AMD$201.99-0.9%Cap: $329.3BP/E: 77.452w: [=======|---](Mar 28)
Verdict: KEEP
No case for removal. +39.5% trailing alpha (strongest in selectable set), no bear catalyst in the 15-week window, and 48.7% idiosyncratic variance means removing AMD is ≈71% a bet against tech — incoherent when we're long QQQ. Zero informational edge on a $329B, 49-analyst name. Market is efficiently priced. Consensus is correct.
Factor Decomposition
Regression (250 trailing days):
| Factor | Beta | Variance % |
|---|---|---|
| XLK (Technology) | +2.43 | 70.9% |
| SPY (Market) | -1.25 | -22.8% (collinear with XLK) |
| MTUM (Momentum) | +0.14 | 3.2% |
| Idiosyncratic | — | 48.7% |
R-squared 51.3%. Alpha +39.5% annualized. Idio vol 45.2%.
Interpretation: AMD is a 2.4x leveraged tech bet. XLK explains 71% of variance — well below the 75% idiosyncratic target. Removing AMD from the basket is functionally ≈71% a bet against tech, which we're long via QQQ. Self-defeating unless you have a specific idio bear thesis. The low idio variance means limited room for stock-specific under/outperformance. AMD moves with tech.
Orthogonal Sharpe: 39.5% / 45.2% = 0.87. Within the plausible 0.5-1.0 range. Not suspiciously high — this is real idio alpha, not misattributed factor exposure.
Net market beta: The regression shows SPY beta -1.25 and XLK beta +2.43. Because XLK itself has ≈1.0 SPY beta, net market beta ≈ -1.25 + 2.43 = ≈1.18, consistent with yfinance's reported 2.02 SPY beta (longer lookback captures more variance from market factor directly). AMD amplifies QQQ in both directions.
FY2025 Financials (10-K filed Feb 4, Q4 transcript Feb 3)
| Metric | FY2025 | FY2024 | YoY |
|---|---|---|---|
| Revenue | $34.6B | $25.8B | +34% |
| Non-GAAP GM | ≈55% | ≈51% | +400 bps |
| Non-GAAP EPS | $4.17 | $3.31 | +26% |
| GAAP EPS | $2.65 | $1.00 | +165% |
| Free Cash Flow | $6.0B+ | $3.3B | +82% |
| Cash & Investments | $10.6B | — | — |
| Data Center Revenue | ≈$16.8B | $12.6B | +33% |
Q4 2025 clean baseline: $10.3B revenue, $1.53 non-GAAP EPS (+16% beat vs $1.32 consensus). But Q4 included $390M MI308 China shipments (not in guide) and a $306M inventory reserve release boosting gross margin. Ex-China, ex-reserve: ≈$9.9B at ≈55% GM. This is the real run-rate for modeling Q1.
GAAP vs non-GAAP gap: Trailing P/E 77.4x is distorted by Xilinx intangibles amortization ($4.7B remaining, ≈$1.5B/year). Non-GAAP trailing P/E: $202 / $4.17 = 48.4x. Forward P/E 18.8x uses NTM consensus of ≈$10.74.
Q1 2026 EPS Bridge
Revenue guidance: $9.8B ± $300M (includes ≈$100M MI308 China).
| Line | Amount | Notes |
|---|---|---|
| Revenue | $9.8B | Midpoint; ex-China ≈$9.7B |
| Gross Profit (55% GM) | $5.39B | Jean Hu's "approximately 55%" historically means 55-56% |
| Operating Expense | ≈$3.05B | |
| Operating Income | $2.34B | 23.9% margin |
| Other Expense | $35M | |
| Pre-Tax Income | $2.305B | |
| Tax (13% ETR) | $300M | |
| Net Income | $2.005B | |
| Diluted Shares | 1.65B | Guided; pre-Meta warrant vesting |
| EPS at guide midpoint | $1.22 | |
| Street consensus | $1.27 | Implies ≈56% GM or ≈$10B revenue |
Consensus bakes in a modest beat. AMD's conservative guidance pattern supports this: 3 of 4 recent quarters beat, and Hu's margin guidance has undershot by ≈100 bps consistently. A normal beat delivers $1.27-$1.33.
