Clean Beat, Cratered Q2 Margin: MI308 China Ban Punches a $1.5B Hole in the FY25 Story — Initiating at Hold

Results vs. Consensus

Q1 2025 was a clean broad-based beat against both Street consensus and AMD's own guidance. The magnitude of the revenue beat (~4.4%) and EPS beat (~3.2%) are typical for AMD's recent cadence; what makes the print qualitatively different is the composition — Data Center and Client both posted growth rates that exceeded even the most optimistic published estimates, while Gaming continued its expected semi-custom-led slide. The "quality of beat" question for AMD this quarter is not whether the upside was organic (it was), but whether it is durable into 2H given the China overhang — addressed in the segment and outlook sections below.

Quality of Beat

• Revenue: Fully organic. No M&A contribution (ZT Systems closed late in Q1 and was immaterial to revenue this quarter; it consolidates as part of the Data Center segment going forward but is essentially a systems-integration capability, not a revenue line). No FX tailwind of consequence. Beat composition skewed to the two segments where it matters most for the AI thesis: Data Center (+57% YoY) and Client (+68% YoY). Some sell-side desks worried about Client tariff pull-forward; CEO Lisa Su explicitly rejected this on the call.
• Margins: Non-GAAP GM of 54% is the structural high-water mark and reflects favorable mix (high-end EPYC, premium Ryzen, growing Instinct ASPs) plus operating leverage. The 200 bps YoY expansion is sustainable on mix; what is NOT sustainable is the absence of the Q2 MI308 reserves charge, which will compress reported GM by ~1,100 bps next quarter. Adjusted for the one-time charge, the GM trajectory remains intact.
• EPS: The $0.03 EPS beat was operationally driven; share count and tax rate were not material swing factors. Below-the-line items were clean.

Segment Performance

Data Center

The segment that matters most printed +57% YoY to $3.7B with $932M of operating income (~25% segment OI margin), exceeding even the higher end of buyside expectations. The growth was a balanced mix of (a) 5th-generation EPYC ("Turin") ramping at every major hyperscaler — AMD added more than 30 new public-cloud instance launches in Q1 alone — and (b) Instinct GPU shipments where MI325X is now in production deployment with major frontier-model developers for both training and inference workloads. EPYC server CPU revenue grew double-digit YoY for the seventh consecutive quarter, and the deployed-customer footprint is now the entire Forbes 2000 top-10 in telecom, aerospace, and semiconductors, plus 9 of 10 in automotive.

"GPUs have a larger role due to the changing models and competitive TCO." — Lisa Su, CEO — defending against the custom-ASIC competitive narrative

Assessment: The +57% Data Center growth print is the cleanest possible answer to the bear question of whether AMD is winning real share in AI accelerators. The segment is now annualizing ~$15B and growing in the high-50s. The bull thesis on Data Center remains intact through Q1; the question for the next two quarters is whether MI355 (mid-year launch) and the MI308 China hole net out positive or negative. Su was explicit that they expect "strong double-digit" Data Center GPU growth for the full year despite the China headwind, implying significant 2H acceleration on MI355.

Client

Client revenue surged +68% YoY to $2.3B, the fifth consecutive quarter of share gains against Intel. The driver was richer mix (high-end Zen 5 Ryzen on both desktop and mobile) producing a record client CPU ASP, not unit-volume optimism. Some Street desks worried this reflected tariff-related pull-in demand; Su was unusually direct in rejecting that interpretation, citing "strong ordering patterns" without "lot of tariff related activity."

"We saw record client CPU ASP, driven by a richer mix of high end desktop and mobile Ryzen processors." — Lisa Su, CEO

Assessment: Client is now the secondary AMD growth engine and its ASP-led story is durable as long as Intel's product cadence remains soft. The 68% growth print is unlikely to be repeated — the comp gets harder — but the share-gain trajectory is structural. The segment is moving from cyclical PC exposure toward a premium-mix story that materially expands the operating model.

