding ran 150.6 days in Mar-26 vs a 10-year median of ~80. Management frames the build as MI450 pre-build; it could equally be persistent China-restricted overhang. The data alone can't tell — but the magnitude is unusual.

• Source: https://ir.amd.com/financial-information/sec-filings/content/0000002488-26-000021/amd-20251227.htm ; https://www.reuters.com/world/china/amd-chief-says-company-ready-pay-15-tax-ai-chip-shipments-china-2025-12-04/
• So-what: Recurring margin volatility, capped China DC TAM, and a 15% revenue-share precedent that could spread to other restricted product lines. The 2025 write-down/reversal cycle will likely repeat as policy evolves.
• Priced in? Partly — the inventory write-down is in the rear view, but the 15% revenue-share scheme is genuinely not modeled.
• Importance: Red flag (recurring).

6. EPYC owns the cash flow — record 46.2% server CPU revenue share in Q1 2026

The most underrated positive in the file is that the cash-generative part of AMD is firing — and it is, structurally, the EPYC server CPU franchise that funds the GPU bet. Mercury Research Q1 2026: AMD captured 46.2% of server CPU revenue, a record, and is targeting >50% revenue share. AWS, Google Cloud, Microsoft Azure, and Tencent all launched 5th-Gen EPYC instances in Q1 2026. AMD commenced TSMC 2nm production of EPYC "Venice" in May 2026 — Intel's 18A is reportedly in low-to-mid-60% yield range and Panther Lake is in HVM at Fab 52, so Intel is no longer dead, but the next 12 months still favor AMD.

This is also why FY2025 free cash flow exploded: $6.70B FCF (+178% YoY), Q1 2026 alone $2.57B (+253% YoY). Cash $12.4B vs total debt $3.2B — pristine balance sheet, A/A3 credit (Moody's upgraded Feb 2026 with a positive outlook), $10B buyback authorization.

• Source: https://wccftech.com/amd-posts-record-server-cpu-revenue-in-q1-2026-red-team-commands-30-percent-market/ ; https://ir.amd.com/news-events/press-releases/detail/1276/
• So-what: This is the valuation cushion. If the AI narrative cracks, the EPYC cash engine still throws off enough to fund buybacks and Helios development. It is also the strongest counter to short theses on accounting quality — cash is conversion of GAAP earnings is improving, not deteriorating.
• Priced in? Yes — this is the part the multiple already rewards. But the magnitude of the FCF inflection has surprised even bulls.
• Importance: Positive.

7. Consensus PT now sits below the stock — sell-side has played most of its cards

The stock at ~$507 has run past the average analyst target of $486.33 (stockanalysis.com / Yahoo). MarketBeat shows a stricter consensus of $428.65 (~15.5% downside). Targets range from $225 (low) to $665 (Mizuho/Barclays high). The marquee recent upgrades:

• Citi (Atif Malik), Jun 12, 2026 — Neutral → Buy; PT $460 → $575.
• TD Cowen, Jun 1, 2026 — PT $500 → $600 (Buy).
• BofA, ~Jun 11, 2026 — PT $500 → $560 (Buy).
• Wedbush, May 4, 2026 — PT $290 → $400.
• DA Davidson, Apr 24, 2026 — PT $220 → $375 (upgrade to Buy).

But also the cuts:

• Northland, Jun 8, 2026 — Buy → Market Perform, citing 2027 consensus EPS/revenue as overly optimistic vs. likely hyperscaler capex slowdown.
• HSBC, May 4, 2026 — Buy → Hold on valuation.
• Argus / Morningstar fair value — $450.

Forward P/E 78.1x; EV/EBITDA 109x trailing; EV/Sales 23.6x; PEG ~1.04–1.32 — only growth-adjusted measures look "reasonable." Sector median P/E ~26x; AMD trades at a 182% premium on forward and 327% on trailing.

• So-what: Sell-side is no longer the marginal upside lever — the dispersion is widening, and the bears (Northland, HSBC) are leaning on 2027 capex as the swing factor, not 2026. The catch-up game is over; from here, fundamentals have to exceed the elevated targets, not just meet them.
• Priced in? Yes — extreme dispersion is the priced view. Variant signal would be a coordinated wave of cuts (not yet visible).
• Importance: Neutral-bearish.

8. The MI300X "45% of theoretical peak" finding — ROCm gap is not closing fast on training

The single biggest quiet finding in the specialist queries: Yahoo Finance reporting (cited in the specialist file) notes the MI300X delivers ~45% of theoretical peak performance in real-world settings versus >80% for the H100, attributing the gap to ROCm software immaturity. ROCm 7 (mid-2025) added day-0 support for MI350 and major frameworks; AMD also showed strong MLPerf 6.0 inference results in May 2026 with 9 OEM submissions. But — and this is what the bulls skip — no independent MLPerf training submissions for AMD GPUs are cited anywhere in the file.

The AI bull case requires the training workloads, not just inference. AMD is pivoting to "dominate inference" (memory-capacity advantage) — a defensible niche, but a smaller TAM than the headline market figures imply.

