AMD (AMD · Nasdaq)

Full report

1. Executive Summary

Key takeaways: At $515.21, Advanced Micro Devices (AMD) shares price in about $137 billion of annual free cash flow (operating cash flow minus capital spending) from fiscal 2032, roughly three times the $47 billion we model for fiscal 2031 after revenue compounds at 27% a year. Management's guidance for revenue growth above 80% in 2027 is supported by recognized Helios revenue and named customers, so we regard the business as exceptional and the price as unsupported. Our weighted value is $208 a share and our fair value range is $190–$320, which is why we rate the shares Underweight. Even our bull case, worth $305 a share, would return 9.3% a year from today's price against our 13.2% cost of equity.

AMD reported Q3 2026 revenue of $13.4 billion on 3 November, up 45% from a year earlier, with Data Center revenue of $8.5 billion (up 97%) and a Q4 revenue guide of about $17 billion, 4% above the $16.3 billion consensus. The shares fell 38% from their 6 October closing high to 5 November and then regained 27% by 10 November, so the price has moved far more than the reported fundamentals. This report builds on our event study of 11 November 2026, which measured how AMD reacted to 13 comparable earnings reports.

The price embeds revenue and cash flow well beyond management's own path, as the expectations gap in the table below sets out.

Over the next one to three months we expect the market to trade first the AI-capital-spending and policy debate, which can drive a repricing of the whole sector, then Nvidia's results on 17 November, and then AMD's Q4 guide on 2 February 2027. Our event study found that AMD's shares lagged the S&P 500 over the two sessions after 9 of 13 comparable reports (median −5.3%) and by a median of −12.0% over the following 21 sessions, so we expect Sideways to lower with very high volatility. The upside invalidation is fiscal 2027 free cash flow above $20 billion, against the $10 billion we model, and the downside invalidation is inventory above $14 billion at fiscal year-end 2026 without a Q1 2027 revenue guide above $20 billion.

Our view: our largest disagreement is that the price requires about $137 billion of annual free cash flow while our bull case reaches $74 billion. The market may be extrapolating management's 2027 guide and discounting cash flows at a return nearer 7% than 13%, and it is giving little weight to the working capital, financing and dilution that growth is consuming. We would learn we are wrong first from the Q1 and Q2 2027 reports if free cash flow margin exceeds 15% while inventory stays flat, and conclusively from fiscal 2027 free cash flow above $20 billion.

The three bull points rest on recognized revenue and signed agreements, and the three bear points rest on arithmetic and balance-sheet trends.

• Bull: Data center AI revenue is now recognized and contracted. Q3 Data Center revenue was $8.5 billion (up 97%), first Helios racks shipped for revenue, and agreements cover up to 14 GW (OpenAI 6, Meta 6, Anthropic 2).
• Bull: Server processor share keeps rising, to 46.2% of x86 server revenue in Q1 2026 (Mercury Research), with more than 240 Venice platforms committed.
• Bull: Management expects 2027 operating expenses to grow less than half as fast as revenue, which should lift profits faster than sales.
• Bear: The price needs $137 billion of annual free cash flow from fiscal 2032, against $6.7 billion in fiscal 2025.
• Bear: Working capital and customer financing absorb the growth, with a 4% Q3 free cash flow margin, inventory of $11.5 billion (up 45% in nine months), up to $5 billion for Anthropic and warrants over 19% of shares.
• Bear: Policy, financing and competition can reset AI capital spending, as the 5 November sector sell-off, Nvidia's Rubin ramp in the same quarters and SpaceX's August choice of Nvidia show.

2. Business Overview

AMD designs processors and accelerators and relies on outside manufacturers for wafers, substrates and advanced packaging, which management said it has secured for 2027 growth. Its products are EPYC server processors, Ryzen client processors, Instinct data center accelerators (including MI455X in the Helios rack-scale system), Radeon graphics, semi-custom chips for game consoles, and adaptive chips from the 2022 Xilinx acquisition. Customers include the five companies running first Helios racks (Anthropic, Meta, Microsoft, OpenAI and Oracle), server makers such as HPE, Dell, Lenovo and Super Micro, and PC and console makers. We could not obtain customer-level concentration percentages from the filings available to us, so we rely on the named agreements.