Gaming

Gaming revenue fell 30% YoY to $647M, entirely driven by semi-custom (console SoC) softness as Microsoft and Sony work through the back end of the Xbox Series X/S and PlayStation 5 cycles. Discrete Radeon shipments were actually up YoY and the Radeon 9070 series posted a record first-week sellout at launch, but the dollar contribution is not large enough to offset the semi-custom decline.

Assessment: Gaming is in the trough of the console cycle, which is an entirely cyclical phenomenon and not a competitive issue. Recovery requires a next-gen console announcement — not in our forecast for 2025. Treat this segment as a slowly-shrinking ballast for now; it is not the marginal driver of the AMD investment case.

Embedded

Embedded (the legacy Xilinx franchise) was down 3% YoY at $823M with $328M of operating income (~40% margin). The revenue weakness is the tail end of the inventory digestion cycle in industrial and comms end markets that began in 2023. Management expects sequential improvement in Q2 and Q3 with a return to growth in 2H.

Assessment: Embedded margins are doing the heavy lifting at 40%+ — this segment is a profit cushion regardless of revenue cycle. Watch for the inflection back to YoY growth as a positive catalyst, expected late-2025.

Key Topics & Management Commentary

Overall Management Tone: Confident on the operating execution but visibly defensive on the China policy variable. Lisa Su opened with "We delivered an outstanding start to the year despite the evolving dynamics related to tariffs and the regulatory environment" — the qualifier "despite" is doing heavy lifting and was repeated through the call. The tone shift versus prior quarters is notable: prior quarters opened with the Data Center growth narrative; this one led with a regulatory caveat.

The MI308 China Export Ban

The single biggest topic on the call. New US export-license requirements imposed in mid-April 2025 effectively block AMD's MI308 (a China-specific export-compliant Instinct variant) from shipping to Chinese customers. The financial impact: ~$700M of Q2 revenue lost, ~$800M Q2 inventory and reserves charge, and ~$1.5B of total FY25 revenue impact concentrated in Q2/Q3. The $800M charge is the optics killer for the Q2 print — it drags reported non-GAAP gross margin from a 54% run-rate to 43% in a single quarter.

"We continue to expect strong double digit percentage revenue growth in 2025." — Jean Hu, CFO — on the FY25 outlook despite the China hole

Management's framing is that the China headwind is "well contained" (Su) and offset by MI355 ramp in 2H. The math implied by the FY25 outlook is that ex-China Data Center GPU revenue accelerates meaningfully in 2H to absorb the missing China dollars and still leave aggregate growth in the "strong double digits." That is an aggressive assumption that depends entirely on MI355 production execution.

Assessment: This is the single biggest variable in the AMD investment case for the next 6–9 months. The financial hole is now quantified, but the policy variable is unquantifiable: licenses could be granted (upside), the ban could be tightened to other SKUs (downside), or the rules could be re-written entirely. We treat the China revenue line as effectively zero through year-end and watch MI355 execution as the only mitigant under management's control.

The MI350 / MI355 / MI400 Roadmap

The roadmap was reaffirmed with no slips. MI350 sampled to customers in Q1 with production ramp targeted for mid-year. MI355 is the high-volume FY25 SKU and is positioned to deliver the 2H Data Center GPU acceleration that is required by the FY25 guide. MI400 remains on track for 2026 with rack-scale capability enabled by the just-closed ZT Systems acquisition.

"We are excited about the MI350 launch, which is on track for midyear. Customer interest is high." — Lisa Su, CEO

Assessment: Roadmap execution is the bull case in this print. MI350/MI355 mid-year launch is the variable that converts Q1's ~57% Data Center growth into the "strong double-digit" FY25 outcome management is committing to. Slippage on this would invalidate both the FY25 guide and any constructive 2H scenario.

ZT Systems and the Rack-Scale Pivot

The ZT Systems acquisition closed late in Q1 and is now being integrated. Management framed it as the missing capability for rack-scale AI deployments — the ability to ship pre-integrated racks of GPUs, CPUs, and networking the way Nvidia ships NVL72 rack-scale systems. JPMorgan's analyst pressed on this point, and Su confirmed customer enthusiasm is "very high" for MI400-era rack-scale offerings enabled by ZT.