Sequential (Q4→Q1): -5% guided. Ex-China: $9.7B vs $9.9B = -2% QoQ. Seasonal pattern, nothing unusual. Q1 has historically been AMD's weakest quarter.
Q1 Earnings Scenarios (May 5)
| Scenario | Prob | EPS | Driver |
|---|---|---|---|
| Beat + raise | 55% | $1.30-1.40 | DC GPU ramp on track, EPYC share gains, GM 56%+ |
| In-line | 25% | $1.24-1.30 | Guide met, no surprise |
| Miss / soft guide | 20% | $1.15-1.24 | MI355 supply constraint, China MI308 < $100M |
None justify removal. Even the 20% miss scenario is a modest ≈$0.05 shortfall — not the kind of structural underperformance (negative trailing alpha, management deterioration) that characterizes our removes.
Meta Warrant (8-K Feb 24, 2026)
160M shares at $0.01/share. Vesting tied to Instinct GPU purchases: 1 GW (binding) through 6 GW (full). Price hurdles escalate to $600 for the final tranche. Exercise window through Feb 23, 2031.
Dilution math:
| Milestone | Shares | Dilution | Implied Price |
|---|---|---|---|
| 1 GW (binding) | ≈26.7M | 1.6% | Current (≈$200) |
| 3 GW | ≈80M | 4.8% | ≈$300-400 |
| 6 GW (full) | 160M | 9.7% | $600 |
Full vesting requires $600 stock price (3x current). Probability-weighted dilution at $202 is well under 9.7%.
Accounting (ASC 606): Warrant fair value is contra-revenue — reduces reported GPU revenue from Meta over the vesting period. This creates a reporting headwind that analysts have already modeled. Market absorbed this on Feb 24 with no lasting impact.
Economics: At full dilution, 160M shares x $200 = $32B dilution cost against potentially $80-180B in cumulative GPU revenue (6 GW at estimated $13-30B/GW). The warrant is shareholder-positive if Meta follows through — you don't give away $32B of equity unless the revenue justifies it.
Insider Activity
| Date | Insider | Action | Amount | Signal |
|---|---|---|---|---|
| Mar 12 | Lisa Su (CEO) | Sale | 85K shs / $16.9M | 10b5-1 plan |
| Mar 11 | Grasby (Officer) | Sale | 7.5K / $1.5M | Routine |
| Mar 4 | Papermaster (CTO) | Sale | 3K / $607K | Routine |
| Feb 27 | Hu (CFO) | Acquire | 132K / $26.7M | RSU vest |
| Feb 27 | Guido (Officer) | Acquire | 45K / $9.1M | RSU vest |
| Feb 24 | Papermaster (CTO) | Gift | 413K / $83.5M | Estate planning |
| Feb 13 | Norrod (EVP) | Acquire | 77K / $15.5M | RSU vest |
Standard RSU vesting → acquire → sell-for-taxes pattern. No unusual clusters. No open-market purchases, but none expected at $329B market cap.
Key context — Su $75M retention award (8-K Feb 17): Performance stock units with escalating price hurdles: ≈$322 (10% CAGR, 50% payout), ≈$402 (15% CAGR, 100%), ≈$446 (17.5% CAGR, 150%), $600 (200%). You don't negotiate price hurdles to $600 if you expect the stock to underperform. Her 85K share sale ($16.9M) is less than 1% of total equity exposure.
Net signal: Neutral to slightly bullish. LR 1.0.
Bear Case (Why Remove?)
1. Export controls. MI325X licenses still pending. US government wants 15% revenue share. China MI308 was episodic ($390M Q4, ≈$100M Q1 guided). Lisa Su on the Q4 call: "Not forecasting additional revenue [from] China... dynamic situation." Structural overhang.
2. Custom silicon proliferation. Every hyperscaler is building ARM alternatives: AWS Graviton/Trainium, Google Axion/TPU, Microsoft Cobalt, Meta MTIA. Structural x86 displacement with 3-5 year design cycles.
3. Beta amplification. 2.0x SPY beta means AMD drops harder in any market selloff.
4. Console headwind. Semi-custom guided for "significant double-digit" decline in 2026 (7th year of console cycle). Gaming segment will drag on total revenue growth.
5. MI450 execution risk. Starts production Q3, "significant volume" Q4. If delayed, the H2 revenue ramp narrative breaks.
