At $562.23 the market has already spent the gigawatt announcements: AMD must compound owner earnings at +35.4%/yr for a decade — roughly the pace of management's own 2027 "more than double" Data Center commentary, sustained for ten years instead of two — and the first 1-gigawatt tranches that decide whether that is real start shipping this quarter.
Pre-earnings baseline for Q3 2026. Every figure below comes from primary sources: SEC filings, company releases and full earnings-call transcripts. No analyst estimates.
AMD designs chips it does not manufacture — Taiwan Semiconductor Manufacturing Company (TSMC) fabricates them — and sells into four reported lines:
Operating income is reported for three segments — Data Center, Client and Gaming combined, and Embedded — with stock-based compensation (SBC), acquisition amortization and deal costs held in an "All Other" bucket (−$4,153M TTM), which is where the gap between generally accepted accounting principles (GAAP) and non-GAAP results lives.
AMD has converted itself from a CPU challenger into the only credible full-stack #2 in AI compute — CPUs, GPUs, networking, rack-scale systems (Helios) and an open software stack (ROCm) — and the three largest AI builders have each signed gigawatt-scale deployment frameworks with it. The operating business is compounding exactly as the story requires: revenue +27.1%/yr from fiscal year 2023 (FY23) to TTM, Data Center operating income up to a 29.4% TTM segment margin, gross margin rebuilt to 54% GAAP / 56% non-GAAP.
The question the price asks is narrower and harder than "is AMD winning?" It is: how much of the announced 14-gigawatt (GW) future must become shipped, cash-collected revenue, and how fast? At management's own revenue-per-gigawatt framing ("double-digit billions," Q2 2026 call), the full framework is a $140B+ multi-year revenue program — more than three years of AMD's current total revenue — of which only the first 1 GW tranches for OpenAI and Meta (shipments beginning second half (H2) 2026) and Anthropic's first GW (first half (H1) 2027) have dated starts. The $562.23 price capitalizes a large share of that conversion happening on schedule, and pays for it partly in future dilution (warrants + World Labs stock) that the headline share count does not yet show. The position is therefore a bet on conversion speed and on Su-era execution continuing to beat its own guides — a bet the track record supports and the valuation fully charges for.
| Metric | Q2 2026 | Q2 2025 | Change |
|---|---|---|---|
| Revenue | 11,536 | 7,685 | +50.1% |
| Gross profit (GAAP) | 6,203 | 3,059 | +102.8% |
| Gross margin (GAAP) | 53.8% | 39.8% | +14.0 pts |
| Gross margin (non-GAAP) | 56.2% | 43.3% | +13.0 pts |
| Operating income (GAAP) | 1,990 | −134 | swing to profit |
| Operating income (non-GAAP) | 3,094 | 897 | +244.9% |
| Net income (GAAP) | 2,297 | 872 | +163.4% |
| Diluted earnings per share (EPS, GAAP) | $1.38 | $0.54 | +155.6% |
| Data Center revenue | 6,718 | 3,240 | +107.3% |
| Data Center operating income | 2,103 | −155 | swing to profit |
| Client revenue | 3,062 | 2,499 | +22.5% |
| Gaming revenue | 779 | 1,122 | −30.6% |
| Client & Gaming operating income | 582 | 767 | −24.1% |
| Embedded revenue | 977 | 824 | +18.6% |
| Embedded operating income | 386 | 275 | +40.4% |
| Free cash flow (quarter) | ~$1.6B | — | per Q2 call |
Q2 2025 carried the ~$800M MI308 China inventory charge, which depresses every YoY GAAP comparison — flagged wherever it matters. Client & Gaming operating income fell despite higher revenue (mix + go-to-market/roadmap investment, per the call).