Assessment: Rack-scale is a real strategic gap that ZT closes — but the meaningful financial payoff is an MI400 / 2026 story, not a near-term lever. For Q1/Q2 2025 ZT is dilutive optically (integration costs) and net-neutral fundamentally.

EPYC Share Gains: Quietly the Best Story in the Print

Server CPU posted its 7th consecutive quarter of double-digit growth, and the deployed-customer narrative continues to broaden: every top-10 telecom, aerospace, and semiconductor company in the Forbes 2000, plus 9 of 10 in automotive, are now EPYC customers. 30+ new hyperscaler instance launches in Q1, 150+ Turin-based platforms shipping from Dell, HPE, Cisco, Lenovo, and Super Micro. This is the share-gain story that does not depend on AI accelerator outcomes.

Assessment: EPYC remains the underappreciated workhorse of the AMD story. It does not need MI355 to ramp to keep growing double-digit, and it provides the structural margin uplift that is showing up in non-GAAP GM expansion. If the AI accelerator story disappoints, EPYC is the floor.

Custom Silicon Threat (ASIC Narrative)

Su spent meaningful airtime defending against the increasingly loud sell-side narrative that hyperscaler-designed custom AI silicon (Google TPU, Amazon Trainium/Inferentia, Microsoft Maia) erodes the merchant-GPU TAM. Her counter: model architectures are still evolving, TCO favors flexible GPU compute, and AMD's GPUs sit alongside — not against — ASICs in most hyperscaler footprints.

"GPUs have a larger role due to the changing models and competitive TCO." — Lisa Su, CEO

Assessment: The ASIC threat is real but slow-moving and AMD is correct that the pie is growing fast enough to accommodate both architectures. We don't think this is a near-term thesis risk; it becomes a 2027+ TAM question.

Guidance & Outlook

The Q2 revenue guide of ~$7.4B is essentially in line with the pre-print Street consensus of $7.35B — an impressive outcome given the $700M China revenue removal, because it implies the rest of the business (Client + Gaming sequential, Embedded recovery, ex-China Data Center) is collectively offsetting the loss. Segment color: Client + Gaming up double-digit sequentially; Embedded roughly flat; Data Center down sequentially (the China hole).

The 43% non-GAAP gross margin guide is the print's gut-punch optic. Excluding the $800M MI308 charge, the underlying GM run-rate is ~54% — right at Q1 levels — so the operating leverage story is not impaired. But the reported number is what algorithms and headlines will see, and a 1,100 bps QoQ GM compression will scare price-sensitive holders even if it is fully one-time.

Implied 2H ramp: To hit "strong double-digit" FY25 revenue growth (call it ~25% to be generous to management's framing), with Q1 at $7.44B and Q2 at $7.4B, the 2H needs to print roughly $17B–$18B combined — or ~$8.5B–$9B per quarter. That requires Q3 sequential growth of ~15% and Q4 sequential growth of ~5–10%. The MI355 ramp is essentially the entire delta.

Street at: Pre-print FY25 consensus revenue was $31.31B and EPS $4.40. The Q1 beat plus the Q2 in-line revenue guide on a $1.5B FY25 China hole means consensus revenue likely settles around $31.0–31.5B and EPS in the $4.20–4.40 range as Street analysts model the GM trough. We would expect FY26 estimates to hold or rise, given MI400 enthusiasm.

Guidance style: Historically AMD has guided conservatively under Lisa Su; the in-line Q2 revenue guide despite the China hole is consistent with that posture but signals confidence in the rest of the business. The "strong double-digit" FY25 framing is more aggressive than typical and depends materially on MI355 execution.

Analyst Q&A Highlights

China / MI308 Hole and Backfill

• Vivek Arya, Bank of America: Pressed on how China restrictions reshape the GPU TAM and FY25 outlook. Su responded that China headwinds were "well contained" and reaffirmed "strong double digits" full-year GPU growth.
  Assessment: Management has staked the FY25 narrative on MI355 backfilling China. The "well contained" framing is optimistic; we'd characterize it as "quantified, not contained."
• Stacy Rasgon, Bernstein: Pressed on visibility into Q3 / 2H given the MI308 hole and what specifically backfills the China revenue. Hu pointed to the MI355 ramp in 2H as the bridge.
  Assessment: This is the bear's right question. The answer depends entirely on MI355 production execution — a roadmap-execution variable rather than a demand variable.

