AMD is a good company in a bad setup, two sessions before a binary event. The business is accelerating for real: trailing revenue up 37.8% to $37.5B, data centre up 57% year on year to $5.8B last quarter, $8.5B of net cash, and an MI400 and Helios roadmap that finally gives hyperscalers a second credible supplier. None of that is in dispute. The price, the share count and the timing are. The stock trades at 107.5x EV/EBITDA, nearly four times NVIDIA's 28.6x, on a 14.4% operating margin against NVIDIA's 65.6% and an 8.1% return on equity against NVIDIA's 114.3%. Above that price sits a warrant issued to OpenAI in October 2025 for up to 160 million shares at a $0.01 exercise price — about 10% of the share count — whose tranches vest against GPU purchase milestones and against AMD share-price targets running up to $600. The dilution arrives precisely if the bull case works. Meanwhile seven insiders sold 690,258 shares for $221.9M across 101 transactions in 180 days with zero open-market purchases. The stock enters Tuesday's print 18.6% below its 52-week high, below its 20-day and 50-day averages, with a negative MACD, after handing back 7.7% intraday from Friday's high. The business we would own. This entry we would not.
AMD designs high-performance processors and graphics chips and has them manufactured by third-party foundries. Four segments matter, but only one currently sets the share price.
Data Center is what sets the price. It generated $5.8 billion in Q1 2026, up 57% year on year, and is now the majority of revenue. It contains two very different businesses that investors routinely conflate. EPYC server CPUs, where AMD has taken share from Intel for eight consecutive years and has real pricing power. And Instinct AI accelerators — MI300X, MI325X, and the new MI400 series — where AMD is the challenger to NVIDIA and competes largely on price, memory capacity and bandwidth rather than on software ecosystem.
That distinction is the crux of the case. The CPU franchise is a durable share-gain story with defensible economics. The accelerator franchise is a land grab against an incumbent whose CUDA software stack remains the industry default. Today's price assumes both work. At its Advancing AI event, management unveiled the full MI400 line, the Helios rack-scale platform and an MI500 roadmap preview; the flagship MI455X is specified at 40 petaflops of FP4 compute, 432 GB of HBM4 memory and 19.6 TB/s of bandwidth.
Client (Ryzen PC processors), Gaming (console semi-custom and Radeon) and Embedded (the Xilinx FPGA franchise) make up the rest. They are cyclical, slower-growing, and matter mainly as a source of cash and as a cushion when the AI narrative wobbles.
| Segment | Revenue |
|---|---|
| Data Center | $5.8B in Q1 2026, +57% year on year. EPYC server CPUs (durable share gains against Intel) plus Instinct AI accelerators (challenger position against NVIDIA). The majority of revenue and effectively all of the valuation. |
| Client | Ryzen desktop and notebook processors. Steady share gains in a low-growth PC market. Cash generative, not a valuation driver. |
| Gaming | Semi-custom console silicon and Radeon graphics. The most cyclical segment, tied to console cycles. |
| Embedded | The Xilinx FPGA franchise acquired in 2022. High gross margin, industrial and communications end markets, in slow recovery. |
| Metric | Value | Signal |
|---|---|---|
| Revenue (TTM) | $37.45B (+37.8% YoY) | Strong acceleration |
| Earnings growth | +91.2% | But see the base effect in Risks |
| Gross margin | 53.1% | NVIDIA 74.1%, Broadcom 76.3% |
| Operating margin | 14.4% | NVIDIA 65.6% — a 51-point gap |
| Net margin | 13.4% | NVIDIA 63.0% |
| Return on equity | 8.1% | NVIDIA 114.3%, Broadcom 37.3% |
| Return on assets | 3.6% | Capital-intensive for the returns produced |
| Cash | $12.35B | Fortress |
| Total debt | $3.87B | Net cash $8.48B |
| Debt / equity | 0.06 | No leverage risk |
| Market cap | $776.41B | Enterprise value $798.59B |
| Shares outstanding | 1.631B (1.791B fully diluted for the warrant) | The 160M-share OpenAI warrant is about 9.8% dilution at a $0.01 strike |
| EV / EBITDA | 107.5x | NVIDIA 28.6x, Broadcom 44.4x, Marvell 56.8x |