Data Center now generates 60% of revenue and nearly all of the growth, as the segment table shows.

By customer location in fiscal 2025, the United States provided 33% of revenue, China 22% ($7.8 billion, up 24%), Taiwan 15%, Singapore 12% and other countries 18%. Because non-US revenue is two-thirds of the total, our cost of capital adds a revenue-weighted country risk premium of 0.48% (see appendix). China exposure also carries export-license risk, which AMD flagged in 2025 for its MI308 accelerator.

3. Business and Competitive Analysis

3.1 Cycle position and operating indicators

AMD sits early in a capital-spending build that its own balance sheet increasingly funds. Q3 inventory reached $11.5 billion, up 45% in nine months and equal to 169 days of cost of sales (154 days at fiscal year-end 2025), and management said demand exceeds supply and that it will keep investing in inventory and capacity for 2027. AMD's own history shows how quickly the cycle can turn: revenue fell 3.9% in fiscal 2023 after rising 43.6% in fiscal 2022, and GAAP operating margin fell from 22.2% to 1.8% over fiscal 2021–2023, partly because of Xilinx amortization. The inventory is being built against signed deployments, which raises the cost of any slippage (operating indicators are in the appendix).

3.2 Competition and moat

AMD competes with Intel in server and PC processors and with Nvidia and custom chips in accelerators. Intel is weakening: it guided Q4 revenue to $15–16 billion on 30 October and warned that its PC slowdown would last through the first half of 2027. In accelerators, AMD says Helios delivers 15% more compute and 50% more memory than Nvidia's Vera Rubin NVL72 rack, but these are company figures with no third-party benchmark yet, and Rubin ramps in the same quarters. SpaceX said in August that it would buy only from Nvidia, and an analyst on the Q3 call asked whether custom chips and competitors' financing guarantees would favor others in 2028.

Technology leadership has translated into share and revenue in servers but not yet into pricing power in accelerators, so we rate the moat Narrow.

4. Management, Governance, and Capital Allocation

Lisa Su has been chair and CEO throughout the period and Jean Hu is chief financial officer. We found no 2026 short-seller report, litigation or regulatory investigation at AMD. A board change was disclosed on 19 August (a director retired and Tim Ryan joined), and Su sold shares in February and March under a Rule 10b5-1 plan (reported as about $26.8 million and $16.9 million). The governance question we weigh most is structural: AMD has issued warrants to two customers and committed to invest in a third, which links its revenue to the financing of its buyers.

Capital allocation has favored reinvestment and stock-funded acquisitions over buybacks, and dilution is the weakest element.

5. Financial Analysis and Earnings Quality

5.1 Financial trends

Revenue has grown at a 20.5% compound rate from fiscal 2021 to fiscal 2025 while GAAP operating margin has recovered from a trough of 1.8% to 10.7%, and free cash flow has returned to 19.4% of revenue after a low of 4.9% in fiscal 2023.

AMD trades at 110.5 times trailing GAAP earnings against 22.6 times for Nvidia, 25.0 times for TSMC and 40.8 times for Broadcom, even though its earnings are depressed by acquisition amortization and the warrant charge (see appendix).

5.2 Earnings quality

We grade AMD's earnings credibility B. The cash tests are clean: accruals were negative in each of the last three fiscal years (−4.6% of average assets in FY2025), operating cash flow was 1.8 times net income, and the Beneish M-Score, a screen for earnings manipulation where values above −1.78 flag risk, was −2.68. The two flags are mild but moving the wrong way: inventory days rose from 130 in FY2023 to 169 in Q3 2026, and Q3 free cash flow fell to $554 million (a 4% margin) as inventory and receivables absorbed cash. The non-GAAP figures that headline AMD's reports also exclude $1.5 billion of stock compensation and $1.2 billion of warrant charges over nine months.

