Bridge Quarter Delivered: Data Center Reaccelerates to +22%, GM Snaps Back to 54%, OpenAI 6GW Deal Adds a Decade-Long Catalyst — and the Stock Sells Off Anyway: Upgrading to Outperform

Results vs. Consensus

Q3 2025 is the cleanest top-to-bottom print AMD has delivered in the recent cycle. Revenue, EPS, gross margin, free cash flow, segment composition, and forward guide all came in above or in line with bullish-end-of-range Street modeling. The contrast with Q2 is sharp: where Q2 delivered the optically-ugly bridge quarter we modeled, Q3 delivered the operational snap-back the Q1/Q2 thesis required. There is no major disappointment in this print to anchor a continued Hold rating against, and there is one significant new positive (the OpenAI deal) that changes the multi-year revenue framing.

Quality of Beat

• Revenue: The $510M beat versus consensus is fully organic and broad-based. Data Center reaccelerated to +22% YoY without MI308 China contribution; Client compounded its first-half strength at +46% YoY; Gaming continued the dramatic recovery at +181% YoY. No single segment is responsible for the magnitude of the beat — this is the kind of broad operating beat that historically supports multiple expansion (or, when valuation is full, reduces drawdown risk). The +20% sequential growth from $7.685B to $9.246B is the single largest QoQ revenue increase in dollar terms in the recent AMD cycle.
• Margins: Non-GAAP gross margin snapped back to 54% from Q2's 43%, a clean +1,100 bps recovery in a single quarter and exactly the trough-is-one-quarter outcome the Q2 framing implied. The gross-margin recovery is the validation that the $800M MI308 charge was fully one-time and contained. The Q4 guide of 54.5% confirms the structural run-rate is now restored and tracking modestly higher into year-end.
• Operating Leverage: Non-GAAP operating margin expanded to 24.2% — a meaningful step-up from Q2's ~12% (charge-impacted) and visible operating leverage relative to the $7.685B / 24% OM run-rate of Q1 (which was effectively the structural baseline). On the Data Center segment specifically, operating income recovered to $1.074B from the $(155)M Q2 loss — a ~$1.23B segment OI swing in a single quarter as the MI308 charge rolled off and MI350 production economics kicked in.
• Free Cash Flow: $1.5B Q3 FCF is a record, and 9-month FCF of $3.437B at a ~14% margin is materially above where we modeled coming out of Q2. Cash + short-term investments of $7.243B against $3.220B of total debt leaves the company with net cash of ~$4B and ample dry powder for either capital returns or M&A. The OpenAI warrant structure also implicitly preserves cash (unlike a pure equity sale).
• EPS: Non-GAAP $1.20 vs $1.16 consensus is a ~3% beat — in line with the operating beat, with no unusual tax or below-the-line drivers. GAAP EPS of $0.75 (up 60% YoY) reflects the same operating recovery without the Q2 GAAP-vs-non-GAAP inversion that flattered the comparable Q2 print. The convergence is a sign of clean, recurring quality of earnings.

Segment Performance

Data Center: The Reacceleration the Thesis Needed

Data Center revenue of $4.341B (+22% YoY, +34% QoQ from Q2's $3.24B) is the single most important data point in the print. Three things matter about this number: (1) it materially reaccelerated from Q2's +14% trough despite still containing zero MI308 China revenue, validating that the ex-China trajectory is intact; (2) it confirms the MI350 series volume production that started ahead of schedule in June is contributing materially to revenue in its second full quarter; and (3) the operating income swing from Q2's $(155)M loss to Q3's $1.074B profit ($1.23B in segment OI in a single quarter) demonstrates that the charge-induced trough was structurally one-time, exactly as the Q1 framing assumed.

"Multiple MI350 series deployments are underway with large cloud and AI providers, with additional large-scale rollouts on track to ramp over the coming quarters." — Lisa Su, CEO — the operationally most important sentence on the call

The +22% YoY headline number understates the underlying strength. With MI308 China contribution at zero (versus the contribution that was in the year-ago Q3 2024 base), the comp is effectively China-disadvantaged. The implied ex-MI308-comp growth in Data Center is in the high-twenties to low-thirties percent range — meaningfully closer to the +57% Q1 print than the +14% Q2 trough suggests. EPYC continued its multi-quarter share-gain march, and the MI350 ramp is contributing at multiple cloud and large-scale customers (the multi-customer framing matters: it argues against the "one or two big hyperscalers" concentration risk that some bears had flagged).