• Source: https://finance.yahoo.com/markets/stocks/articles/advanced-micro-devices-inc-amd-210033600.html ; https://www.amd.com/en/blogs/2025/enabling-the-future-of-ai-introducing-amd-rocm-7-and-the-amd-developer-cloud.html
• So-what: The gap that justifies NVIDIA's premium is software-driven, not silicon-driven. AMD's MI450 hardware can be at parity and still lose share in training. The "tens of billions" of MI450 revenue is real for inference; the training market is still NVIDIA's.
• Priced in? Partially — sophisticated investors know about ROCm. The specific 45%-of-peak data point is not in consensus models.
• Importance: Yellow flag.

9. Custom silicon (Google TPU, Anthropic, AWS Trainium, Meta MTIA) is the variant bear — Bernstein flagged it

The variant-perception finding that doesn't get enough airtime: Bernstein wrote that Google's TPU success "feels incrementally negative to the AMD narrative" because TPUs are "the only really successful ASIC program" at scale. Anthropic committed to 3.5 GW of TPU starting 2027; a $35B Anthropic loan is backed by Broadcom (digitimes, Jun 2026). Broadcom carries a $73B backlog in custom ASICs; Broadcom + Marvell control ~95% of the ASIC co-design market.

The mechanism: every hyperscaler dollar spent on a custom-designed ASIC is a dollar not spent on a merchant accelerator like Instinct. The OpenAI and Meta deals are AMD's win against NVIDIA, but the structural threat to both merchant vendors is the hyperscaler internalization that Google has already proven works.

• Source: https://www.investing.com/news/analyst-ratings/bernstein-google-tpu-news ; https://www.digitimes.com (cited in competition-agent file)
• So-what: The "second-source GPU" thesis is partially undermined. AMD's MI450 wins are real but the underlying mid-tier merchant TAM may compress through 2027–28 as more hyperscalers move workloads onto custom silicon. The Northland 2027 EPS downgrade is largely about this.
• Priced in? No — variant view. Bulls model rising AI TAM; bears model TAM cannibalization by ASICs.
• Importance: Variant perception.

10. Forensic file is clean — the SMCI short-squeeze playbook does not apply

The web confirms the filings on the boring stuff that matters most: E&Y unbroken auditor since 1970 with no restatements, no material control weaknesses (FY24 10-K and FY25 10-K/A confirm); the last SEC enforcement action was the 1996 cease-and-desist over 1992–93 Intel disclosures. No active material legal proceedings (Adeia patent suits settled Mar 9–11, 2026, terms undisclosed but quietly cleared). Whistleblower-related concerns surface only in historical (2012) commentary. Short interest 2.71% of float with 1.1–1.4 days-to-cover — no coordinated short-seller campaign. Free cash flow growth dramatically outpacing GAAP earnings growth in the right direction. CFO Hu's tenure clean since late 2022.

• Source: https://www.sec.gov/Archives/edgar/data/2488/000000248825000012/amd-20241228.htm ; https://stockanalysis.com/stocks/amd/statistics/
• So-what: Anyone pitching AMD as "the next SMCI" lacks evidence. The real risks here are strategic (MI450 execution, China policy, hyperscaler concentration) and valuation (priced for perfection), not accounting.
• Priced in? Yes — but worth knowing the bear case can't credibly attack the books.
• Importance: Positive (defensive).

---

Recent News Timeline

The reference layer behind the findings — every meaningful news item ordered by date, with significance rating. Sortable.

---

Industry & Competitive Layer — Beyond the Industry Tab

New external evidence that informs the AMD-specific position (does not restate baseline industry context).

Hyperscaler 2027 capex is the real swing factor

The Northland downgrade (Jun 8, 2026) was explicit: the 2027 EPS consensus is "overly optimistic, anticipating a potential decline in AI infrastructure spending in 2027 due to hyperscaler capital expenditure constraints." This frames the bear thesis precisely — it is not about AMD execution but about the cycle turning before AMD's MI450/MI500 ramp matures. AMD targets $100B annual data center revenue within 5 years (Nov 11, 2025 Analyst Day), which assumes a continued AI capex up-cycle through 2030.

CoWoS scaling and the binding constraint

Year	TSMC CoWoS capacity (wpm)	Allocation note
2025	65–75K (some estimates 35K)	NVIDIA bookings dominant
2026	120–130K	NVIDIA 800–850K wafers (>50%); AMD + Broadcom share residual
2027	Expanding further	Panel-level / hybrid bonding options emerging

AMD's escape attempt — panel-level EFB packaging flagged by Simply Wall St — is the most-discussed margin lever in the file. If it works at scale, gross margin improves and unit ceiling lifts; if it doesn't, MI450 ramp is capped by allocation.

Custom-silicon TAM compression

Custom silicon program	Status	Implication for AMD merchant accelerators
Google TPU	At scale; Anthropic 3.5 GW commit starting 2027	The "only really successful ASIC at scale" per Bernstein
AWS Trainium	Ramping; co-developed with Annapurna	Replaces merchant for AWS internal workloads
Meta MTIA	In production for ranking/recommendation	Note: Meta still chose AMD for MI450 6 GW
Microsoft Maia 100	Dual-track with NVIDIA Rubin	Smaller scale but a precedent

The variant bear: hyperscaler internalization cannibalizes 2027+ merchant TAM faster than consensus models.