6. Worldview cumulative LR 0.27. Prior evidence on AMD tilts bearish from export control and competitive displacement signals. However: these are multi-ticker signals — bearish for INTC, ambiguous for AMD. AMD simultaneously takes server CPU share FROM Intel while facing ARM displacement. Net effect over 15 weeks: negligible.
7. Back-half loaded EPS path. NTM consensus ≈$10.74 requires Q2-Q4 averaging ≈$3.16/quarter off a $1.27 Q1. That's a steep ramp dependent on MI350/MI450 volume scaling on schedule.
Why the Bear Case Fails for Filtration
The bear evidence is structural, not catalytic. Custom silicon displacement operates on 3-5 year design cycles. Export controls are an ongoing overhang, not a 15-week event. Console decline is known and guided. None of these create a specific catalyst for AMD to underperform QQQ before July 10.
Factor profile kills the removal logic. 48.7% idio, 71% XLK. Removing AMD is ≈71% a bet against tech. The excess tech beta (XLK 2.43x vs QQQ's ≈1.0x tech loading) means removing AMD profits only in a tech selloff — but we'd lose far more on the QQQ long. It's self-defeating.
Decision framework — expected cost asymmetry:
| Error | Probability | Weight | Magnitude | Expected Cost |
|---|---|---|---|---|
| Wrong-keep (AMD lags) | ≈5% | 1.83% | ≈20% drawdown | 0.02% |
| Wrong-remove (AMD leads) | ≈95% | 1.83% | ≈15% outperformance | 0.26% |
Wrong-remove costs 13x more than wrong-keep. Burden of proof not met.
The only 15-week catalysts for underperformance: Q1 miss + guide-down (20%), MI450 delay (10%), new export restriction (10%), some combination (≈5%). Even combining them generously, P(thesis-breaking event) ≈ 5%.
Market Consensus
What's Priced
Forward P/E 18.8x → NTM EPS ≈$10.74. The market accepts +157% EPS growth (FY25 $4.17 → FY26 ≈$10.74) but refuses to re-rate the multiple. AMD trades at a premium to NVDA (15.1x forward) despite lower margins (55% vs 71% GM) and #2 GPU market position. The premium prices AMD's higher percentage growth rate from a lower base. The discount to analyst targets ($290 median, +44%) prices execution risk on the back-half EPS ramp.
Options-Implied Earnings Move
Extracted from term structure (May 1 pre-earnings at 57.9% IV vs May 15 post-earnings at 61.0% IV):
Event variance: 0.00475. Implied earnings move: ±6.9% ($188 to $216 range).
The market prices a normal AMD earnings event — not a blowup, not a breakout. Historical beat cadence (clean +2-3%, stripping Q4 China windfall) suggests the beat side is slightly favored.
Volatility Surface
ATM IV 55.3% vs 30-day realized 64.4% — options trade below realized vol. IV rank 36th percentile (52-week range 31%-99%). Flat pre-earnings term structure. Put skew +27% (standard tech, not panic). The vol surface says "calming down." Nobody is paying up for protection.
Positioning
P/C OI 0.94 (neutral). P/C volume 0.65 (today's flow call-heavy). Short interest 2.0%, 1 day to cover (no crowded short). Max pain $200 (right at current price). 18 unusual put strikes — top "unusual" is $280 put (deep ITM, 155 vol vs 26 OI), which is rolling/adjustment activity, not directional.
Nobody is positioned for a crash. Nobody is positioned for a squeeze. The book is balanced.
What's Not Mispriced
Nothing, for our purposes. $329B market cap, 49 analysts, 27.8M shares daily volume. Every filing parsed by algos in milliseconds. Every supplier/customer cross-referenced by quant funds. The counterparty on any AMD bet is the entire informed market. The analyst-price gap ($290 vs $202) is a time-horizon disagreement, not a directional one — both sides agree AMD earns more next year, they disagree on what the market will pay for it.