Assessment: Data Center is back on the trajectory the bull case needed. The +22% headline plus the implicit ex-China underlying suggests a clean sequential ramp through Q4 as MI350 deployments continue, with MI355 ramping into 1H 2026. Combined with the OpenAI deal as a 2H 2026 anchor for MI450, Data Center revenue trajectory through 2026 is materially better-supported than at our Q2 update.

Client: Quietly Compounding

Client at $2.75B (+46% YoY) is the third consecutive quarter of mid-40s-or-higher growth on top of an already-elevated comp. The Q1 ($2.3B, +68%) → Q2 ($2.5B, +67%) → Q3 ($2.75B, +46%) sequence shows continued sequential dollar-growth (~$200M+ each quarter) with growth-rate normalization that is exactly what we'd want to see — comp-driven decel without dollar-flow weakness. Ryzen 9000 desktop strength and continued ASP expansion remain the drivers. The "tariff pull-forward" thesis we noted as buried after Q2 stays buried — three consecutive quarters of share-gain-led growth is now structural.

Assessment: Client is now annualizing $11B+ at high-30s%-plus YoY growth. Even if the comp gets harder in Q4 and 1H 2026, the dollar-flow trajectory implies $13–14B in FY26 Client revenue at scale, with mid-teens segment operating margin contribution. This is the floor under the AMD investment case if AI accelerator economics ever wobble.

Gaming: The Big Surprise Continues

Gaming printed $1.298B (+181% YoY), continuing the dramatic Q2 inflection (+73%) at an even more extreme rate. The driver mix is meaningful: semi-custom (console SoC) appears to be running materially ahead of the cycle pattern that prevailed in 2024, and discrete Radeon (RX 9000 series) is taking share at retail. Combined Client+Gaming operating income of $867M is the single largest non-Data-Center contribution AMD has reported, and it's coming with reasonable margin expansion rather than mix-induced compression.

Assessment: Gaming's +181% is partially comp-flattered (Q3 2024 was the trough of the previous semi-custom cycle), but the underlying inflection is real. We continue to view this segment as more cyclically stable than its name suggests — the semi-custom contracts have multi-year life, and Radeon discrete is now meaningfully more competitive. We don't expect +181% to repeat in Q4 (or Q1 2026 once comps normalize), but the dollar-revenue level should hold or grow modestly.

Embedded: Now A Definitive Drag

Embedded at $857M (-8% YoY) is the third consecutive quarter of failing to inflect to growth, and the trajectory is mildly worsening (-3% Q1 → -4% Q2 → -8% Q3). Operating income of $283M at ~33% segment OM holds up reasonably, but the recovery framing that management has held to since Q1 needs a credibility reset. With Embedded at ~9% of revenue and ~13% of segment OI, the drag is manageable in aggregate but is now a clearly-identified pressure point.

Assessment: Embedded is now the segment most at risk of a downward revision in management's near-term framing. The Q4 guide does not break out Embedded explicitly, so we'll know more in February whether management resets the recovery cadence. This is a watch item but not yet thesis-altering — Embedded would need to fall to mid-single-digit percent of total OI to become a real concern.

The OpenAI Partnership: Decade-Defining Customer Commitment

On October 6, 2025 — one month before the Q3 print — AMD announced a strategic partnership with OpenAI to deploy 6 gigawatts of AMD Instinct GPUs over multiple years, structured around an initial 1GW deployment of MI450 series accelerators beginning in 2H 2026. The deal is paired with a warrant issued to OpenAI for up to 160 million shares of AMD common stock, vesting in tranches tied to (a) deployment scale milestones, (b) AMD share-price targets, and (c) OpenAI technical/commercial milestones. If fully exercised, OpenAI would own approximately 10% of AMD.