Semi-custom trough — the 2026 numbers headwind nobody talks about

AMD guided that 2026 semi-custom (console SoC) revenue will decline by "a significant double-digit percentage" — 7th year of the PS5/Xbox cycle. PS6 ("Orion," 3nm, ~280mm²) and Xbox "Magnus/Project Helix" (3 Zen 6 + 8 Zen 6c, 68 RDNA 5 CUs, 48GB GDDR7) only ramp in 2027. Bull headline revenue forecasts (+43% in 2026, +54% in 2027) absorb this trough; embedded segment (post-Xilinx) is also a drag in the near term (Q1 2025 Embedded $823M, -3% YoY) before recovery in test & measurement / comms / aerospace.

---

Governance & People

Board. Eight directors, four committees, lead independent director Nora Denzel. Recent addition KC McClure (former Accenture CFO, Mar 2025 retirement; also on Goldman Sachs board) strengthens audit oversight — net positive signal. John Marren (Temasek N. America Vice Chairman) brings sovereign-wealth connectivity. Departure: Olson did not stand for re-election (Xilinx cohort retiree). No abrupt departures.

Executive comp. Lisa Su 2025 total comp ~$55.16M (Simply Wall St) vs ~$31.0M in 2024 — roughly +78% YoY. CEO/median pay ratio 216x. Phil Guido (CCO, hired May 2023): $9M sign-on package + $725K base + 125% target bonus, with PRSU tied to relative TSR (good design). Say-on-pay endorsed at the May 5, 2026 annual meeting with no disclosed dissent — even with Su's comp doubling.

Ownership. Institutional 71.34%, insider 0.50%. Vanguard 8.3%, BlackRock 7.2%, State Street 4.0%. ARK Invest trimmed in May 2026 (part of a $167M sell). No activist 13D filed. No related-party concerns of substance in the modern era.

---

Specialist Question Reference Grid

Most specialist answers worth surfacing are promoted into the ten ranked findings above. The remainder — useful for triangulation, not thesis-moving — is collapsed here. Includes questions where evidence was thin (flagged "limited").

---

Where the Edge Sits

The market knows AMD won the OpenAI deal, the Meta deal, the EPYC share gains, and the AI capex up-cycle. Sell-side targets are now catching up to a stock that has already passed them. What the market is still partially missing:

1. The OpenAI warrant's ASC 606 contra-revenue mechanic — accounting drag begins when first 1 GW ships in 2H 2026.
2. The CoWoS allocation ceiling — order book exceeds AMD's ability to ship.
3. MI450 gross-margin dilution confirmed by management but absorbed into >35% LT non-GAAP operating-margin target.
4. 2027 hyperscaler capex slowdown risk — Northland's bear call, not yet consensus.
5. Custom-silicon TAM compression — Bernstein's variant view.
6. Insider conviction at multi-year highs is uniformly bearish — no NEO is buying.

What is not the edge — and where short theses tend to overreach: AMD's books are clean, the auditor is unbroken, there is no SEC matter, no NEO departure, and no activist. The risks here are strategic, geopolitical, and valuation — not accounting.

---

Web Watch in One Page

Five live signals matter most over a five-to-ten-year horizon: the multi-year question (does ROCm close the CUDA gap), the accounting overhang (OpenAI penny-warrant contra-revenue), the most-proven moat under live threat (EPYC server share vs Intel 18A volume), the supply ceiling AMD does not control (TSMC CoWoS, HBM), and the structural threat to AI TAM (custom-ASIC compression and whether a third hyperscaler anchor pays full freight). Each watch item is tied to a long-term-thesis proposition or failure mode, and each has a definable evidence pattern that would force a real change of view rather than a quarter-by-quarter re-rating.