| Metric | FY22 | FY23 | FY24 | FY25 | TTM |
|---|---|---|---|---|---|
| Data Center revenue | 6,043 | 6,496+7.5% | 12,579+93.6% | 16,635+32.2% | 22,214+33.6% |
| Client revenue | 6,201 | 4,651−25.0% | 7,054+51.7% | 10,640+50.8% | 11,794+10.8% |
| Gaming revenue | 6,805 | 6,212−8.7% | 2,595−58.2% | 3,910+50.7% | 3,640−6.9% |
| Embedded revenue | 4,552 | 5,321+16.9% | 3,557−33.2% | 3,454−2.9% | 3,657+5.9% |
| Total revenue | 23,601 | 22,680−3.9% | 25,785+13.7% | 34,639+34.3% | 41,305+19.2% |
| GAAP gross margin | 44.9% | 46.1%+1.2 pts | 49.3%+3.2 pts | 49.5%+0.2 pts | 53.2%+3.7 pts |
| GAAP operating income | 1,264 | 401−68.3% | 1,900+373.8% | 3,694+94.4% | 6,488+75.6% |
| Net income | 1,320 | 854−35.3% | 1,641+92.2% | 4,335+164.2% | 6,434+48.4% |
| Operating cash flow | 3,565 | 1,667−53.2% | 3,041+82.4% | 7,709+153.5% | 10,080+30.8% |
| Capital expenditures | 450 | 546+21.3% | 636+16.5% | 974+53.1% | 1,655+69.9% |
| Free cash flow (FCF) | 3,115 | 1,121−64.0% | 2,405+114.5% | 6,735+180.0% | 8,425+25.1% |
| Stock-based compensation | 1,081 | 1,384+28.0% | 1,407+1.7% | 1,638+16.4% | 1,895+15.7% |
| Owner earnings (FCF − SBC) | 2,034 | −263swing to loss | 998swing to profit | 5,097+410.7% | 6,530+28.1% |
| Data Center operating income | 1,848 | 1,267−31.4% | 3,482+174.8% | 3,603+3.5% | 6,528+81.2% |
| Client & Gaming operating income | 2,143 | 925*−56.8% | 1,187+28.3% | 2,855+140.5% | 2,749−3.7% |
| Embedded operating income | 2,252 | 2,628+16.7% | 1,421−45.9% | 1,243−12.5% | 1,364+9.7% |
| All Other operating loss | −4,979 | −4,41911.2% narrower | −4,1905.2% narrower | −4,0074.4% narrower | −4,1533.6% wider |
*FY23 Client & Gaming operating income is the sum of the separately reported Client (−$46M) and Gaming (+$971M) lines under the prior four-segment presentation; FY22 segment operating income is likewise from the FY23 Form 10-K. From FY24 AMD reports Client and Gaming as one segment (revenue still split). TTM = FY25 − H1 2025 + H1 2026 from the Forms 10-Q; TTM column changes are vs FY25. TTM gross profit and operating figures are four-quarter sums (Q3 2025–Q2 2026) and tie to the same totals.
Segment margins (TTM): Data Center 29.4% · Client & Gaming 17.8% · Embedded 37.3% — before the All Other bucket (acquisition amortization $2,214M TTM + SBC $1,895M TTM + deal/other costs) that takes consolidated GAAP operating margin to 15.7%.
"For the third quarter of 2026, AMD expects revenue to be approximately $13 billion, plus or minus $300 million. … Non-GAAP gross margin is expected to be approximately 56%."
The Q3 guide midpoint implies +40.6% YoY and +12.7% sequentially.
The same pattern holds back through the Q1 guides of 2023–2025 (four straight Q1 guides met or beaten) and the 2024 Data Center GPU outlook, raised three times and delivered. AMD gives no formal full-year guidance — the discipline is real but the commitment horizon is one quarter.
Management also states it is tracking "materially ahead" of the Nov 2025 Financial Analyst Day model (>35% revenue growth, >$20 annual non-GAAP EPS within the strategic timeframe) and expects to "significantly exceed" the $20 EPS target. Client: management plans for a softer H2 2026 PC market on memory/component costs.