"This partnership brings the best of AMD and OpenAI together... We are thrilled to partner with OpenAI to deliver AI compute at massive scale." — Dr. Lisa Su, CEO

"We got a structure that has extremely aligned incentives. Everybody wins, right?" — Lisa Su, responding to Ross Seymore (Deutsche Bank) on warrant differentiation

The financial framing AMD provided is "tens of billions of dollars in revenue" with the deal "anticipated to be highly accretive to AMD's non-GAAP earnings-per-share." Some sell-side framing has put potential cumulative hardware revenue at ~$90B over the life of the deal — we treat that as a high-end estimate but the order of magnitude is consistent with management's guidance.

Three structural implications:

• Validation of the AMD roadmap. OpenAI's commercial mandate is to deploy compute as fast and cheaply as possible. They would not commit 6GW of capacity to MI450/MI355 unless internal benchmarks confirmed AMD can deliver competitive performance per dollar at hyperscale. This is the most credible third-party validation of the AMD AI accelerator thesis to date — arguably more meaningful than the "7 of top 10 model builders use Instinct" framing AMD has been giving for two quarters.
• Aligned-incentive structure rather than dilution. The warrant gives OpenAI ~10% ownership only at the back-end, contingent on AMD delivering and on shares appreciating. This is a structurally different cost than issuing stock today; it is closer in spirit to a customer success-fee than to dilutive financing. The accretion-vs-dilution math should be net-positive even at full warrant exercise, given the revenue scale involved.
• Multi-year revenue visibility that the Nvidia narrative has but AMD historically did not. Nvidia trades at a premium in part because its hyperscaler concentration produces multi-year revenue visibility. AMD has historically lacked this. The OpenAI deal gives AMD — for the first time at this scale — a multi-year anchor that supports forward-period revenue modeling with much greater confidence. That alone is worth a multiple step-up.

What it does NOT do: It does not change the FY25 print materially (1GW MI450 doesn't begin shipping until 2H 2026), and it does not eliminate the customer-concentration question (it arguably increases it for the back-end of the deal). But the FY26–FY28 revenue trajectory implied by the deal is materially higher than where consensus was pre-October-6, and more importantly, it is much more visibility-supported than the speculative MI400 ramp we were modeling at our Q2 update.

Key Topics & Management Commentary

Overall Management Tone: Notably more confident than even Q2 — the OpenAI deal in the rearview, the bridge quarter delivered, and the Q4 guide above Street give Su the cleanest framing she's had in a year. Opening line: "The demand for compute has never been greater" and "Our data center AI business is entering its next phase of growth." This is no longer the defensive China-overhang framing of Q1; this is forward-looking ramp narrative with concrete operational and customer evidence. The bar this sets for Q4 is high — the Q4 guide is explicit confidence rather than measured caution.

MI350 → MI355 → MI450 Roadmap Cadence

Management was unusually concrete on the product roadmap progression: MI350 series in volume production now (record DC quarter is the evidence); MI355 ramps in 1H 2026; MI450 series begins shipping at-scale in 2H 2026 (with OpenAI's first GW as the anchor deployment). Each transition is positioned as additive rather than substitutive — AMD is now operating a multi-product Instinct portfolio with overlapping product lifecycles, which reduces the "single-SKU bet" risk that characterized the MI300 era.

"MI355 continue to ramp in the first half of 2026. And then... MI450 series comes online in the second half of 2026." — Lisa Su, responding to Vivek Arya (BofA)

Assessment: The roadmap is the most de-risked it has been at any point in the AMD AI accelerator cycle. Three concrete products at three different lifecycle stages with operational evidence on the leading edge (MI350 in production) is the kind of cadence-of-execution that historically supported Nvidia's multiple, and AMD is now demonstrating it. We treat the roadmap risk as materially reduced from our Q2 view.

ROCm 7: The Software Story Catches Up

AMD launched ROCm 7 in the quarter, claiming "up to 4.6x higher inference and 3x higher training performance compared to ROCm6." Buried in the call but operationally critical: the historical CUDA-vs-ROCm software gap has been the single most-cited bear point on AMD's AI accelerator thesis. ROCm 7 is the largest single performance step AMD has shipped on the software side, and the workload-specific claims (4.6x inference) translate directly to better total cost of ownership for customers running production AI workloads on Instinct.