| EV / Revenue | 21.3x | NVIDIA 18.6x on far better margins |
| Trailing P/E | 158.7x | On GAAP EPS of roughly $3.00 |
| Forward P/E | 34.5x | NVIDIA 15.6x, Broadcom 20.0x, TSMC 18.7x |
| Price / book | 11.5x | Book value $39.55 per share |
| PEG ratio | 1.25 | Best bull argument — but its denominator is the same 91.2% growth the base effect discredits |
| Free cash flow yield | 0.87% – 1.10% | 91x to 115x free cash flow. Two published figures disagree by 27%, so we give the range, not a point estimate |
| Beta | 2.47 | Moves roughly 2.5x the index |
| Dividend | None | All capital reinvested |
| Analyst mean target | $573.15 | +20.4% — 28 Buys vs 8 Holds |
| Q1 2026 (5 May) | Revenue $10.3B (+38%), data centre $5.8B (+57%) | Guided Q2 to $11.2B |
| Q2 2026 consensus (4 Aug) | Revenue ~$11.32B (+47.3%), EPS $1.55–$1.62 | The street sits above the company's own $11.2B guide |
| What the price implies | ~18% FCF growth, forever | Reverse discounted cash-flow: at a 19.3% cost of capital, $776B requires free cash flow to compound at about 18% a year in perpetuity. At a gentler 1.5 beta it still needs ~12.5% |
| Cost of capital (our estimate) | 19.3% | Beta 2.469 against a 4.7% risk-free rate. Debt is negligible in the weighting |
| Interest coverage | 41.2x | EBIT $5.39B against $131M of interest expense (FY2025 annual report). Debt service is a non-issue |
| Date | Insider | Type | Shares | Value |
|---|---|---|---|---|
| 2026-07-15 | Mark Papermaster | Sale | 6,000 | $3.34M |
| 2026-06-15 | Mark Papermaster | Sale | 6,000 | $3.22M |
| 2026-06-10 | Lisa Su | Sale (27 lots) | 125,000 | $57.6M |
| 2026-06-02 | Nora Denzel | Sale (2 lots) | 10,447 | $5.45M |
| 2026-05-20 | Forrest Norrod | Sale (9 lots) | 19,487 | $8.41M |
| 2026-05-13 | Lisa Su | Sale (22 lots) | 125,000 | $55.7M |
| 2026-05-08 | Paul Darren Grasby | Sale | 24,376 | $10.83M |
| 2026-04-24 | Mark Papermaster | Sale | 31,320 | $10.96M |
| 2026-03-12 | Lisa Su | Sale | 125,000 | $16.9M |
| 2026-02-17 | Jean Hu | Sale (9 lots) | 19,956 | $4.02M |
| 2026-02-11 | Lisa Su | Sale | 125,000 | $26.8M |
bearish
39.77M shares short against a 1.62B float. Cheap to borrow, barely over one day to cover. There is no squeeze setup here in either direction, and no meaningful bear position to squeeze.
Unusual Activity: The single most anomalous contract in the chain is the $425 put: 7,173 contracts traded against 315 of open interest, a 22.8x multiple, at 94.2% implied volatility. The $425 strike sits 0.2% above the 29 July low of $424.03 — money is paying up for protection at precisely the recent floor. Nineteen put strikes were flagged for unusual volume against eleven call strikes. Other large anomalies, all puts: $465 (3,777 vs 192 OI, 19.7x), $490 (1,950 vs 99, 19.7x), $487.50 (1,222 vs 65, 18.8x), $457.50 (1,065 vs 80, 13.3x). IMPORTANT CAVEAT, and it changes the reading: every one of these contracts expires Monday 3 August, the day BEFORE the print. This is not an earnings hedge. It is either positioning for Monday alone — ISM manufacturing at 16:00 CET and Palantir's results after the close, both read-across events for the AI complex — or it is roll activity into later expiries our feed does not carry. We flag the pattern; we do not claim to know which.
| RSI (14) | 45.2 |
| EMA 20 | $503.96 |
| EMA 50 | $486.27 |
| EMA 200 | $342.99 |
| MACD | -10.320 |
| Signal | -1.320 |
| ATR (14) | $40.33 |
The stock sits below its 20-day ($503.96) and 50-day ($486.27) averages and 38.8% above its 200-day ($342.99): a broken short-term trend inside an intact long-term one. RSI at 45.18 is neutral. MACD at -10.32 is below its signal line at -1.32, so momentum is negative and not yet turning. Average true range is $40.33, 8.5% of the share price, which is what a double-digit implied earnings move looks like in practice. The most informative bar is Friday 31 July: open $510.28, high $515.62, close $476.15. Handing back 7.7% from the high on the last session before a print tells you who was selling into strength. Levels that matter: $486 then $504 on the way up, the two moving averages in that order; $455, then $429.56 and the 29 July low of $424.03 on the way down. The heaviest put activity in the chain clusters on that last level.