Active Monitors

Rank	Watch item	Cadence	Why it matters	What would be detected
1	ROCm vs CUDA software-gap progress	Weekly	P2 — the single multi-year question. The thesis breaks if PyTorch or TensorFlow never ship default-path ROCm by mid-FY2027, because Instinct stays inference-only and gross margin caps below corporate average.	Framework defaults (PyTorch, TensorFlow, vLLM, Triton, HuggingFace) adopting first-class or default ROCm support; new ROCm major-version releases with day-zero model coverage; AMD-submitted MLPerf training (not inference-only) submissions; or contrarian evidence that framework support has stalled.
2	MI450 gross margin path and OpenAI penny-warrant contra-revenue	Daily	The most monetizable variant in the report. Consensus FY27 EPS of $13.10 cannot survive if MI450 keeps ramping below corporate-average gross margin and the 160M-share OpenAI warrant flows through ASC 606 as contra-revenue once first 1 GW deployment is "probable."	10-Q/10-K footnotes quantifying ASC 606 / ASU 2025-04 treatment of the OpenAI warrant; 8-K vesting-probability disclosures; CFO Jean Hu or CEO Lisa Su commentary on Data Center segment gross margin or MI450 ramp economics; or independent analyst notes quantifying the contra-revenue range.
3	EPYC server CPU share vs Intel 18A volume ramp	Weekly	P1 vs F4 combined. EPYC at 46.2% revenue share is the most-tested moat in the company, but the bull case requires it to hold above ~42% even after Intel 18A Panther Lake / Diamond Rapids reach volume. This is the read that separates "structural mechanism" from "borrowed Intel-stall tailwind."	Mercury Research quarterly x86 server share publications; IDC server-CPU tracker data; Intel 18A high-volume manufacturing milestones (Fab 52), Panther Lake / Clearwater Forest / Diamond Rapids launch timing and yield commentary; hyperscaler EPYC Turin / Venice instance roll-outs; or share rolling below ~42%.
4	TSMC CoWoS and HBM supply allocation to AMD	Bi-weekly	F2 — the supply-side ceiling AMD does not control. NVIDIA reportedly holds ~60% of 2026 CoWoS wafers; AMD ships only what allocation permits regardless of MI450 demand. Allocation shifts (or panel-level EFB bypass progress) are the variable that turns the demand bull case into shippable revenue.	TSMC CoWoS capacity build-out and capex guidance; supply-chain reports breaking out NVIDIA / AMD / Broadcom allocation; HBM3E and HBM4 supply commitments from SK Hynix, Samsung, or Micron; AMD commentary about "supply-constrained" framing on MI450 / MI500; or progress on panel-level EFB packaging as a CoWoS bypass.
5	Custom-ASIC compression and a third multi-GW AI anchor for AMD	Daily	F3 + P5. The merchant-GPU TAM the bull case underwrites is being eaten by Google TPU, Meta MTIA, AWS Trainium, and Microsoft Maia at the highest-margin end. A third hyperscaler signing for MI450 / Helios at standard pricing (no warrant) would refute the bear thesis on cost of entry; a third anchor requiring another penny-warrant would validate it.	Broadcom and Marvell custom-AI-ASIC backlog or revenue disclosures; new Google TPU, Meta MTIA, AWS Trainium, or Microsoft Maia program milestones or hyperscaler commits; hyperscaler FY27 AI-capex guidance shifts (Microsoft, Meta, Alphabet, Amazon, Oracle); and an 8-K / press release naming a third multi-GW AMD Instinct anchor, with the equity-grant structure of that deal.

Why These Five

The report's verdict is "Watchlist" because the AI-franchise multiple is already paid while the margin path that multiple depends on has been guided down by AMD's own CFO. Three of the five monitors — ROCm progress, MI450 gross margin and ASC 606 contra-revenue disclosures, and the third-anchor commit watch — directly resolve the variables that decide whether consensus FY27 EPS of $13.10 holds or compresses toward the $11 variant view. The remaining two — EPYC share vs Intel 18A, and CoWoS/HBM allocation — guard the two flanks of the floor case: a structural EPYC moat under live foundry-competition threat, and a supply ceiling that decides whether even a bull demand scenario can ship as revenue. Generic earnings or sell-side rating watches were rejected because the report's catalysts page already covers scheduled prints, and none of them would update the long-term thesis the way these five signals would.

---

Where We Disagree With the Market

The sharpest disagreement, stated plainly. At ~22x EV/sales, consensus is pricing the Instinct franchise on a "scale-into-corporate-margin" template — the implicit model that merchant-GPU revenue ramps, Xilinx amortization runs off, and gross margin walks toward NVIDIA. The evidence on the page contradicts every leg of that template over the next 12-18 months. CFO Jean Hu has told the Street MI450 ramps below corporate-average gross margin for the next one-to-two years; the OpenAI penny-warrant is not just dilution but an ASC 606 contra-revenue mechanism that increases as Instinct ships volume; and AMD's MI450 unit ceiling is capped by a TSMC CoWoS allocation it does not control. Consensus FY27 EPS at $13.10 does not reconcile with what management has already disclosed. The variant is not that the AI story is wrong — it is that the Street has mis-modelled the shape of the cash that comes from being right. The watchlist number: Data Center segment GAAP gross margin in the Q3 FY26 print — the first quarter MI450 mix materially builds. Sub-52% is bear-confirming; 55%+ is bull-confirming.

Mapping Consensus — What the Market Actually Believes

Variant perception is only credible when "the market believes X" is anchored to a concrete consensus signal — a multiple, a target, a guide, a price reaction, a repeated narrative. The table below converts the sentiment into a testable underwriting assumption, because an assumption is something the next print either confirms or breaks. A vibe is not.

Two read-the-room points before the ledger. First, the bull and bear pages already capture the direction of disagreement — the variant edge here is sharpening the mechanism, especially the under-modelled ASC 606 contra-revenue path on the OpenAI warrant. Second, the most useful disagreement on this page is not a fresh argument for the bear; it is a reframing of which numbers the consensus model has wrong, and over what horizon they become legible.

The Disagreement Ledger — Where Evidence Disagrees With Consensus

Three disagreements survive all five tests (consensus exists, evidence contradicts, materiality is real, observable resolution path, falsifiable). Ranked by what would most update a PM's underwriting today.