"We launched ROCm7, our most advanced and feature-rich release to date, delivering up to 4.6x higher inference and 3x higher training performance compared to ROCm6." — Lisa Su, CEO

Assessment: ROCm has historically been the soft underbelly of the AMD AI thesis. ROCm 7 is not "we've caught up to CUDA" — CUDA's installed base and tooling depth still leads materially — but it is the strongest closing of the gap AMD has shipped, and combined with hyperscaler-grade customer adoption (OpenAI for the largest workloads), it removes the "ROCm is unusable in production" framing that some sell-side desks were still carrying.

Helios: Rack-Scale Going Commercial

The MI400-era story extends with Helios, AMD's rack-scale AI platform "supporting Meta's new open rack wide standard." Helios is the systems-integration capability AMD acquired with ZT Systems materializing as a customer-facing product. With Meta as a named open-standard partner and OpenAI as the anchor customer for deployment scale, AMD now has the rack-scale ecosystem positioning that Nvidia has via NVL72 / NVL576 platforms.

Assessment: Helios is FY26+ revenue, not FY25. But it is the architectural answer to the "AMD ships chips, Nvidia ships systems" critique that has been a structural bear point for two years. The Meta open-standard partnership additionally signals that hyperscalers want a non-Nvidia rack-scale alternative, which is the customer pull AMD needs to ramp Helios at scale.

Sovereign AI: Continuing to Build Quietly

The 40+ active sovereign AI engagements globally that we flagged as a positive new thesis vector in Q2 continue to advance. Su didn't quantify a new engagement count on this call, but did reiterate sovereign as an additive customer category to hyperscalers and to OpenAI. The multi-region nature of sovereign AI deployments structurally supports geographic revenue diversification — a hedge against any single-customer or single-region concentration risk.

Assessment: Sovereign AI was the positive surprise of Q2; in Q3 it's a continuation rather than a step-change. The contribution to FY26 revenue is increasingly real but unquantified publicly. We treat it as a free option on the model.

Margin Framework Forward

C.J. Muse (Cantor) pressed Su on the gross margin framework heading into 2026, given the Q4 guide of 54.5% and the OpenAI deal economics. Su's response was telling: "Our priority in data center GPU business is to really expand the top-line revenue growth and the gross margin dollars." The framing of "dollars over rate" is consistent with rapid scale-up — AMD will accept some margin-rate compression to win revenue scale at the OpenAI level. This is the right strategic call but does set up a possible "GM rate disappointment but OI dollar beat" pattern in 2026 prints.

Assessment: Investors should be modeling FY26 GM in the 53–55% range with possible quarterly volatility tied to MI350 vs MI450 mix. The OI-dollar trajectory is what matters; we'd characterize the "dollars over rate" framing as a small risk for sentiment but the right operating call.

Q4 2025 Guidance and the FY25 Wrap

The Q4 guide of ~$9.6B ±$300M is constructive in absolute dollar terms (above $9.15B Street by ~$450M) but is the source of the post-print sentiment unease. Two specific concerns from the sell-side response:

• Sequential growth deceleration to ~+4% QoQ after Q3's +20% QoQ. This is a natural product-cycle pause — MI350 shipments are well-distributed across customers and a sequential breather is normal — but bears will frame it as "the bull case requires +10%+ sequential growth and the guide doesn't deliver."
• GM guide of 54.5% met but did not beat the StreetAccount 54.5% consensus. JPMorgan's note that "lack of stronger operating leverage in the model appears to be the most prominent concern" captures the bull-leaning desk view that GM should be expanding faster as Data Center mix improves. Our take: 54.5% in Q4 with the OpenAI ramp ahead is structurally fine; the operating leverage materializes as MI450 hits scale in 2H 2026.

FY25 implied: With 9-month revenue of $24.13B (Q1 $7.44B + Q2 $7.69B + Q3 $9.25B) and a Q4 midpoint of $9.6B, full-year FY25 lands near $33.7B — tracking +30% YoY to the FY24 base of ~$25.8B. This puts FY25 firmly in the upper half of the $32–33.5B range we modeled coming out of Q2 and confirms the "strong double-digit FY25 growth" framing management committed to in Q1. The MI308 China hole is fully absorbed and the print still tracks +30% — a meaningful demonstration of the underlying operating quality.

