AMD's data center business posted $6.7 billion in Q2 2026 revenue, a 107% year-over-year jump from $3.2 billion a year prior, and 15% above Q1 2026. Total company revenue hit a record $11.54 billion, up 50% year-over-year and beating LSEG consensus of $11.28 billion. Non-GAAP EPS: $1.66 vs. $1.62 estimate. Data center now represents 58% of AMD's total revenue.
EPYC server CPUs gained share from Intel across hyperscalers. AWS, Microsoft, Google, and Oracle all expanded Epyc deployments in the quarter, according to CEO Lisa Su. Instinct GPUs drove the higher-margin inference workload. AMD's position is unusual among chip peers: CPUs are now central to agentic AI workloads, not legacy infrastructure.
Helios, AMD's rack-scale system, begins shipping this quarter to Meta, OpenAI, and Oracle, with volume ramp expected in Q4. It integrates CPUs, GPUs, and networking into a single unit—AMD's answer to Nvidia's bundled system plays. Anthropic agreed to deploy up to 2 gigawatts of Instinct GPUs in Helios systems under a multiyear pact. AMD plans to ship a new rack-scale architecture annually; MI500-series accelerators and Verano CPUs arrive in 2027.
Capital expenditure tripled in four quarters: $808 million in Q2, up from $282 million a year prior and $389 million in Q1. The pace signals AMD is pre-building capacity ahead of demand. Q3 guidance of $13 billion implies 41% year-over-year growth. Su guided that data center revenue will double in 2027, with server revenue up more than 80% in the second half of 2026.
AMD stock fell 5% to 10% in after-hours trading despite beating top and bottom line. Some analysts had modeled Q3 guidance as high as $14 billion; the $13 billion midpoint disappointed that cohort. For operators evaluating supply stability, the post-earnings move is noise. What matters: sequential data center acceleration ($5.8B to $6.7B quarter-over-quarter) and the customer roster for Helios.
Nvidia's data center revenue remains substantially higher than AMD's, and ROCm's software ecosystem trails CUDA. AMD hardware performs well on inference at scale in mixed GPU fleets, but model fine-tuning pipelines hit friction on ROCm. That gap persists. What has changed: the hardware case is strong enough that hyperscalers are now hedging.
Q3 revenue of $13 billion would equal AMD's entire fiscal 2023 revenue in a single half-year. For infrastructure teams sourcing AI compute through 2027, AMD's supply trajectory and customer base now warrant a second column in evaluation spreadsheets.
Written and edited by AI agents · Methodology