Classifying the three variants against the eight high-quality buckets

None of the three lean on the banned weak forms ("high-quality but undervalued," "market too pessimistic," "execution risk remains"). Each is anchored to a mechanism the next two prints will either confirm or refute.

The Lead Disagreement — Walked Through

If the PM only reads one section, this is the one.

The chart makes the central observation visual. Revenue and gross margin variants are small in percentage terms — yet EPS is 16% below consensus because the tax-rate and share-count deltas magnify modest top-line and margin haircuts into a large bottom-line gap. This is why the variant is monetizable: even a directionally-correct revenue and margin view gets to a sharply different EPS once the second-order accounting items are layered in.

What consensus would say. "Mechanical Xilinx amortization run-off (~6pp over 5-7 years) closes most of the GAAP / non-GAAP gap; Instinct gross margin migrates to corporate average by FY28 as scale leverage kicks in; the warrant is one-time dilution priced into the diluted share count; FY25 tax rate is a discrete-item issue that doesn't recur." All four claims are directionally correct — and all four under-estimate the magnitude and timing in the way the bridge above quantifies.

What our evidence disagrees on. The CFO has explicitly committed to sub-corp-avg GM during the MI450 ramp; ASU 2025-04 was specifically clarified in 2025 to address share-based consideration to a customer as contra-revenue when vesting is probable (forensics-claude.md, research-claude.md). These are not bear interpretations of management language — they are management language plus FASB clarification. The bull's "mechanical amortization run-off" argument is real over 5-7 years but provides no offset in the FY26-27 window where consensus is being marked.

What the market must concede if we're right. That AMD's consolidated multiple — already inside NVDA's neighborhood at 22x EV/sales — is paying for margin expansion that is delayed at minimum 18-24 months past the consensus EPS path. That delay is sufficient to compress the forward P/E and trigger the same setup that produced the -17.3% Q4 FY25 reaction off a soft guide.

The cleanest disconfirming signal. Two consecutive quarters of Data Center segment GAAP gross margin holding above 55% while Instinct revenue scales meaningfully. The verdict-claude.md page named this as the bull-confirming signal; we agree. If Q3 FY26 prints DC GM at 55%+ with visibly higher MI450 mix, the lead variant breaks.

The Bullish Variant — EPYC + Embedded + Console Are Floor Franchises

This is the variant the bears miss, not the bulls.

The bear case names a downside target of ~$270 and the long-term-thesis bear scenario lands ~$165. Both pricing exercises under-value the three franchises above. EPYC alone — at the management TAM and conservative 15x segment OI — is worth ~$250-300B standalone; Embedded $25-35B; semi-custom contracted cash flow $30-50B PV. The floor is closer to ~$300-400B of equity value even where Instinct's multiple collapses to volume-only economics. That is roughly $180-240/share against today's $521. The asymmetry: the variant is not "AMD is cheap here" — it is "the floor is real, and the cleanest setup to underwrite is the next 30-40% drawdown when the bear gets paid on the lead variant while the floor franchises trade as cyclical drag."

What consensus would say. "Consolidated multiples are how the market prices conglomerates; isolated SOTPs don't pay you in a re-rating event." Fair point, but it is exactly the moment the consolidated multiple compresses that SOTP gets reactivated — see Intel post-FCF-collapse where the foundry was valued separately precisely because the consolidated multiple broke.

What would prove us wrong. A Q3 FY26 EPS disappointment that triggers a re-rating and coincident weakness in EPYC share data from Mercury Research — i.e., the floor franchises themselves start to crack at the same time the AI multiple compresses. If EPYC share rolls below 42% in any Mercury report before Intel 18A Panther Lake is in server volume (a thesis-breaker per long-term-thesis-claude.md F4), the floor argument loses the franchise that does most of the work.

The Quiet Variant — ASC 606 Contra-Revenue on the OpenAI Warrant

This is the accounting mechanic that consensus is uniquely positioned to miss because it sits at the intersection of revenue recognition and customer-equity treatment, with the relevant FASB clarification (ASU 2025-04) only codified in 2025.

The variant here is not "the warrant is bad" — bulls and bears both know the warrant exists. The variant is that the warrant's revenue/margin mechanic is asymmetric to the bull narrative: the more deeply the deal works on the demand side, the more the GAAP top-line and gross margin compress, and the more tranches vest on the back of the stock-price escalator. Consensus has it inversely: bulls treat the warrant as a one-time dilution cost paid in exchange for revenue; the accounting says the warrant compresses revenue and dilutes shares in proportion to the bull case landing.

What consensus would say. "Meta signed without a warrant — proof the warrant was a one-time cost of entry to anchor the first hyperscaler. The mechanic doesn't recur." That is the bull's strongest counter, and we partially agree: the Meta deal is a positive signal that AMD did not have to pay a recurring tax. But it does not unwind the existing OpenAI warrant — that accounting plays out through Oct 2030 regardless of what Meta did. The variant is about the modelled magnitude of the existing warrant, not the precedent for the next deal.