Street will land: Pre-print FY25 consensus was around $32.5–33B revenue / $4.50–4.75 EPS. The Q3 beat plus Q4 guide above Street should push FY25 estimates toward $33.5–34.0B revenue and $5.10–5.40 EPS. FY26 estimates — which are the more important re-rate variable — should move higher on the OpenAI deal recognition and the ROCm 7 / MI355 ramp, with consensus moving toward $40–43B revenue and $7.00–8.00 EPS over the next several weeks of Street model updates.

Analyst Q&A Highlights

OpenAI Deal Structure and Differentiation

• Ross Seymore, Deutsche Bank: Asked Su to differentiate the OpenAI partnership from a typical merchant-supplier relationship and to characterize the warrant structure. Su responded that the structure has "extremely aligned incentives. Everybody wins."
  Assessment: The "aligned incentives" framing is doing a lot of work here. The warrant tranching means OpenAI's economic upside scales with both AMD's deployment success AND share-price appreciation, creating a structural reason for OpenAI to be a vocal AMD reference customer rather than purely a customer of last resort. This is the kind of structural alignment that historically only Nvidia has had with its largest hyperscalers.

MI350 Customer Adoption and MI450 Transition

• Vivek Arya, Bank of America: Pressed Su on the CPU/GPU mix shift and the MI355-to-MI450 transition strategy. Su confirmed MI355 ramps in 1H 2026 and MI450 in 2H 2026.
  Assessment: The roadmap cadence is now the most concrete it has been at any point in this AI cycle. MI355 in 1H 2026 + MI450 in 2H 2026 is overlapping product lifecycle execution, not bet-the-quarter single-product launches.
• Thomas O'Malley, Barclays: Asked about discrete sales versus system sales for Helios. Su clarified that early MI450 customers will primarily be rack-scale solutions buyers, with discrete form factors also available.
  Assessment: Rack-scale-first deployment positioning is the right go-to-market strategy for displacing Nvidia at hyperscale — hyperscalers want full-stack platforms, not chips. AMD has finally architecturally aligned to that customer preference.

CPU Sustainability and ROCm Competitive Position

• Joshua Buchalter, TD Cowen: Probed the durability of the EPYC CPU share-gain story and ROCm's competitive positioning. Su pointed to ROCm 7's 4.6x inference / 3x training improvements vs ROCm 6.
  Assessment: The ROCm performance step is the largest single jump AMD has shipped on the software side. It does not close the CUDA gap on tooling depth, but it does close it materially on production performance for the workloads that matter most to scaled AI deployments. Combined with OpenAI as a marquee production user, the ROCm narrative is materially better than three months ago.

Margin Framework for 2026

• C.J. Muse, Cantor Fitzgerald: Asked about gross margin framework for 2026 given the Q4 guide and OpenAI deal economics. Su: "Our priority in data center GPU business is to really expand the top-line revenue growth and the gross margin dollars."
  Assessment: "Dollars over rate" is the right strategic posture during a hyper-scale ramp but does set up a possible quarterly GM-rate disappointment dynamic in 2026. Investors should be calibrated to model FY26 GM in the 53–55% range with quarterly variability rather than a clean expansion path.

AI TAM Sizing

• Aaron Rakers, Wells Fargo: Questioned AI TAM expansion implications of the OpenAI deal. Su: "the AI compute TAM just going up... there is a larger opportunity for us over the next few years."
  Assessment: The TAM commentary is consistent with what hyperscaler capex disclosures and Nvidia's data center commentary have been signaling for several quarters. AMD's increasing share of a still-growing pie is the simplest version of the bull case.

What They're NOT Saying

1. FY26 explicit revenue guide: Management did not provide an FY26 framework on the Q3 call — understandable timing-wise (full FY guide typically arrives with Q4 print in February) but notable given the OpenAI deal would warrant earlier framing. The market will fill the gap with bullish bottom-up models; the risk is they overshoot what AMD can comfortably deliver in 2026.
2. OpenAI deal revenue cadence by year: The "tens of billions of dollars" framing is intentionally non-time-specific. We know 1GW of MI450 in 2H 2026 (call it $2–4B contribution) but the back-end ramp through 2027–2030 is unspecified. Sell-side models will land in a wide range.
3. MI308 China licensing pathway: Continues to be an unresolved policy variable. AMD's continued exclusion of MI308 China revenue from forward guides suggests no near-term licensing optimism, but management has not closed the door publicly. The free option remains in the model.
4. Embedded recovery cadence reset: Three consecutive quarters of failing to inflect, and management did not reset the framing publicly. This will become a credibility issue if Q4 prints another decline.
5. Hyperscaler customer concentration: Even with OpenAI now disclosed, the broader Instinct customer concentration across the "7 of top 10 model builders" remains opaque. The OpenAI commitment is enormous but it does increase rather than decrease customer concentration in the Instinct franchise.
6. Capex trajectory and ZT Systems integration spend: No specific guide on capex acceleration to support the OpenAI / Helios / rack-scale buildout. AMD's historically capital-light model may need to evolve as it takes on more systems-integration scope.