| Indicator | Value | Signal |
|---|---|---|
| Market regime | RECOVERY 0.72 / risk-on 57.9% | Supportive but not euphoric |
| VIX 9-day | 13.05 | The index market prices almost no risk for the next two weeks |
| Semiconductor group, last week | -3.7% | Rallied 6.9% on 30 July, gave it back |
| AMD vs its own group | -8.8% vs -3.7% | Underperformed the sector by 5 points in one week |
| 10-year Treasury | 4.745% | Long-duration growth is the first casualty when rates rise |
| 30-year Treasury | 5.275% | Highest close since July 2007 — a direct headwind to 100x multiples |
Regime: recovery
AMD is a maximum-duration asset: 2.47 beta, no dividend, and a valuation that discounts cash flows many years out. The macro backdrop of the past week has been quietly hostile to exactly that profile. The 30-year Treasury closed at 5.275%, its highest since July 2007, and the 10-year at 4.745%. When the discount rate on long-dated cash flows rises, the stocks that lose most are those whose value sits furthest in the future — which is precisely a company at 107x EV/EBITDA. Note too that three Federal Reserve officials dissented in favour of a rate hike on 29 July, so the familiar reflex of buying weak macro data on the expectation of cuts is not currently reliable. That is the backdrop into which AMD reports.
A high-beta, high-multiple asset reporting binary results in two sessions, in a broken short-term trend, with unanimous insider selling, a defensively positioned options chain, and a 160-million-share warrant struck at a cent sitting above the price. The balance sheet is clean; the share count is not. Nothing here threatens solvency. Everything here is about what you pay and when.
Why downgrade the same company from B to C+ while its revenue accelerates? Because a grade is not a verdict on a business, it is a verdict on the risk and reward available to a buyer today. In February we graded AMD B at $207, with a forward multiple that left room. Since then the business improved and the share price tripled off its low. The four eliminatory criteria in our A-grade screen now read: guidance raised — pass; five consecutive earnings beats — fail, our verified history shows three beats and one marginal miss across four quarters; forward P/E under 35x — pass, at 34.45x, by half a point; price within 3% of its 20-day average — fail, at 5.52% below. Two of four fail, so the A grades are off the table by construction, and the composite lands at 50 out of 100. The facts that made the company more attractive are the same facts that made the entry less attractive. We grade the entry.
Two plans, chosen on Wednesday morning by what the print actually says rather than by what we expect it to say. PRIMARY — the pullback. If the results are good but the stock sells into $430–441 while holding above the 29 July low of $424.03, that is a confirmed catalyst bought at a price the event itself created. Entry zone $430–441 (limit only, hard cap $441), stop $412, first target $486 at the 50-day, second target $520 — scaling out into $512–515 rather than resting size at $520, which sits 0.85% above Friday's $515.62 swing high. Reward-to-risk runs 1.64 at a $440 fill, 1.55 at the $441 cap and 3.11 at $430, so where you get filled decides most of the outcome. Be aware the $412 stop is 0.69 times the $40.33 average true range — a deliberately tight stop on a 2.47-beta name in its most volatile window, which buys a defined loss at the cost of a higher stop-out rate. Size in two tranches for that reason. The options market has already marked this zone: the heaviest unusual put volume of the whole chain sits at the $425 strike. ALTERNATIVE — the reclaim. If the print is strong and the stock closes back above its 20-day at $503.96, enter there with the stop under the 50-day at $483, which is the genuine invalidation of a reclaim: 1.96 to $545 and 3.85 to $584.73. Half size, because a reclaim after a broken trend fails more often than it works. WHAT WE DO NOT DO: buy before Tuesday's close, hold through the print, or use a market order at Wednesday's open. On a name with an 8.5% average daily range, the opening auction is where retail pays the spread.
This analysis is provided for informational and educational purposes only. It does not constitute financial advice, investment recommendation, or solicitation to buy or sell any security.
Past performance is not indicative of future results. All investments involve risk, including the possible loss of principal. Always conduct your own research and consult a licensed financial advisor before making investment decisions.
Data sourced from real-time market data, Yahoo Finance, SEC EDGAR, and public market data. Accuracy is not guaranteed.