Every named AI-accelerator framework in the filings and releases, classified by what it legally is:
| Customer | Announced | Scale | Instrument status | First deployment | Warrant |
|---|---|---|---|---|---|
| OpenAI | Oct 6, 2025 | 6 GW, multi-year, multi-generation (MI450 first) | Definitive agreement (Form 8-K + supply agreement; "OpenAI to deploy 6 gigawatts") | 1 GW from H2 2026 | Up to 160M shares @ $0.01, tranches vest on purchase milestones scaling to 6 GW + AMD share-price targets + OpenAI technical/commercial milestones; expires Oct 5, 2030 |
| Meta | Feb 24, 2026 | "Up to" 6 GW, multi-year, multi-generation; custom MI450-based GPU on Helios; Meta also lead customer for EPYC Venice/Verano | Definitive partnership agreement (Form 8-K), framed as "deploy up to 6 gigawatts" | First 1 GW shipments from H2 2026 | Up to 160M shares @ $0.01, performance-based, vesting on Instinct shipment milestones + AMD stock-price thresholds + technical/commercial milestones; expires Feb 23, 2031 |
| Anthropic | Aug 4, 2026 (Q2 release/call) | "Up to" 2 GW of MI450 in Helios + multiyear ROCm/Claude engineering collaboration | Announced strategic partnership; no Form 8-K agreement or warrant filed as of this dive | First 1 GW from H1 2027 | None disclosed |
Quantification: named frameworks total 14 GW — OpenAI 42.9%, Meta 42.9%, Anthropic 14.3% of the announced gigawatt book. At management's "double-digit billions" of revenue per gigawatt, the full book is a $140B+ multi-year revenue program if every gigawatt ships — vs TTM total company revenue of $41,305M. Only 3 GW of that book has a dated first-tranche start (OpenAI 1 GW + Meta 1 GW from H2 2026; Anthropic 1 GW from H1 2027). What is contracted in the accounting sense is not disclosed: AMD reports no backlog or remaining performance obligations for these agreements, "no customer accounted for at least 10% of consolidated net revenue in fiscal 2025 and 2024" (FY25 Form 10-K), and the warrant vesting schedules (Exhibits E/F to the OpenAI warrant) are redacted in the public filings. Microsoft (Helios on Azure), Oracle and others are named Helios deployers without gigawatt figures. Beyond the equity: AMD has issued ~$4.1B of guarantees in connection with commercial partners' data-center lease obligations (terms up to 15 years), which "may be issued in exchange for warrants" (Q2 2026 Form 10-Q, commitments note) — contingent exposure that grows with the same build-out.
Per-share consequence: if all 320M warrant shares vest (they vest only alongside the purchases — the dilution arrives with the revenue), the count rises ~19.6% to ~1,952M before World Labs (~15M shares) and ongoing SBC issuance. The reverse DCF's per-share lens prices that path explicitly below.
Cash $5,086M + short-term investments $8,025M = $13,111M liquid; total debt principal $3,250M → net cash ≈ $9,861M. The revolver and commercial-paper program were undrawn in H1 2026.
| Debt stack by maturity | Principal |
|---|---|
| 4.212% notes due 2026 (matured Sep 24, 2026, after period-end) | $875M |
| 4.319% notes due 2028 | $625M |
| 2.375% notes due 2030 | $750M |
| 3.924% notes due 2032 | $500M |
| 4.393% notes due 2052 | $500M |
| Total debt principal | $3,250M |
Post-period: Aug 17, 2026, AMD issued $4.75B of senior notes — $1.25B 4.600% due 2029, $1.50B 5.000% due 2031, $1.00B 5.250% due 2033, $1.00B 5.500% due 2036 — for "general corporate purposes, which may include the repayment of debt." The maturity wall is now termed out; interest cost rises modestly (interest expense was $74M in H1 2026).
Working capital is being consumed by the ramp, deliberately: inventories $8,468M (+$548M in H1), accounts receivable $7,281M, prepayments/other assets up ~$991M in H1 (supply commitments), payables up $2,170M in H1 funding much of it. H1 operating cash flow was $5,321M on net income of $3,680M.
Off-bridge items, disclosed not buried: $4.1B maximum guarantee exposure on partner data-center leases (recorded as a credit derivative in other long-term liabilities, changes through other income); ~$2.9B of long-term investments (incl. $1,220M of marketable equity securities at fair value — an investee that listed publicly in Q2 2026, generating $549M of H1 mark-to-market gains inside the $763M of H1 other income — and $1.7B of non-marketable equity); goodwill $25,470M and acquisition intangibles $15,635M net (the Xilinx legacy); a $100M term-loan receivable from an investee maturing Oct 2026.
Verdict on health: unambiguously strong — net cash, termed-out maturities, no reliance on the revolver. The balance-sheet risks are contingent (guarantees) and valuation (goodwill), not liquidity.
