After the close on May 5, 2026, AMD reported its first quarter: data center segment revenue of $5.8 billion, up 57 percent year over year, on total revenue of $10.25 billion, up 38 percent. The stock jumped more than 14 percent in after-hours trading. But the structural signal was buried in the earnings call: CEO Lisa Su roughly doubled her server CPU market growth forecast for the next three to five years, from 18 percent to 35 percent annually, putting the market above $120 billion by 2030.
Su called the quarter “a clear inflection in our growth trajectory and a structural shift in our business” — data center is now the primary driver of AMD’s revenue and earnings. Against two years of narrative in which the AI infrastructure race was simply a GPU race, this report pulls CPUs back to the center of AI infrastructure.
The Numbers: Data Center Now Half the Company
- Data center revenue of $5.8bn, up 57 percent YoY, now more than half of total revenue, carried by Epyc server CPUs and ramping Instinct GPU shipments
- Net income of $1.4bn, up 95 percent YoY but down 8 percent sequentially; non-GAAP gross margin of 53 percent, one point lower than Q4, with operating income down 16 percent QoQ
- Operating expenses up 34 percent YoY; CFO Jean Hu pointed to record quarterly free cash flow
- Q2 guidance of $11.2 billion — roughly 46 percent YoY growth and 9 percent sequential, with non-GAAP gross margin around 56 percent
The sequential dips are not weakness. Inventory build, capacity expansion, and AI software ecosystem investment all land in the expense lines, and the market’s attention has already shifted to the MI450 ramp in the second half.
Why CPUs: Agentic AI Reshapes Server Demand
Su flagged in March that server CPU demand had “far exceeded” expectations, driven mainly by agentic AI applications. On this call she reiterated that inference and agentic workloads are increasing the need for server CPU compute, and said AMD is working with supply chain partners to “meaningfully increase” wafer and back-end capacity. She expects server CPU revenue to grow more than 70 percent YoY in Q2, with momentum running through 2027.
Analysts put it more bluntly. Stephen Sopko of HyperFrame Research argues inference and agentic AI “aren’t just incremental — they’re structurally shifting the CPU-to-GPU ratio higher” and expanding the total addressable server CPU market. Matt Kimball of Moor Insights says that as inference takes over, “that CUDA moat becomes less treacherous to cross”: token economics and performance-per-dollar are displacing raw performance as the procurement metric.
MI450, Helios, and the 6GW Deployments
- Meta (the 6GW agreement signed in February 2026): shipments begin in the second half of the year — custom Instinct GPUs based on the MI450 architecture, paired with sixth-generation Epyc “Venice” CPUs on the Helios rack-scale architecture, at power densities analysts estimate at 120–140 kW per rack with liquid cooling
- Helios timeline: initial volume in Q3, a significant ramp in Q4, continuing into Q1 2027
- OpenAI (the 6GW deal from October 2025): “going really well” per Su; OpenAI has previously said it will build a 1GW data center on MI450
- Tata Consultancy Services will design Helios systems aimed at enterprise and sovereign AI; AWS, Google Cloud, and Azure are expanding Epyc-based instances in parallel
Supply Chain and Power Visibility
Su’s answer to supply chain worries is worth keeping on record: AMD’s visibility into 2027 deployments now runs “down to which data centers the GPUs are going to be installed in” — necessary, she said, given all the constraints out there. She acknowledged tightness in the supply chain and in data center construction, but said AMD is confident it can supply to and exceed the growth levels it is forecasting, adding that “there is much more power that’s coming online in 2027.”
What It Means for Developers and Procurement Teams
Three practical effects. First, a real second source for GPUs: with Meta and OpenAI both placing gigawatt-scale orders with AMD, single-vendor architecture risk has an alternative, and ROCm maturity is the next thing to watch. Second, CPUs are a first-class citizen of AI clusters again — orchestration and real-time inference push Epyc demand up alongside GPUs, so cost models that only count accelerators are wrong. Third, 120–140 kW racks and liquid cooling are entering mainstream deployments, which invalidates a lot of standing assumptions in facility design and capacity planning. As our 2026 opening outlook argued, infrastructure constraints set this year’s rhythm — and this quarter is the latest footnote.
Sources
- AMD posts Q1 2026 data center revenue of $5.8bn, forecasts $120bn server CPU income by 2030 — Data Center Dynamics
- AMD’s AI Infrastructure Push Drives 57% Data Center Growth — Data Center Knowledge
AI-assisted summary compiled from the sources above, reviewed by a human before publishing.