FCF Tripled, But The Back-Half Growth Hasn't Shipped Yet
They doubled the server CPU TAM in six months and couldn't cite a single ASP number to justify it.
Thesis: The EPYC franchise is printing real cash and taking real share: server CPU +50% YoY, 1,600 cloud instances, record enterprise sell-through, FCF tripled to $2.6B. This is the base case and it does not require MI450 to succeed. The asymmetry sits in the back-half GPU ramp: Helios production hasn't started, MI450 ships below corporate GM, and the 2027 'tens of billions' rests on Meta/OpenAI forecast commitments. Own the stock for EPYC. Size the position knowing Q4 margin guide has a binary element.
Verdict: LONG — Conviction: MEDIUM
Catalyst: Advancing AI event in July: quantifies MI450 order book and the data center AI path to 'tens of billions in 2027.' Venice launch with more customer validations than any prior EPYC generation confirms the CPU flywheel.
Key Risk: Helios production shipment slip in H2 2026. MI450 dilutes Q4 gross margin per Jean Hu, and the entire DC AI growth story depends on a product that has not begun production. Combined with memory cost inflation guiding gaming down 20%+, a GPU miss removes both the growth narrative and the consumer buffer.
The Tell: Lisa Su was asked directly about the agentic AI ASPs underpinning the $120B TAM revision and answered: 'I don't have a number that I can tell you in terms of relative ASPs.' The entire TAM doubling rests on this component and she could not quantify it. The revenue growth she does cite is explicitly unit-driven.
Detected Patterns
Beat and Raise Machine: Revenue beat high end of guidance at $10.3B, Q2 guide up 46% YoY to $11.2B, GM guided to 56%. Fourth consecutive record server CPU quarter. Consistent pattern of exceeding and raising.
Margin Expansion Engine: GM up 170bps YoY to 55% on data center mix, guided to 56% in Q2. Operating margin 25%, EPS up 43%. FCF $2.6B confirms margin quality. Operating income scaling faster than revenue.
Structural Demand Shift: Server CPU TAM revised from $60B to $120B+ by 2030. CPU-to-GPU ratio moving from 1:4/1:8 toward 1:1 as agentic AI spawns CPU tasks. Every hyperscaler expanded EPYC footprint in Q1.
Backlog Fortress: Meta deal for up to 6GW of Instinct GPUs spanning multiple generations with custom MI450 accelerator. OpenAI partnership. Lead customer forecasts on MI450 now exceed initial plans.
AI Label Without Substance: TAM doubled from $60B to $120B in months. When pressed on ASPs for the agentic AI component: 'I don't have a number that I can tell you in terms of relative ASPs.' Revenue growth is unit-driven, pricing is cost passthrough.
Unsustainable Trend Confidence: 2027 'tens of billions' in DC AI rests on lead customer forecasts for MI450, which ships from a single point of failure: Helios second-half ramp that hasn't begun production. Forecast cancellable engagements with concentrated counterparties.
Capacity Ceiling: Memory is tightest in a decade. Gaming guided down >20% H2 vs H1 on memory costs. Client planning lower H2 shipments. Memory vendors are capturing value; AMD absorbs cost on consumer parts it cannot pass through.
China Revenue Decline Masked: Data Center AI declined sequentially in Q1 due to China revenue rolling off. Acknowledged only under direct analyst questioning by Lisa Su, never in prepared remarks. The Data Center segment still grew 7% sequentially on server CPU strength.
Mix Shift Spin: Data Center segment masks Data Center AI weakness ex-China. Jean Hu frames gaming decline as gross margin tailwind ('gaming coming down helps margin') while it actually represents revenue loss on the consumer side.
Say/Do Gap: OpEx grew 42% YoY vs 38% revenue growth. SG&A outpaced R&D in Q1, contradicting stated strategy of investing R&D ahead of sales. Jean Hu had to promise R&D will outgrow SG&A going forward.
Rising Customer Switching Costs: Customers standardizing on AMD across on-prem, hybrid, and AI workloads. 1,600 EPYC cloud instances up ~50% YoY. Record enterprise sell-through with new wins across financial services, healthcare, industrial.
Friction Level: MODERATE_FRICTION — Both sides agree on the numbers: FCF $2.6B, GM 55%, server CPU +50%. Disagreement is whether the TAM revision to $120B and the 2027 DC AI guide rest on forecast-cancellable hyperscaler commitments or structural demand.
Report not found
The report data is no longer available. Please return to the archive.
FCF Tripled, But The Back-Half Growth Hasn't Shipped Yet | Silicon Signal