Inverting the price on enterprise value of $907.7B — market capitalization $917.6B + debt $3.25B − cash $5.09B − short-term investments $8.03B — owner earnings, defined as free cash flow minus stock-based compensation, must compound at:
Actual compounding beside the bars:
Owner earnings compounded +39.6%/yr FY22 → TTM — off a small base (FY23 owner earnings were negative, −$263M, so no FY23-based owner-earnings compound rate is meaningful). The Q3 2026 guide (+40.6% YoY at midpoint) and management's 2027 commentary both run ahead of the +35.4% bar near-term.
| Base (exact definition) | Base $M | Implied growth | CAP +25% | CAP +30% | CAP +35% | TV % EV | Term. EV/EBIT |
|---|---|---|---|---|---|---|---|
| Owner earnings, reported (TTM FCF $8,425M − SBC $1,895M) | 6,530 | +35.4%/yr | 16.0 yrs | 13.0 yrs | 11.0 yrs | 73.2% | 12.8× |
| Owner earnings, normalized (reported owner earnings excl. the +$360M Q4 2025 MI308 charge reversal inside the TTM) | 6,170 | +36.2%/yr | 16.0 yrs | 13.0 yrs | 11.0 yrs | 73.6% | 12.1× |
| Raw TTM free cash flow, SBC not subtracted (memo only) | 8,425 | +31.7%/yr | 14.0 yrs | 11.0 yrs | 9.0 yrs | 71.6% | 16.6× |
The MI308 normalization is deliberately small: the TTM contains no MI308 charge (the ~$800M charge sits in Q2 2025, outside the window) but does contain Q4 2025's ~$360M reversal as those written-down parts shipped. Removing it raises the bar 0.8 pts — the diagnosis is "temporary, policy-driven," and the honest statement is that MI308 no longer distorts the base much in either direction. The larger structural fact needs no normalization: TTM owner earnings are earned at a 15.8% margin on revenue while the business mix (Data Center at a 29.4% segment margin and 58% of revenue) is still rotating — the growth bar is where that rotation is priced. CAP = competitive advantage period; TV = terminal value.
Demonstrated net share-count change: the count has risen every year since FY22 (1,612M → 1,630M by end-FY25 → 1,632M at Jun 27, 2026) despite $8.6B of FY21–FY25 repurchases, because SBC and deal issuance outrun the buyback; H1 2026 repurchases were $221M (1.1M shares) against 4M shares issued under equity plans. $9.2B of authorization remains, but at $562.23 that retires ~16M shares (~1.0%) — a level effect, not a shrink rate, and management's revealed pace at high prices is minimal. Per-share inversion on the reported owner-earnings base: flat count +35.6%/yr; at the demonstrated +0.3%/yr net-issuance pace +36.0%/yr; with the full 320M warrant shares phasing in over years 2–6 (illustrative vesting path) +38.1%/yr — i.e., full warrant vesting adds ~+2.5 points to the required per-share compounding, arriving only if the gigawatt purchases that trigger it also arrive.
Holding owner-earnings margin ramping linearly from today's 15.8% to a mature margin by year 10, the price requires revenue compounding of +32.2%/yr (20% mature owner-earnings margin), +29.3%/yr (25%), or +26.9%/yr (30%) for ten years — year-10 revenue of roughly $450–640B depending on the margin endpoint. Management's 2027 commentary (Data Center more than doubling) is consistent with the first two years of that path; no semiconductor franchise has yet held the full decade of it.
| Required return | 8% | 9% | 10% | 11% | 12% |
|---|---|---|---|---|---|
| Implied growth | +29.9% | +32.7% | +35.4% | +37.9% | +40.3% |
Reconciliation to the Oct 7, 2026 desk run: at $633.91 that run printed ~+33.6%/yr. Rebuilt on today's identical TTM reported owner-earnings base, the same price implies +37.2%/yr — the October note did not archive its base, and a base ~$800M higher (roughly the size of the MI308 charge add-back then sitting in trailing figures) reproduces its number; the saved figure should be treated as a normalized-base bar. On the rebuilt, like-for-like model, the move from $633.91 to $562.23 (−11.3%) lowers the reported owner-earnings bar from +37.2% to +35.4%.