Because AMD invests in and issues warrants to its customers, we measure free cash flow before and after those commitments: $4.68 billion in nine months of 2026, $3.17 billion after stock compensation, and negative $1.83 billion after the up to $5.0 billion Anthropic investment (see appendix). We could not obtain receivables days by customer, the terms of the Anthropic investment or any off-balance-sheet guarantees.

6. Valuation

Key takeaways: Four methods place AMD's fair value at $190–$320 against a price of $515.21, and none of them reaches the price, including the bull scenario at $305. The conclusion holds across discount rates, because the probability-weighted value is $310 even at a 10% WACC. We label the shares Expensive.

6.1 What the price implies

A reverse DCF solves for the cash flows that justify today's price. At a 13.0% WACC (weighted average cost of capital, here equal to our cost of equity) and 3.0% terminal growth, the enterprise value of $847 billion requires free cash flow of $137 billion a year from fiscal 2032. A second test, the present value of growth opportunities (PVGO, the part of the price that pays for growth not yet earned), shows that 88% of the price is growth: zero-growth earnings of $8.0 a share (our fiscal 2026 non-GAAP estimate) are worth $62 at a 13.0% discount rate. Revenue of $457–686 billion at 30–20% free cash flow margins is 2.7–4.0 times our base case (see appendix).

For perspective, our base case already assumes a 27% revenue CAGR from a $52 billion base, which is better than the 97th percentile of historical outcomes (Mauboussin's base-rate research finds that only 1–3% of companies growing above 20% sustain it for ten years). We justify it with signed agreements for up to 14 GW and an analyst-cited range of $15–20 billion of revenue per gigawatt, but it remains a demanding path.

6.2 Scenario DCF

The three scenarios differ in AI capital spending, share, margins and warrant vesting. Free cash flow margins come from an explicit bridge (operating income, tax at 15%, capital spending and working capital at 20–30% of incremental revenue), which is why they start near 10% in fiscal 2027 and rise toward 27%.

The bear scenario models an AI capital-spending digestion, with revenue falling from $72 billion in fiscal 2027 to $62 billion in fiscal 2029, because every previous semiconductor expansion has been followed by an inventory correction and AMD's own revenue fell in fiscal 2023. Its evidence grade is Medium: inventory of $11.5 billion at 169 days, a 4% free cash flow margin in Q3, the 5 November policy sell-off, and customer financing concerns. The bull case delivers management's 2027 guide and a $265 billion revenue base by fiscal 2031. Its evidence grade is Medium-strong: recognized Helios revenue, a Q4 guide above consensus, five named customers and demand above supply. Under the house rule a one-step gap favoring the bull case gives weights of 20% bear, 50% base and 30% bull; equal grades would give $196 and a two-step gap $213, so the weighting moves the value by no more than 6%. Across a WACC of 12–14% and terminal growth of 2–4% the base case is worth $176–$253, so no cell reaches the price (see appendix).

6.3 Cross-checks and composite range

Earnings power value (EPV, normalized earnings capitalized with no growth) is $57 a share and acts as a floor. The economic value added (EVA) method, which values the franchise as invested capital plus the present value of returns above the cost of capital as they fade over ten years, gives $82–$96 at a 27% return on capital. A peer test applies the interquartile range of peer trailing P/E multiples (13.5–25.0x) to consensus fiscal 2027 earnings of $15.72 and gives $212–$393. A Monte Carlo run with growth, margin and discount-rate distributions gives a median of $230, and 98% of the 5,000 runs are below the price.

The composite implies an expected return of −60% to the weighted value. Our view: the valuation rests less on this quarter's revenue than on whether AMD can sustain a 25–30% free cash flow margin on more than $150 billion of revenue through a full cycle, and the price leaves that outcome no room to fall short. AMD's current earnings are not at a cyclical low, so we normalize earnings power value with a 25% operating margin rather than applying a mid-cycle multiple to peak profit. If revenue, gross margin, inventory days, Helios shipments or customer capital spending breach the thresholds in section 9, we would cut our long-term assumptions further.