Market Reaction

• After-hours move (Nov 4 evening): Stock declined approximately 3.7% in after-hours trading, to roughly $250.47 (some quotes had it down 3.9% to ~$246 by 7:42 PM ET). Initial reaction was driven by the Q4 GM guide of 54.5% being merely in line with StreetAccount consensus rather than ahead, and the perception that operating leverage in the Q4 model was insufficient given the recent stock rally.
• Pre-market and intraday Nov 5: Stock fell more than 5% in premarket and continued under pressure through the Wednesday session. Cumulative decline in the days following the print is approximately 9% per follow-up coverage.
• Pre-print context: AMD had rallied ~24% on Oct 6 alone on the OpenAI deal announcement, taking the stock to all-time highs heading into the Nov 4 print. Forward P/E entering the print was approximately 41x — elevated by historical AMD standards and specifically vulnerable to "sell the news" dynamics on any guidance that did not exceed already-bullish expectations. Market cap into the print was over $400B.
• Cited reasons for decline (sell-side aggregation): (1) Q4 GM guide met but did not beat StreetAccount consensus; (2) lofty pre-print valuation (forward P/E ~41x); (3) JPMorgan's "lack of stronger operating leverage" framing; (4) broader AI sell-off catalyzed by Michael Burry's disclosure of bearish options on Nvidia and Palantir; (5) "sell the news" dynamic post the Oct 6 OpenAI rally.
• Volume: Materially elevated vs 30-day average on both the AH session and the Nov 5 regular session.

The post-print sell-off is rational in our view: the stock entered the print at all-time highs after a one-month rally driven by the OpenAI announcement, and the Q4 guide — while above Street — did not deliver the kind of 10%-above-Street upside surprise that maximally-AI-bullish positioning was demanding. The Q3 print is operationally excellent and structurally re-rates the FY26+ thesis, but the immediate-term price action is being driven by valuation-reset dynamics rather than fundamental concerns. This is exactly the setup we listed as upgrade trigger (c) in our Q2 note: "a 15-20% drawdown without thesis impairment, to upgrade to Outperform." The drawdown is at ~9% as of T+1 cutoff and may extend further; the thesis is materially un-impaired (and in fact materially improved by the OpenAI deal and the Data Center reaccel evidence). The combination is the upgrade trigger firing.

Street Perspective

Debate: Is the Q4 GM guide of 54.5% a disappointment or appropriate normalization?

Bull view: 54.5% is a clean +50 bps expansion from Q3's 54%, sustains the snap-back from the Q2 trough, and represents normalized run-rate expansion as MI350 mix improves. The bears framing this as "in line is a disappointment" are setting an unrealistic bar of always-beat-on-guide; AMD has historically delivered modest beats on conservative guides, and 54.5% is structurally where we want margin to be heading into the MI450 ramp in 2H 2026.

Bear view: AMD's own October investor framing pointed to operating leverage acceleration as MI350 scaled. The Q4 guide of 54.5% (vs ~54% Q3 actual) is materially less expansion than that framing implied. JPMorgan's "lack of stronger operating leverage" critique is the right read; if MI350 is ramping at the pace AMD is signaling, GM should be expanding 100–150 bps sequentially, not 50.

Our take: The bull is roughly right. The 54.5% guide reflects mix headwinds from MI308 absence (which would have been higher-margin DC revenue had it been in the comp) plus the start of the OpenAI prep-investment phase. Pure quarterly GM-rate expansion will be choppy through 2026 as MI350/MI355/MI450 mix evolves; investors should be modeling OI dollars, not GM rate. The Q4 GM guide is fine.