Client Sustainability

• Joshua Buchalter, TD Cowen: Asked whether the +68% Client beat reflected pull-forward from tariff anticipation. Su was unusually direct: "strong ordering patterns" without "lot of tariff related activity," citing ASP expansion (high-end Zen 5 mix) and desktop channel outperformance.
  Assessment: We give management the benefit of the doubt here, but the Client growth print compares against an easy YoY comp. Watch Q2 sequential Client — if it's only modestly up, the tariff-pull-in skeptics will get more vocal.

MI400 and Rack-Scale

• Harlan Sur, JPMorgan: Questioned whether MI400 platform deployment will match Nvidia's NVL72-class rack-scale offerings. Su confirmed ZT Systems is the rack-scale enabler and customer enthusiasm for MI400 is "very high."
  Assessment: Rack-scale is the strategic gap that ZT closes for the MI400/2026 cycle. Today's Q1 print does not depend on this; the FY26 narrative does.

Server CPU Share

• Aaron Rakers, Wells Fargo: Pressed on EPYC share gains and durability. Su highlighted the 7th consecutive quarter of double-digit growth and the breadth of customer adoption across enterprise and cloud.
  Assessment: This is the most under-discussed positive in the print. EPYC is doing the structural margin work and is independent of AI accelerator volatility.

Gaming Trough

• Timothy Arcuri, UBS: Asked when Gaming bottoms. Su pointed to semi-custom softness in 1H and seasonal recovery in 2H.
  Assessment: Gaming is cyclical; the next leg up requires a next-gen console announcement which is a 2026 event at earliest.

What They're NOT Saying

1. Specific China license outlook: Management did not signal whether they have applied for MI308 export licenses, what dialogue (if any) they are having with the Commerce Department, or what scenario they are modeling for license outcomes. The complete absence of this discussion suggests the licensing pathway is either unclear or unhelpful, and it implies management is modeling zero China revenue for the rest of FY25.
2. MI308 inventory disposition: Management took the $800M reserves charge but did not articulate what happens to the physical inventory — whether it can be re-spec'd for non-China markets, written off entirely, or held in hopes of a license. The lack of clarity here suggests the asset has limited residual value.
3. Hyperscaler customer concentration in Instinct: The Instinct GPU revenue continues to be discussed in aggregate without specific customer-concentration disclosure. This matters because the bull case for a multi-customer Instinct franchise differs sharply from a story dependent on one or two large frontier-model labs.
4. Operating expense growth assumptions: R&D and SG&A guidance was not granular — given the ZT Systems integration and the MI355/MI400 development push, opex leverage assumptions for 2H are unclear.
5. Embedded H2 inflection magnitude: Management said Embedded returns to growth in 2H but did not quantify the recovery magnitude or attribute it to specific end markets. Given Embedded carries 40%+ operating margins, the magnitude of this recovery has outsized profit implications.

Market Reaction

• After-hours move (May 6 evening): Initial pop of ~+4% on the headline beat, then receded to roughly flat / slight decline once Lisa Su walked through the China export-control impact and the 43% Q2 GM guide on the conference call.
• Next-day intraday (May 7): Pre-market opened lower as headlines digested the $1.5B FY25 China hole; by mid-morning Wednesday shares were down approximately 7–8%.
• Volume: Materially above 30-day average on May 7 as the Street re-modeled FY25 estimates for the China impact and the GM trough.
• Pre-print context: AMD entered the print already down ~16.7% YTD vs. the S&P -3.9% YTD — the market had partially anticipated the China bad news (the $800M charge was first disclosed April 16), but not the magnitude of the Q2 GM trough.