7. Analyst View Summary

Sell-side opinion is the opposite of ours: 51 of 60 analysts rate the shares buy, none rate them sell, and Street targets range from $525 to $1,250 with a mean of $655.11, 27% above the price (Robinhood, 9 November). Oppenheimer upgraded the shares to "Outperform" with a $600 target on 9 November, citing the Q3 call and modeling 2027 Data Center revenue growth of 135%. The only published estimates we obtained pre-date the Q3 report: consensus fiscal 2027 earnings of $15.72 and fiscal 2028 earnings of $22.82 (6 October), so post-print revisions are probably higher and we flag that as a gap. Our fiscal 2027 revenue of $88 billion is 6% below management's guide and our earnings of about $14 are 10% below consensus (see appendix).

The expectations gap in section 1 is where we differ: the Street and we agree that revenue growth will be extraordinary and disagree on what that revenue is worth, because we use a 13.0% discount rate, include working capital and warrant dilution, and cap margins at the top of AMD's own history.

8. Recent News and Catalysts

AMD's recent news has been dominated by the sector, and the company-specific events mostly support revenue while raising financing questions. The largest sessions since July were the 1.5% gain on the Anthropic agreement (22 July), the 10.9% fall on Intel's guidance (30 October), the 17.3% fall after the midterms (5 November) and the 13.3% rebound on an Oppenheimer upgrade and a Google Cloud agreement (9 November); the full table is in the appendix.

The next scheduled tests are Nvidia's results and AMD's own Q4 report, with product and policy milestones in between.

8.1 Forecast register

Each forecast below is dated, scoreable and kept unchanged once published.

9. Investment Conclusion, Counter-Case, and Position Sizing

9.1 Conclusion

Key takeaways: We rate AMD Underweight because the shares at $515.21 trade 61% above the top of our $190–$320 fair value range and 69% above our bull case. The business is delivering, which is why we expect the shares to stay volatile rather than fall steadily, and our direction for the next one to three months is Sideways to lower with very high volatility. Holders should reduce, and we would turn Neutral near $320 and call the shares Cheap only below about $160.

Decision triad. Intrinsic value view: Expensive. 1–3 month market trading direction: Sideways to lower, very high volatility. Investment action: Underweight, holders reduce. The three judgments differ for clear reasons. The intrinsic value label is Expensive because the price is more than 1.15 times the top of the range. The direction is Sideways to lower rather than Lower because management's guidance, an Oppenheimer upgrade and a 27% rebound can sustain the shares for a while, while the Q4 report has historically been followed by underperformance. The rating is Underweight because the matrix for an Expensive label with very high uncertainty is Avoid, and no earnings-credibility veto applies. If this were a cash decision today, we would not buy the shares at $515.21, because even our bull case returns less than the 13.2% we require.

9.2 Counter-case and pre-mortem

Our rating fails if the market accepts a return nearer 8% than the 13% we use, or if Helios revenue exceeds our bull case with higher cash conversion than we model. The strongest counter-evidence is Medium-strong: AMD recognized its first MI455X and Helios revenue in Q3, guided Q4 above consensus, and says it has secured supply for 2027. We responded by raising the bull weight to 30% and by choosing Sideways to lower rather than Lower for the one-to-three-month direction. We did not change the label, because every scenario remains below the price.

The prior ABiS event study (11 November 2026) is the only earlier work on AMD and contains no equity-research forecasts, so there is nothing to review.

9.3 Position thinking

The probability-weighted expected return is −60%, and 98% of Monte Carlo runs value the shares below the price. All three scenarios lie below the price, so the upside-to-downside ratio is not defined and the Kelly-style sizing the model computes is zero. Our assessment: the risk-reward supports no long position, and we advise holders to reduce. A short position is not our recommendation, because the shares' beta of 1.65 (3.2 on recent daily data), a 27% three-day rebound and policy-driven swings make the path hard to hold.

9.4 Monitoring list

• Upside invalidation: free cash flow margin above 15% in Q1 or Q2 2027, or fiscal 2027 free cash flow above $20 billion.
• Downside invalidation: inventory above $14 billion at fiscal year-end 2026 without a Q1 revenue guide above $20 billion, a Helios delay of more than one quarter, or enacted limits on data-center construction or chip exports.
• Data Center revenue below $9.5 billion in Q4 2026, or growth under 90% year on year, would weaken the bull scenario.

