Financials

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.