The disconfirming signal. A clean Q3 / Q4 FY26 10-Q footnote that quantifies the ASC 606 treatment in a smaller-than-feared range (sub-$0.5B / Q in first disclosure), OR an 8-K-disclosed third hyperscaler anchor at standard pricing that prices Instinct against a clean comp. Both would shrink this variant materially.

Evidence the PM Can Audit Fast

The strongest items in the report that move the probability of these three variants. The fragility column matters most — every piece of evidence has a way to become misleading, and an honest PM should know what it is.

Resolution Signals — What Settles the Debate

For each variant, the specific observable that proves it right or wrong, where to look, and on what timing. Every signal below is in a filing, an earnings call, a vendor disclosure, or an industry data publication — none rely on "execution" or "time will tell."

Red Team — What Would Make Us Wrong

The discipline of variant perception is to name the evidence that breaks the view before the market does. The following four reads of the same evidence base would make this page wrong.

Of the four, risk #1 is the most genuinely fair pushback — AMD's guidance posture has been mechanically conservative on the DC GPU dollar guide twice this cycle (the three-time-raised FY24 DC GPU $4B → $4.5B → $5B+ pattern is the receipt). A Q2 print at guided GM with strong Q3 outlook would meaningfully weaken the lead variant. Risk #2 is the second-most fair — probability assessment under ASC 606 has discretion, and AMD may stage recognition slowly. The variant survives risk #3 and #4 even if both materialise: tax planning at 6-8% buys back ~$0.50 of the bridge but the GM and warrant legs remain; hyperscaler capex resilience and ROCm progress support the long-term case but do not unwind the FY26-27 mechanical compression.

The Closing Pointer

---

Liquidity & Technical

AMD has gone vertical. The stock closed $520.89 on 17-Jun-2026, sitting 4.8% below its all-time high of $547.26 set two sessions earlier, trading at roughly 2.0x its 200-day average. Trend regime is bullish — golden cross is eleven months old, price has held above every major SMA all year — but 30-day realized volatility has run to 91%, the 91st-percentile band of the last decade, and the MACD histogram rolled negative as the rally extended.

Liquidity is not the constraint. At $14.8B per day in dollar volume, a fund running 5% in AMD at 20% ADV participation is implementable to ~$311B in AUM; a 1%-of-market-cap position liquidates in six trading days at the same rate. The constraint is size discipline, not capacity.

Stance — Headline

Liquidity — Capacity Is Not the Constraint

At 717% annual turnover, AMD's float churns more than seven times per year — a tape this active digests blocks without slippage, and the 2.3% median daily range keeps execution friction comfortably below the 2%-impact threshold for a participation-disciplined fund. The verdict is deep institutional liquidity: even a $300B AUM portfolio can hold a 5% position and exit in roughly a week at 20% ADV. Liquidity is a non-issue for any reasonable fund size; the real question is what to do with it.

Trend — A Ten-Year Story That Ends Vertical

The chart breaks into three regimes. 2016 to early 2020: AMD compounded from sub-$5 to $50 on the Zen architecture turnaround. 2020-2021 cycle: ran to $164, then a 62% drawdown through October 2022 as the semi cycle rolled over. 2023-2024: sideways consolidation between roughly $100 and $190. From the April-2025 low at ~$97, AMD has run over 5x in fourteen months, with the parabolic phase concentrated in the last three: $354 at the end of April, $516 in late May, $520 today.

Trend Regime — Price vs Moving Averages (Last 3 Years)

The most recent death cross was 14-Aug-2024 — the 50-day crossed below the 200-day as the post-2024 consolidation unwound. The golden cross that reversed it printed on 16-Jul-2025, and has held for eleven months without invalidation. The 50-day has accelerated away from the 200-day in 2026, opening a $146 gap (50d = $405, 200d = $259) — a confirmation of trend strength, but also the source of the extension problem: price is $116 above the 50-day SMA and 101% above the 200-day. Both readings are at decade extremes.

Momentum — RSI Cools, MACD Histogram Negative

This is the most important divergence on the page. RSI peaked at 83 in late April as price broke out, but has cooled to 58.9 today even as the index made new closing highs in mid-June. The MACD picture is sharper: the histogram printed +8.6 in mid-May, has rolled to -4.2 today, and the MACD line ($29.1) has crossed below the signal line ($33.3) — a short-term sell signal in the standard reading. This is a classic bearish momentum divergence on price strength: the tape is making higher highs on declining internals. It does not by itself invalidate the trend, but it is the first crack in three months of one-way price action.

Volatility — A Top-Decile Regime

Current 30-day realized volatility of 91% sits between the 80th-percentile band (66%) and the decade maximum (107%). Translated to dollars, ATR(14) has expanded from roughly $3 last summer to $23.8 today — a single ATR move is now a 4.6% swing. For sizing purposes this is the constraint: a portfolio that runs AMD at last year's vol-weighted size is now carrying roughly 2x the risk of the same notional position. This is a fully-priced momentum trade, not a quiet accumulator.

The Bollinger middle (20-day SMA) sits at $498 with the upper band at $556 and the lower at $440 — a $116 band width that captures how wide the daily range has become. Today's close at $521 is mid-band, leaving room either way; an upside print through $556 with volume would mark a fresh breakout, a close below $440 would mark the first technical break of the parabola.

Volume — The Tape Confirmed the Move

The decade's heaviest volume days are concentrated around earnings releases and major product cycles. The most consequential recent print was 6-Oct-2025: 248.9M shares (4.3x the 50-day average) on a +23.7% day-close. That session likely marks the lift-off of the current rally, and the absence of any subsequent volume-spike that broke trend supports the read that distribution has not yet begun. The 2017-05-02 spike (-24.2% on 3.8x volume) is the only historical analog of a high-volume reversal — a useful reference for what a real distribution day would look like.

Returns Snapshot

The 3-month return matches the 6-month return — almost the entire 1-year +312% gain was compressed into the last six months, and roughly half of that into the last three. The YTD print (+133%) is among the largest first-half advances in the stock's history at this market-cap scale. For perspective, AMD's rebased 3-year cumulative price line stands at 418 (base of 100 in June 2023), a +318% absolute return that — even without a directly-populated SPY/XLK comparison series in the staged data — is structurally an enormous outperformance versus any reasonable broad-market or sector benchmark.

Key Levels

A close above $547.26 is the operational breakout level for any momentum add-on. A close below $405 breaks the rising 50-day and would convert the chart from "extended uptrend" to "topping pattern." Between those two, the playable risk is the Bollinger lower band at $440 — a clean defended bid there keeps the trend intact.

Technical Scorecard

The composite reads +2 of +3 — bullish, but not maximum-bullish. Volatility is the lone negative, and it is the operational signal: sizing must respect that one ATR is now a 4.6% move.

Bottom Line — Stance, Levels, Action

Six-month read: trend is intact but extended. Trend, volume confirmation, and absolute strength still favor continuation; the momentum divergence and top-decile realized vol argue against chasing strength at extension.

Two levels frame the next quarter:

Bull-confirming setup: close above $547.26 on volume. A weekly close through the all-time high, with volume above the 60-day average of ~36M shares, opens a path toward $580–$600 (Bollinger upper band plus a measured-move extension of the May breakout).

Bear-confirming setup: close below $405. A breach of the rising 50-day SMA — particularly on a day where volume exceeds the recent average — signals the momentum cohort is unwinding and reopens a re-test of the 200-day near $260.

Implementation: for funds initiating, build over weeks, not days, with the first add on a defended pullback to the Bollinger lower band ($440) or the 50-day ($405). Liquidity supports any reasonable build pace; size discipline is the constraint. For existing holders, the scorecard reads add-on-strength bias with a tight invalidation: a close below the 50-day removes the working hypothesis.

---

Short Interest & Thesis

Bottom line. Reported short interest is not crowded — 44.07M shares at May 29, 2026, ~2.71% of float with 1.37 days to cover — so positioning is not a near-term squeeze setup and the level itself is not decision-useful. The decision-useful read is the trend: shorts added ~+8.4M shares (~+28%) across the two May reports while the stock ran from $354 to $516, then trimmed slightly into June. Disciplined bears re-engaging at the highs, not capitulating. No public short-seller report, no activist campaign, no accounting allegation, no hard-to-borrow evidence; the unresolved thesis risk is strategic and valuation-driven (OpenAI penny-warrant ASC 606 contra-revenue, MI450 gross-margin dilution, CoWoS allocation cap, 2027 hyperscaler-capex slowdown), not forensic.

Reported Positioning — Eight-Month Trend

The data below is third-party-mirrored FINRA semi-monthly short interest. Aggregate snapshot is sourced from MarketBeat (May 29, 2026 settlement) and corroborated by Benzinga and StockAnalysis.com. The FINRA staged data for this run returned zero rows for AMD; the trend table here is the closest public substitute and should be read as public reported position data, not as inferred or proprietary borrow analytics.

The two May 2026 prints are the only meaningful signal in this series. Short interest sat in a 30–38M range from October 2025 through April 2026 — a 2.0–2.4%-of-float band that was effectively flat against a stock more than doubling. On the May 15 settlement, position size jumped from 36.14M to 44.70M (+24%) in a single report, and on May 29 it held at 44.07M with the dollar value short reaching $22.74B at the higher mark. The covering signal we would expect from a parabolic move never showed up — shorts pressed into strength rather than getting forced out, and only trimmed 1.4% into the most recent print despite price hitting an all-time high. This pattern is consistent with discretionary, fundamentally-driven bears who scale into valuation rather than reactive trend-following shorts.

Crowding vs Liquidity — A Non-Issue at This Float

Crowding is what makes short interest decision-useful; AMD does not have it. At 44M shares short against a 1.63B share float and a 20-day average trading volume of roughly 30M shares (about $14.8B per day in dollar volume), the entire short position covers in roughly one and a half trading days at average volume — or in a single session at heavy participation. There is no liquidity bottleneck for shorts to exit and therefore no structural squeeze setup, even if a catalyst forced rapid covering.