9.5 Confidence

Appendix: supporting exhibits

Operating indicators and recent news that support chapters 3 and 8 are set out first.

AMD's multiple is the highest of the large semiconductor peers except Arm.

Free cash flow turns negative once the Anthropic commitment is counted.

The price requires three to four times our base-case revenue.

No combination of discount rate and terminal growth in the grid reaches the price.

The Street and ABiS differ most on what the revenue is worth.

Quarterly results show revenue growth accelerating while gross margin (quarterly, GAAP) has stayed near 54%.

The cost of capital follows the house defaults, and the single largest sensitivity is the discount rate.

Key valuation assumptions are in the scenario exhibit above and the cost of capital build here. Our checks of the financial inputs found no material problems; the one item noted is the M-Score of −2.68, which is below the −1.78 threshold.

Method and data notes

Method and data notes: price is the 10 November 2026 close of $515.21 (Yahoo Finance), the last official close before work started on 11 November. Shares are 1.661 billion diluted weighted-average shares from the Q3 2026 10-Q, held flat in the first year and diluted afterward by stock compensation, customer warrants and the World Labs acquisition. Net cash of $8.7 billion is cash and short-term investments of $15.8 billion less debt of $7.1 billion at 26 September 2026. The explicit forecast covers fiscal 2027–2031 and excludes the Q4 2026 stub. The cost of capital uses a 4.98% risk-free rate (10-year Treasury par yield, 10 November 2026), a beta of 1.65 (two-year weekly regression, raw 1.96, adjusted with the 0.67 × raw + 0.33 formula), the 4.5% equity risk premium plus a 0.48% revenue-weighted country premium from Damodaran's January 2026 table, and a 13.0% WACC after rounding 13.2% to the nearest half point. Terminal growth is 3.0%. All discounted cash flow, earnings power, economic value added and Monte Carlo values come from our valuation model, with assumptions saved for re-running. Our calibration rule places the price in the "significantly overvalued" band (above 1.5 times the top of the range), which we publish as Expensive. Tax in the free cash flow bridge is 15%, an ABiS assumption, and working capital intensity of 20–30% of incremental revenue is also ours. We could not obtain customer concentration, post-Q3 consensus, Morningstar's rating or any Q3 non-GAAP operating expense history. This is historical and model-based analysis and is not investment advice.

Inputs: price $515.21 (10 November 2026 close, Yahoo Finance), risk-free 4.98%, beta 1.65, equity risk premium 4.5%, WACC 13.0%, terminal growth 3.0%.

Disclaimer: not investment advice. ABiS Capital Research is not a licensed adviser, and nothing here is a recommendation to buy or sell any security.

Sources

Industry appendix: semiconductors. Sources and timestamps: AMD Q3 2026 results release (3 November 2026, via MFN) and call transcript (Motley Fool, 4 November 2026); AMD 10-K (filed 3 February 2026) and 10-Q (filed 3 November 2026) and earlier 10-K (31 January 2024) via SEC XBRL, retrieved 11 November 2026; Robinhood quotes, fundamentals, analyst ratings and earnings calendar (9–10 November 2026); Yahoo Finance adjusted closes and earnings calendar (retrieved 11 November 2026); U.S. Treasury par yield curve (10 November 2026); Damodaran country risk premiums (January 2026); TIKR on consensus earnings (6 October 2026); CNBC on Oppenheimer's upgrade (9 November 2026); Stocktwits on the Q4 consensus (4 November 2026); Motley Fool on the 5 November sell-off; Mobile World Live and The Next Web on Anthropic (July 2026); Reuters via The Star on the $4.75 billion bond sale (14 August 2026); crnasia and Verdict on World Labs (September 2026); Mercury Research via Tom's Hardware and technews.tw on processor share; Futurum via CNBC and Data Center Knowledge on Helios pricing and timing; MarketBeat on Lisa Su's share sales (February–March 2026); Honolulu Star-Advertiser on Intel (30 October 2026); ORTEX on short interest (2 November 2026).