Debate: Does the OpenAI deal change AMD's fair value framework?

Bull view: The OpenAI deal changes the multi-year revenue trajectory from speculative to anchored. With "tens of billions" of revenue contractually structured over the life of the partnership and OpenAI as a direct equity-aligned reference customer, AMD now has a multi-year revenue visibility profile that approaches Nvidia's. That justifies a multiple step-up of 4–7 turns of forward P/E vs the pre-deal framework. Combined with raised FY26+ revenue assumptions, fair value moves from the $130–160 we framed in Q2 to $230–280.

Bear view: The OpenAI deal is impressive but speculative until 2H 2026 deployment actually occurs. Customer concentration risk is now meaningfully higher (one customer for 6GW). Warrant dilution at full vesting reduces per-share economics. And historical precedent (e.g., past hyperscaler-supplier mega-deals) shows execution risk on multi-year deployments at this scale is real.

Our take: The bull is more right than wrong, but the bear's execution-risk point is real. We'd characterize fair value as moving to $200–240, allowing for some warrant dilution drag and execution discount, but acknowledging the deal is a structural re-rate. At the post-print pricing in the low-$240s area, the stock is roughly at our framework fair value — not screamingly cheap, but no longer clearly expensive given the strengthened multi-year visibility.

Debate: Is the Data Center +22% reaccel sustainable or peak quarter?

Bull view: +22% YoY with zero MI308 China and only the second full quarter of MI350 production is the new floor, not the peak. MI350 ramp continues through Q4 and into 1H 2026; MI355 layers on top in 1H 2026; MI450 adds OpenAI scale in 2H 2026. Each transition is additive, not substitutive. Data Center YoY growth should accelerate from +22% Q3 toward +30%+ by mid-2026.

Bear view: The +22% figure is partially flattered by sequential easy-comp dynamics from the Q2 trough. As AI infrastructure capex normalizes from the current peak buildout phase, AMD's incremental share gains may not translate into the +30%+ DC growth the bull case assumes. Hyperscaler ASIC programs (TPU, MTIA, Trainium) may also start to bite into merchant GPU TAM in 2026.

Our take: The reaccel is real and tracks our Q1/Q2 framing. ASIC competition is a 2027+ concern, not a 2026 one. We'd model FY26 DC growth in the +35–45% range with the OpenAI 2H contribution layered on, well above the bear case but below the most aggressive bull modeling.

Model Update Needed

Valuation impact: Net of the changes, fair value moves from our Q2 framework of $130–160 to $200–240. Drivers: (1) FY26 EPS estimates moving to ~$7.50 midpoint, (2) appropriate forward P/E expansion to 28–32x given improved visibility from the OpenAI deal, (3) Data Center growth trajectory more credibly tracking through 2026–2027. At post-print pricing in the low-to-mid $240s area, the stock is right around our framework fair value — with the OpenAI deal providing structural multi-year support that the prior "Hold" framework did not credit. The thesis is no longer constrained by valuation in the way it was through Q2.

Thesis Scorecard Post-Earnings

Overall: Eight bull points strongly confirmed (six existing + two new), one bear point worsened (Embedded), one bear point newly identified but well-managed (OpenAI concentration), and the central bear narrative (China overhang + execution risk) materially de-risked relative to our Q2 framing. This is the cleanest thesis-favoring print of the cycle to date.

Action: Upgrading from Hold to Outperform. Two of three pre-stated upgrade triggers fired this quarter: (a) tangible Data Center reacceleration evidence (+22% YoY with zero China contribution), and (c) meaningful drawdown without thesis impairment (~9% cumulative post-print on valuation reset, not fundamentals). The OpenAI deal arguably constitutes a fourth upgrade trigger we did not preview — a structural customer commitment that re-rates the multi-year revenue trajectory in a way our prior model did not credit. The setup post-print is the most attractive risk/reward we've seen on AMD in our coverage: operating thesis materially strengthened, multi-year visibility re-rated upward, and entry point improved by the post-print drawdown. We move to Outperform with conviction on the operating story and calibrated humility on near-term price-action volatility — the AI complex remains positioning-sensitive and Burry-style bear narratives are unlikely to fully fade through year-end.