Assumptions: 10% required return, 10-year window, 2% terminal growth, terminal return on invested capital fading to the cost of capital, no exit multiple, solved on EV of $907.7B. Base = owner earnings (free cash flow minus stock-based compensation). SBC gets its own line: TTM SBC $1,895M = 4.6% of TTM revenue. Acquisition-intangible amortization (~$2.2B/yr) is non-cash and excluded from the cash base; its economic cost is graded in Management, not double-counted here. Plausibility verdict: the bar is extraordinary but not fantastical — it is, almost exactly, "management's 2027 commentary, extended." The two ways it breaks are visible in advance: conversion (gigawatts that stay frameworks) and duration (an AI capital-expenditure digestion year anywhere in the decade, which a 73%-terminal-value price cannot absorb).
Lisa Su (Chair & chief executive officer (CEO) since Oct 2014; Chair since Feb 2022) and Chief Financial Officer Jean Hu (since Jan 2023). The turnaround record stands: revenue $5.51B (FY14) → $34.64B (FY25), +18.2%/yr for eleven years; Data Center $6.50B → $16.64B in two years (FY23→FY25), $22.2B TTM. The load-bearing facts were re-verified against the Q2 2026 Form 10-Q and the 2026 Forms 8-K; nothing found moves the grade.
What the refresh adds. (1) The World Labs acquisition (merger agreement Sep 26, 2026) — ~$8.2B for an AI research company (spatial intelligence), payable entirely in AMD stock with the share count set by closing volume-weighted average price (VWAP) — a second very large paper deal whose return cannot be assessed from filings (no World Labs financials disclosed); it extends, rather than breaks, the Xilinx-pattern question graded B. (2) Executive compensation recalibration: the Jul 1, 2026 Form 8-K set Su's Aug 2026 long-term incentive target at $36.0M (75% performance-based restricted stock units (PRSUs) / 25% restricted stock units) vs the FY25 PRSU grant-date value of $41.67M; the 2026 PRSU design pays 0–200% on relative total shareholder return vs S&P 500 constituents with a 0/25/50% kicker on FY28-vs-FY26 non-GAAP EPS, and caps payout at 100% of target if absolute total shareholder return is negative — structure supports the B− alignment grade without improving it. (3) Board: Tim Ryan (Citi technology head, ex-PwC US chair) appointed Aug 19, 2026; Joseph Householder retired after 11+ years.
Standing debits. Four named insiders sold $428.8M of stock Oct 2024–Oct 2026 with zero open-market purchases; Su's FY25 pay of $55.2M (341:1) stepped up ~78% in a +34% revenue year; no disclosed successor or chief operating officer since Victor Peng's Aug 2024 retirement; Chair/CEO combined.
Standing credits. Buyback timing has been counter-cyclical (FY22 ~$102, FY25 ~$106 average prices vs $562.23 today); the ZT Systems deal remains the structuring template (bought for $4.41B filed consideration Mar 2025; manufacturing resold to Sanmina Oct 2025 for $2.4B cash + stock + up to $450M earn-out, as pre-announced); guidance misses are confessed early (Oct 6, 2022 pre-announcement pattern; the MI308 charge was quantified a quarter ahead in May 2025).
Charts: Qualtrim.
What is durable: the EPYC franchise (x86 server share gains across five straight record quarters, an installed enterprise/cloud footprint of 1,600+ public cloud instance types, Venice extending the lead on 2-nanometer technology); the #2-with-scale position in accelerators — AMD is the only alternative supplier with the silicon, networking (Pensando), systems (ZT design team, Helios) and software pieces in-house; and an open-ecosystem posture (ROCm open source, Open Compute Project rack designs with Meta) that hyperscalers prefer as a counterweight to a monopolist.
What is not yet durable: the software moat. ROCm has reached what management calls an inflection (3M+ models running out of the box, open-source contributions up tenfold in a year, ROCm.ai launched Jul 2026) — but NVIDIA's CUDA (Compute Unified Device Architecture) ecosystem remains the default, AMD's own risk factors concede the gap, and Data Center AI gross margins run "slightly below company average" (Hu, Q2 call) where NVIDIA prints ~75% gross margins at the corporate level. AMD's structural ceiling so far is a good #2's economics: Q2 non-GAAP gross margin 56.2% vs NVIDIA's ~75% class.
The honest moat statement: AMD's moat today is execution + incumbency-in-CPU + being the industry's chosen second source. That is a real and widening position — but second-source status means its largest customers negotiate with warrants, which is the moat's price made visible.
Q3 2026, reported Tue Nov 3, 2026, after the close — nine checkpoints: