$10.3B Record Quarter, Data Center +39%, FY26 Growth "Could Exceed 60%" — and the Stock Falls 17% on the Whisper-vs-Guide Gap: Maintaining Outperform

Results vs. Consensus

Q4 2025 is the cleanest beat AMD has delivered in our coverage. Revenue beat consensus by 6.2%, EPS by 16%, gross margin held at the structural run-rate restored in Q3, free cash flow set a new record, and segment composition was broad-based across Data Center, Client, and Gaming. There is no operating soft spot in this print to anchor a downgrade against, and there are several specific positive surprises (the magnitude of the DC growth reacceleration, the MI308 China $100M return, the FY26 DC 60%+ framing) that extend the bull case. The narrative-driven post-print sell-off captures none of this; it is a positioning event, not a fundamentals event.

Quality of Beat

• Revenue: The $600M beat is broad-based. Data Center contributed the largest share via the +39% YoY ramp and the MI350 production scale, but Client and Gaming both held the elevated run-rates established in Q2 and Q3. The print shows AMD operating at a clean $40B+ revenue run-rate exiting 2025 — a level the bull case was modeling for late-2026 only six months ago.
• Margins: Non-GAAP gross margin of ~55% is a +100 bps step from Q3's 54% and meets the Q4 guide upper bound, indicating mix improved as Data Center grew sequentially and Client/Gaming margin held. The "dollars over rate" framing Lisa Su gave on the Q3 call is materializing as exactly that — revenue scale is driving operating-margin dollar contribution while rate continues modest expansion.
• EPS: $1.53 vs $1.32 consensus is a 16% beat, the cleanest EPS beat AMD has delivered post-IPO of MI300. The combination of revenue upside, margin expansion, and operating-expense discipline produced operating leverage that consensus had not modeled. Non-GAAP EPS up 40% YoY against revenue up 34% YoY confirms positive operating leverage at scale.
• FCF: $2.0B Q4 FCF is a record and brings FY25 FCF to $5.4B, well above the $4–4.5B range we modeled exiting Q3. AMD enters 2026 with cash + short-term investments of approximately $7.7B against $3.2B of debt — net cash of $4.5B+ and meaningful capacity to fund Helios capex without issuing equity (an important positive given the OpenAI warrant overhang that already exists).

Segment Performance

Note: Client / Gaming / Embedded specific Q4 figures estimated from total revenue minus disclosed Data Center; reconciliations may shift modestly when 10-K is filed.

Data Center: The Reacceleration Compounds

Data Center revenue of $5.4B (+39% YoY, +24% QoQ) is the standout of the print. The trajectory through 2025 — Q1 +57% (with MI308 China still in the comp), Q2 +14% (China-ban trough), Q3 +22% (reacceleration), Q4 +39% (full ramp visibility) — tells the underlying operating story unmistakably: the ex-China growth franchise is materially stronger than the headline figures alone suggest. Management's Q4 framing that FY26 DC growth "could exceed 60%" is the explicit version of what the operating data was already telling us.

"Data center revenue more than doubled in 2025... we expect 2026 to be another year of significant growth, with data center as the lead engine." — Lisa Su, CEO

Drivers of the +39% Q4: (1) MI350 series in full production scale across multiple cloud and AI customer deployments (the multi-customer framing that emerged in Q3 is now operational); (2) EPYC server CPU continuing the multi-quarter share gain march, including continued momentum in cloud-instance proliferation; (3) MI308 China contribution returning at modest scale ($100M in Q1 2026 guide; some Q4 contribution may exist though management framed the China revenue as primarily Q1-loaded); (4) ROCm 7 production deployments at top-tier customers, validating the software ecosystem step that shipped in Q3.

Assessment: Data Center is now in the volume scale-up phase. The bridge thesis (Q1–Q3 of "wait for MI355 ramp evidence") is closed; the new question is the FY26 trajectory and the Helios/MI450 inflection in 2H 2026 with OpenAI as the anchor. Both look on schedule. We model FY26 Data Center revenue moving toward $32–36B at the high end of management's "could exceed 60%" framing — meaningful step-up from the $25–28B range we set at our Q3 update.