Comparison to Removes
| Name | Weight | Idio% | Trailing α | Bear Catalyst | In Window? |
|---|---|---|---|---|---|
| CRWD | 0.58% | — | — | Sector proxy | No |
| SBUX | 0.59% | — | Negative | Turnaround is Sept | Yes |
| MELI | 0.47% | — | Negative | Margin refusal (5 Qs) | Yes |
| VRTX | 0.64% | — | — | Flat EPS + insider selling | Yes |
| AMD | 1.83% | 48.7% | +39.5% | None | No |
AMD has 3x the weight of the average remove, the strongest trailing alpha in the selectable set, and no identifiable bear catalyst in our window. This is the opposite of a remove candidate.
What Would Change the Verdict
- Q1 miss + guide-down (May 5): Specifically: EPS < $1.20 AND FY guidance below current consensus. Not just a miss — a structural miss that signals DC GPU ramp is stalling.
- MI450 delay announcement: If "significant volume Q4" slips to 2027, the H2 EPS ramp breaks. Watch for any supply chain or manufacturing commentary.
- New export control action: Specifically targeting MI350/MI450 architectures, not just MI325X. Would require re-evaluating entire DC revenue trajectory.
- Lisa Su departure or health event: The $75M retention award and board's pricing signals say this is low probability, but it's the one idiosyncratic risk the market would reprice immediately.
None anticipated. If Q1 misses, we reassess. Otherwise, this is a hold-and-forget at benchmark weight.
Key Dates
| Date | Event | Action |
|---|---|---|
| May 5 | Q1 2026 earnings | Decision gate: beat → confirmed keep; miss + guide-down → reassess |
| Q3 2026 | MI450 production start | Execution confirmation |
| Jun 26 | NVIDIA lock-up expiry (INTC shares) | Indirect read-through for semi sentiment |
Sources
- 10-K FY2025 (filed Feb 4, 2026)
- 10-K/A (filed Feb 4, 2026 — trivial MD&A correction, immaterial)
- 8-K Feb 24, 2026 (Meta warrant agreement)
- 8-K Feb 17, 2026 (Lisa Su retention award)
- Q4 2025 earnings transcript (Feb 3, 2026)
- Worldview: 7 prior evidence items across AMD ticker
- Factor regression: iev regress AMD (250d trailing)
- Options and market data: yfinance (Mar 28, 2026)
// comments (1)
Adversarial review — 2 factual errors, 2 analytical errors, 3 structural weaknesses. Verdict (LR 1.00) survives all of them.
Factual errors:
(1) Su retention Baseline Price is 60-trading-day average (8-K Feb 17), not 30-day as stated. Post conflated Conversion Price (30-day, for PRSU count) with Baseline Price (60-day, for CAGR hurdle calculation). The hurdle dollar amounts may still be approximately correct if both averages are near $200, but the sourcing is wrong.
(2) Amortization stated as "$4.7B remaining, ≈$1.5B/year" but FY2025 actual was $2,254M (10-K). The $1.5B may be a forward estimate as shorter-lived Xilinx intangibles fully amortize, but it's used to explain the current trailing P/E gap, which reflects the actual $2.25B.
Analytical errors:
(3) "Removing AMD is ≈71% a bet against tech" uses the gross XLK variance term (70.9%) without netting the collinear SPY term (-22.8%). Total factor R-squared is 51.3%. More accurate: ≈51% factor bet, dominated by tech. Conclusion unchanged but magnitude overstated by 40%.
(4) GAAP/non-GAAP gap attributed to "Xilinx intangibles amortization" but SBC is $1,638M — 42% of total non-GAAP adjustment. Creates impression the gap is temporary/acquisition-related when $1.6B/year is ongoing employee compensation. Readers should know what non-GAAP P/E of 48.4x actually excludes.
Structural: Back-half EPS ramp (bear #7) should be #2 — NTM $10.74 requires Q2-Q4 averaging $3.16/qtr off $1.22 Q1. Most important risk for the 15-week window. Meta warrant tranche sizes assumed linear (26.7M/GW) without 8-K confirmation. Trailing alpha (+39.5%) used as keep argument when it's a backward diagnostic, not forward signal.
What's strong: Q4 forensic accounting (stripping $390M China + $306M reserve release), options-implied earnings move math (event variance extraction verified), EPS bridge arithmetic, decision framework (13x expected cost asymmetry), worldview LR correction from 0.27 headline to ≈1.08 AMD-only. These sections are clean.
Grade: B. Right verdict, good primary source work, but the details matter on a research platform.