The reaction makes sense in context: this is not a "bad quarter" reaction — the operating beat was real — it is a "the next two quarters look ugly and the policy variable is unquantifiable" reaction. The market is rewarding the EPYC and Client execution but penalizing the loss of FY25 visibility. We think both reactions are defensible; the key question is whether the post-print drawdown overshoots the underlying impairment to FY25 earnings power. Our view: the FY25 EPS revision is real (call it ~5–10% lower than pre-print consensus), the FY26 EPS revision is small or zero (MI355/MI400 still ramp), and the policy tail risk is unquantifiable but not zero.

Street Perspective

Debate: Is the $1.5B China hole a one-time event or a multi-year impairment?

Bull view: The hole is fully quantified at $1.5B for FY25, concentrated in Q2/Q3, and will be backfilled by MI355 ramp in 2H. There is also an upside scenario where licenses are eventually granted and some China revenue returns. The structural Data Center growth trajectory is unchanged.

Bear view: The China ban is a forward signal — today MI308, tomorrow potentially other Instinct SKUs — and the policy uncertainty creates a structurally lower China contribution to AMD's TAM for the foreseeable future. The MI355 backfill is a roadmap-execution promise, not a demand confirmation.

Our take: The bull view is correct on the immediate $1.5B math; the bear view is correct on the policy tail risk. We treat China revenue as a binary policy variable that is properly modeled at zero through year-end. The MI355 backfill is the right plan, but plans are not prints.

Debate: Has AMD durably won merchant-GPU share against Nvidia?

Bull view: +57% Data Center growth, MI325X in production at frontier-model builders, MI355 mid-year, MI400 + ZT rack-scale in 2026 — AMD is the credible #2 in AI accelerators and capturing meaningful merchant share as customers diversify away from Nvidia single-source.

Bear view: AMD's Data Center GPU is still <10% of Nvidia's at scale, software ecosystem (ROCm vs. CUDA) remains a meaningful moat for Nvidia, and the customer narrative is concentrated rather than broad-based. The growth rate looks impressive off a small base.

Our take: Both are true at the same time. AMD has won the credibility battle (the "second source" narrative is now consensus) but the absolute share gain is still modest. The story for the next 18 months is whether Instinct can scale from a single-digit-billion-dollars franchise to a clear $15B+ franchise. MI355 and MI400 will determine the answer.

Debate: Is the EPYC share-gain story underappreciated by the market?

Bull view: 7th consecutive quarter of double-digit server CPU growth, broadest Forbes 2000 deployment footprint AMD has ever had, 30+ new hyperscaler instances in a single quarter. EPYC is producing structural margin expansion that is independent of any AI accelerator outcome, and the market is paying nothing for it.

Bear view: Intel's roadmap recovery (Sierra Forest, Granite Rapids) will eventually arrest AMD's share-gain pace; the easy comps are in the rearview mirror; ARM (Graviton, etc.) creates competitive pressure on the upper end.

Our take: The bull view is closer to right. EPYC is the structural floor in this story and is significantly underweighted in the bull/bear AMD discourse, which is dominated by Instinct. If the AI accelerator narrative cools, EPYC is the durable franchise that justifies a baseline valuation regardless.

Model Update Needed

Valuation impact: FY25 EPS revision of ~5–7% lower (China hole + GM trough) partially offset by FY26 EPS holding or rising (MI355/MI400 ramp). At a 20–25x forward P/E on FY26 estimates, fair value is in the $115–145 range — a band that brackets the current trading level after the post-print drawdown. The valuation is no longer obviously stretched, but it is also not yet obviously cheap.

Thesis Scorecard Post-Earnings

Overall: Operating thesis is strengthened on Data Center, EPYC, and margin trajectory; impaired on China policy variable and Q2 reported optics. Net: thesis is roughly unchanged in long-term economic value but visibly impaired in near-term optics and visibility.

Action: Initiating at Hold. We want to own the franchise but not at this entry point given the policy overhang and the visible Q2 GM trough. We would upgrade to Outperform on (a) clarity on the US-China licensing posture, (b) tangible MI355 production ramp evidence in the Q2 print, or (c) a 15–20% drawdown without thesis impairment.