Client & Gaming: Floor Confirmed Higher

Client + Gaming combined remains the operating floor under the AMD investment case. Q4 specific figures await the 10-K but our reconciliation from the $10.27B total minus $5.4B Data Center minus the modest Embedded contribution implies Client+Gaming printed roughly $3.8–4.0B, sustaining the elevated Q3 run-rate and growing meaningfully YoY on a comp that is now starting to get harder. With four consecutive quarters of strong execution in this complex (Q1 $2.9B → Q2 $3.6B → Q3 $4.0B → Q4 ~$3.8–4.0B), the segment is now a $15B+ run-rate franchise growing at robust YoY rates — the kind of non-AI revenue floor that materially de-risks the bull case if the Instinct ramp ever wobbles.

Assessment: Client+Gaming is doing what we hoped — growing without dramatic surprises and providing operating-margin contribution at a meaningful share of total OI. The risk into 2026 is comp normalization (the Q1 2026 +32% YoY guide implies some sequential decel) but the absolute dollar level holds.

Embedded: Recovery Reset to 2H 2026

Embedded continues its multi-quarter laggard pattern. Management's framing on the Q4 call shifted from "expect 2H 2025 inflection" (the Q1–Q3 framing) to "expect to return to growth in 2H 2026" — effectively a one-year recovery slip. This is not catastrophic given Embedded is ~10% of revenue and ~15% of OI, but it's a real reset on management credibility for this segment specifically. We've now had four consecutive quarters of Embedded missing the recovery cadence; the Q4 reset to 2H 2026 acknowledges what the data has been showing.

Assessment: Embedded is the lone clear demerit on this print. The new framing is more honest than the prior quarters' "imminent recovery" stance, but the credibility cost was real. The segment's operating margin profile (~30%+) means even a sustained -5% YoY level is manageable in aggregate. Watch for Q1 2026 segment-specific commentary; if the trajectory worsens further, this becomes a real issue.

FY 2026 Framework: "Data Center Growth Could Exceed 60%"

Management did not provide a formal full-year FY 2026 revenue figure on the Q4 call. The framework was qualitative but pointed: Data Center growth "could exceed 60%" in 2026, anchored by (a) MI350 sustained through 1H, (b) MI355 ramp through 1H, (c) MI450 + Helios platform ramps in 2H 2026 alongside the OpenAI first-tranche 1GW deployment, (d) sustained EPYC server CPU share gains, and (e) modest MI308 China resumption (Q1 ~$100M baseline; full-year contribution unclear).

Q1 2026 specific guide: Revenue ~$9.8B ±$300M (above $9.38B consensus by $420M); non-GAAP gross margin ~55%; includes ~$100M MI308 China. Sequential decline of -5% QoQ from the Q4 $10.27B base reflects normal Q1 seasonality across Client/Gaming and the absence of the Q4 holiday consumer pull. YoY growth of +32% confirms the underlying ramp continues.

The Q1 guide is the source of the post-print sell-off, in our reading. AMD-momentum desks heading into the print were modeling Q1 closer to $10B+ on the assumption that the AI demand strength would over-ride seasonality. The $9.8B midpoint is structurally fine (above LSEG by $420M) but signals normal seasonality is in play, which the most-bullish positioning had implicitly bet against.

The OpenAI Deal: On Schedule Through Q4

Lisa Su confirmed on the Q4 call that the first 1GW MI450 deployment for OpenAI is "tracking to plan for second half 2026" with no material change to the warrant tranche structure or deployment scale described in the October 6, 2025 announcement. This is exactly the operational status we wanted to see at Q4 — on track, no slippage, and integrated into the company's FY26 framework rather than a separate sidecar narrative.

"The OpenAI partnership is tracking to plan with the first gigawatt of MI450 deployment scheduled for second half 2026, and we continue to see expanding engagement with multiple hyperscalers around the broader Instinct portfolio." — Lisa Su, CEO

The "expanding engagement with multiple hyperscalers" framing is critical: it pushes back against the customer-concentration concern we flagged at Q3 by signaling that the OpenAI commitment is part of a broader merchant-share traction rather than a standalone bet. As MI450 ramps approach in 2H 2026, sell-side desks will increasingly model multi-customer Instinct revenue contribution at hyperscaler scale — the pattern that historically supported Nvidia's multiple.

Key Topics & Management Commentary

Overall Management Tone: The most confident and forward-looking framing of the cycle. Su repeatedly used the framing "year 2 of a 10-year AI build-out" — an unusually long-cycle posture for an AMD CEO and one that explicitly sets up the multi-year revenue trajectory the OpenAI deal supports. Less defensive on China, no executional caveats, clear roadmap commitments. The tone shift from Q1 (defensive) → Q3 (constructive) → Q4 (assertively bullish) maps the operational story arc precisely.

MI308 China: Limited Resumption

The most concrete new operational data point: AMD has obtained sufficient export licensing to ship approximately $100M of MI308 to China in Q1 2026. Management did not extrapolate beyond Q1 because of the "very dynamic situation" with US-China policy posture. This is meaningfully less than the Q1 2025 contribution (likely $400–600M run-rate) but is a return from zero and an option-creating data point.

Assessment: The MI308 China story is no longer pure overhang — it is a constrained-but-positive contribution starting Q1 2026. We don't model material upside from China beyond the Q1 $100M because management explicitly declined to project further, but the trajectory has improved from "indefinite zero" to "limited but real."

ROCm 7 Production Adoption

Su disclosed that ROCm 7 is now in "production deployments at multiple top-tier customers" — the operational follow-through to the Q3 launch. This is the specific evidence we wanted on the ROCm narrative: a software release at scale being used by hyperscaler-grade customers in production AI workloads. Combined with the "7 of top 10 model builders use Instinct" framing maintained through 2025, the ROCm-vs-CUDA narrative continues to shift in AMD's favor at the production-customer level.

Helios Open Standard: Multi-OEM Participation

The Helios rack-scale platform that AMD positioned around Meta's open rack standard in Q3 has expanded to "multiple OEMs participating" per management's Q4 framing. We don't have specific OEM names but the multi-vendor ecosystem positioning is the operational signal we wanted — Helios is becoming a standards-based merchant platform rather than a single-customer reference design.

Capital Allocation

FY25 FCF of $5.4B and net cash of $4.5B+ exiting 2025 give AMD ample capacity to fund Helios capex, MI450 production ramp, and OpenAI-deployment-related infrastructure investment without equity issuance. Management did not announce material new buybacks or capital-return programs — appropriate given the OpenAI warrant overhang and the strategic capex requirements ahead. The fact that AMD can self-fund a Helios + MI450 ramp at this scale without dilution is a structural positive vs the bear-case framing of "AMD will need capital to compete with Nvidia at hyperscale."

Analyst Q&A Highlights

FY 2026 Framework Specifics

• Multiple analysts pressed Su on FY26 revenue specifics, OpEx growth, and margin expansion trajectory. Su held to qualitative ">60% Data Center growth possible" framing without committing to a formal full-year revenue figure. The reluctance to provide a formal FY26 number is consistent with her Q1–Q3 posture and likely reflects (a) genuine uncertainty about MI308 China license trajectory through the year, and (b) preference to under-promise and over-deliver after the messy MI308-charge optics of 2025.
  Assessment: The qualitative framing is sufficient. The implied FY26 trajectory is materially above where Street had been heading into the print, and the lack of formal commitment is conservative posture rather than weakness.

MI450 Customer Diversity Beyond OpenAI

• Pressed on whether the OpenAI commitment crowds out other hyperscaler MI450 demand, Su confirmed "active engagements with multiple hyperscalers" beyond OpenAI for MI450/Helios deployments in 2H 2026. No specific customer disclosures.
  Assessment: This is the right answer to the customer-concentration question. We treat it as directionally credible given AMD's overall hyperscaler engagement breadth, but specific customer reveals will be needed in 1H 2026 for the bear case to fully retreat.

Helios Open-Standard Ecosystem

• Asked about Helios partner breadth, management noted "broader than just Meta; multiple OEMs participating." No specific OEM names disclosed on the call.
  Assessment: Multi-OEM framing is structurally better than a single-customer reference. The opacity on specific OEMs is a minor demerit but consistent with AMD's typical disclosure cadence.

Embedded Recovery Timing Reset

• Acknowledged that Embedded recovery has slipped to "2H 2026 return to growth" — one year later than Q1 2025 framing.
  Assessment: A reset, but a clearer one. Management is now explicitly resetting expectations rather than deflecting. This is actually a small positive for the credibility ledger going forward.

What They're NOT Saying

1. Formal FY 2026 revenue and EPS figures: Conspicuously absent for a Q4 call. Management's "could exceed 60% Data Center growth" framing is bullish but not committal. The Street will fill the gap with a wide range of bottom-up models, which creates ambiguity around buy-side positioning into 1H 2026 prints.
2. OpenAI warrant tranche-by-tranche revenue cadence: Still no public schedule of when each warrant tranche vests vs deployment milestones. The "tens of billions over multi-year" framing remains. This opacity is acceptable today but will need to firm up by mid-2026 as the first 1GW deployment scales.
3. FY 2026 Data Center mix between EPYC, MI350/MI355, and MI450: Management did not break out expected segment composition. With three accelerator generations active in 2026 plus the EPYC franchise, the mix matters for both revenue forecasting and margin modeling.
4. MI308 China full-year contribution: $100M Q1 only, with no FY framework. Could be $400–1000M FY26 contribution depending on license cadence; could be much less. Management's "very dynamic situation" framing is honest but model-affecting.
5. OpEx growth assumptions for FY26: No formal OpEx guide. With Helios + MI450 ramps requiring meaningful R&D and possible additional engineering hires, OpEx growth could be a 20%+ figure that compresses operating leverage if revenue surprises to the downside.
6. Capital allocation specifics: No buyback authorization update, no dividend conversation, no specific capex framework for FY26. The implied posture is "self-fund growth" but the framework would help model FCF projections.

Market Reaction

• After-hours move (Feb 3 evening): Stock fell sharply post-print despite the headline beats. The AH move was driven by the Q1 2026 guide being above LSEG consensus but below the most-bullish AI-momentum desk whisper expectations.
• Feb 4 intraday: -17%. The largest single-day post-print decline AMD has experienced in our coverage cycle. Volume materially elevated. The AI complex was broadly weak on the day, with some additional pressure from continued Burry-style AI-bubble narrative concerns active in markets through early 2026.
• Pre-print context: AMD entered Feb 3 having recovered from the Q3 9% post-print drawdown, trading roughly in the mid-$240s to high-$240s area. Forward P/E heading into the print was approximately mid-30s. The recovery from the Q3 drawdown plus the elevated multiple created the setup for a "guide-vs-whisper" reset on any guide that did not exceed bullish positioning.
• Cited reasons for the drop (sell-side aggregation): (1) Q1 guide above LSEG but below most-bullish whisper at $10B+; (2) Sequential -5% QoQ revenue decline implied by guide raised "back-half loading" concerns despite management's seasonal explanation; (3) MI308 China contribution at $100M Q1 was conservative vs some buyside hopes for $300–500M contribution; (4) Continued AI-bubble narrative pressure (Burry short repositioning, broader AI-complex rotation); (5) Forward valuation reset to a more sustainable mid-20s P/E from mid-30s entering the print.

The 17% drop is a positioning event, not a fundamentals event. Q4 actuals beat consensus by every measurable metric, FY26 framework is more bullish than at any prior point in the cycle, OpenAI is on schedule, and the MI308 China story has improved from "indefinite zero" to "limited but real." None of those are consistent with a 17% downward revision in fundamentals; what changed was the buyside positioning around how aggressive an AMD beat-and-raise was supposed to be. This is a textbook capitulation flush in a quality name — the kind of single-day move that historically marks the trough or near-trough of a multi-quarter consolidation in a stock whose underlying business is materially stronger than the price action suggests.

Street Perspective

Debate: Does the 17% drop change the multi-year framework?

Bull view: No. Operating fundamentals improved on every measurable axis. The drop is positioning unwinding by AI-momentum desks who had over-sized AMD on whisper-driven Q1 expectations. With shares now in the low-$200s area against an unchanged-to-improved FY26 trajectory and the OpenAI 2H 2026 catalyst still ahead, the entry point is meaningfully better than at any point since the Q2 2025 drawdown low.

Bear view: A 17% post-print drop in a beat-and-raise quarter signals the market is starting to price in (a) a peak AI capex cycle in 2026 rather than a sustained 10-year build, (b) hyperscaler ASIC ramp eroding merchant GPU TAM faster than AMD's roadmap can offset, and (c) the OpenAI deal warrant dilution being more material than the back-end deployment economics. The drop is informational, not noise, and the multiple compression deserves to stick.

Our take: Bull is closer to right. The bear's points are valid medium-term considerations but none of them have specific Q4 evidence supporting them — Data Center +39%, ROCm 7 production deployment, Helios multi-OEM ecosystem, and OpenAI on-schedule are all directly contrary to the bear case. The drop is a positioning unwinding event around an unchanged-to-improved fundamental thesis. We treat the post-print level as a buying opportunity within our Outperform framework.

Debate: Is the FY26 Data Center "could exceed 60%" framing aggressive or conservative?

Bull view: Conservative. Q4 just printed +39% YoY with MI350 still ramping; MI355 layers in 1H 2026; MI450 scales with OpenAI in 2H 2026; ROCm 7 in production at top-tier customers; sovereign AI engagements expanding. The 60%+ framing is the floor of a realistic FY26 trajectory; the upside scenario is closer to +75–90% Data Center growth if MI450 ramps faster than the Q4 framing implies.

Bear view: Aggressive. Q1 sequential -5% suggests the front-half is compositionally weak, which means FY26 60%+ DC growth requires a violent 2H ramp that concentrates execution risk into the MI450 launch window. If MI450 slips even by a quarter, the +60% FY26 framing breaks decisively.

Our take: The bull is right on substance; the bear is right that 2H concentration of growth is a risk to monitor. We model FY26 DC growth at +50–60% as our central case with upside skew to the upper end of the range. The execution-risk concentration in MI450/2H is real but well-flagged; if any slippage emerges, it would show up in the Q1 or Q2 2026 prints with enough lead time to reset.

Model Update Needed

Valuation impact: Our Q3-update fair-value framework was $200–240 based on a ~28–32x forward P/E on FY26 EPS of ~$7.50. Updating for the higher FY26 EPS midpoint of ~$9.00 and holding the multiple range constant, fair value moves to $250–290. We're conservative on the multiple given the post-print sentiment reset and apply a modest discount for execution risk concentration in MI450/2H 2026, putting our fair-value range at $235–280. At the post-print intraday low in the low-$200s, the stock trades meaningfully below our framework fair value — the most attractive risk/reward setup in our AMD coverage to date.

Thesis Scorecard Post-Earnings

Overall: Nine bull points confirmed (eight existing strongly + one new), three bear points either narrowing or contained, one bear point worsening (Embedded), and one new bear point identified (positioning sensitivity / whisper-driven volatility). The operating thesis is unequivocally stronger than at the Q3 upgrade, while the entry point has materially improved on the post-print drawdown.

Action: Maintaining Outperform. The Q4 print is the cleanest beat of the cycle, FY26 framework is bullish, OpenAI is on schedule, MI308 China is starting to resume, ROCm 7 is in production at top-tier customers, and the post-print 17% drawdown is a textbook positioning unwind in a quality name. Our fair-value framework moves to $235–280 and the post-print level in the low-$200s is the most attractive risk/reward we've seen in AMD coverage. We don't upgrade beyond Outperform (3-tier framework) and we don't downgrade because nothing about the operating story warrants it. We sit at Outperform with elevated conviction.