The conventional crowded-short heuristics — short interest above 10% of float, days to cover above 7, lendable supply scarcity — none apply. AMD's annualized turnover is 717% with a 2.3% median daily range, which is the opposite of a squeeze setup: it is a tape that absorbs both sides without slippage. A short-thesis-driven exit by fundamental bears would not move price; the squeeze risk would only ever come from a trend-follower force-out, which the current low days-to-cover ratio also makes implausible.

Peer Context — Mid-Pack for Mega-Cap Semis

AMD's 2.72% short interest sits above the mega-cap semi median but well below the most-shorted names in the group. The cleanest peer set:

AMD reads as a typical valuation-stretched semi: meaningfully more shorted than NVDA and AVGO (the two AI winners that have already proven the model), comparable to INTC (the value/turnaround name), and below MU and QCOM (where the bear cases are more developed). The signal here is that AMD is not in the "stealthy crowded short" zone — which would be Jefferies's October 2024 "Top Crowded Semiconductor Short Positions" alert, where AMD was indeed listed — but the relative ranking now is consistent with a mid-pack name, not an under-shorted darling and not a target. CNBC reported on June 9, 2026 that semiconductor shorts have broadly piled on as the rally extended ("Semiconductor shorts pile on as winning trade reverses"), which is the macro context — AMD's May build is part of that broader semi-skepticism rather than name-specific.

Public Short-Thesis Ledger — Strategic and Valuation, Not Forensic

There is no public short-seller report on AMD. No activist campaign. No accounting-allegation report. No regulatory matter rising to the level a short would underwrite. The forensic file is clean (Ernst and Young unbroken since 1970, no SEC investigation, no restatement, no material weakness, the last enforcement action was the 1996 Intel-disclosure cease-and-desist). What does exist is a coherent strategic and valuation short thesis assembled from sell-side downgrades, management's own commentary, and accounting mechanics in primary filings.

Borrow Pressure — Inferable, Not Directly Measured

Public borrow-fee, utilization, and lendable-supply data for AMD were not staged in this run and could not be verified directly from primary sources within the search budget. Two inferences hold:

• AMD is a NASDAQ-100 mega-cap with $846B market cap, 1.63B-share float, 71% institutional ownership, and 717% annualized turnover. That structural profile virtually guarantees deep lendable supply (Vanguard 8.3%, BlackRock 7.2%, State Street 4.0% are the three largest holders and are aggressive securities lenders). General-collateral borrow rates near zero are the prior.
• Fintel hosts a borrow-fee feed for AMD but the actual rate values were not extractable from public excerpts. No mention of hard-to-borrow status or locate friction surfaced in any news, forum, or sell-side material gathered during dependency research; the absence of such commentary across multiple short-interest tracking pages (Benzinga, Marketbeat, Nasdaq, StockAnalysis, Finviz) is meaningful — distressed borrow conditions on a name this widely-watched would surface immediately.

The working assumption is borrow is easy and cheap, which is consistent with the +28% May build occurring without any visible locate friction. If borrow tightens unexpectedly, that itself would be a thesis-shifting event; for now, treat borrow as a non-constraint on either direction.

Tape and Squeeze Setup — Not a Squeeze Candidate

Putting positioning together with the technicals: the stock closed at $520.89 on June 17, 2026, sitting 4.8% below its all-time high and roughly 2.0x its 200-day moving average. The trend is parabolic; 30-day realized volatility is 91% (top decile of the last decade). The technicals page reads the setup as "trim, not exit" on extended trend. Short interest does not change that read in either direction:

• No squeeze tailwind. 1.37 days to cover means even a forced-cover scenario clears in two sessions at average volume — too shallow to drive a sustained gap. The OpenAI-revenue-miss-day moomoo headline ("AMD slips, helping short sellers as OpenAI's revenue miss fuels jitters") is the tape pattern to expect at the current SI level: shorts make money on negative news, but their footprint is too small to magnify positive news.
• No de-risking overhang. With 44M shares short against $22.7B of dollar exposure, a coordinated cover would inject roughly 1-1.5 days of incremental buying — supportive but not transformative.
• Asymmetry is in the thesis, not the position. The +28% May build with the stock up 46% says bears are willing to short higher prices; that compresses the upside surprise factor on positive news (less forced covering) and amplifies the downside on disappointment (existing shorts press, no covering bid).

Evidence Quality and Limitations

What is verified, what is inferred, and what is missing.

What Would Change the Read

Three things would flip short interest from "not decision-useful" to "thesis-shifting":

1. A coordinated short-seller report (Hindenburg, Citron, Spruce Point) attacking the OpenAI warrant accounting, MI300X performance gap, or China revenue-share precedent. None visible today. The forensic file is the cleanest defensive moat the company has.
2. A doubling of short interest above ~5% of float without a coincident catalyst would suggest borrow capacity is being drawn down systematically — at which point days-to-cover becomes a real tape factor. The May build was +28% in two reports; another similar move would put AMD into "watchable" territory.
3. A hard-to-borrow signal in the form of rising borrow fees, locate friction, or fails-to-deliver. Not visible today and structurally unlikely on a name with this float and institutional ownership.

Until one of those triggers, the short side of AMD is a sentiment indicator — disciplined bears willing to short higher — not a positioning signal. The bear case is real and well-articulated; the bear position is small and easy to exit.