AMD Posts Record $13.4 Billion Quarter as Data Center Revenue Nearly Doubles

Quick Facts

• AMD reported record Q3 2026 revenue of $13.4 billion, up 45% year-over-year, beating analyst estimates of $13.06 billion.
• Data center segment revenue hit $8.5 billion, up 97% year-over-year, driven by Instinct MI450 GPU sales and EPYC CPU demand from agentic AI deployments.
• Non-GAAP EPS came in at $2.09, topping Wall Street's $1.94 estimate, with non-GAAP operating income of $3.9 billion.

AMD delivered its strongest quarter on record in Q3 2026, with agentic AI workloads driving demand across both its GPU and CPU product lines. The results, detailed in AMD's official earnings release, pushed data center revenue to $8.5 billion, roughly 64% of total company revenue.

CEO Lisa Su said the company exceeded the high end of its own guidance. "Demand for our Instinct and EPYC products continued to strengthen, and we expect growth to accelerate further in the fourth quarter as we scale Helios and Venice production," she said.

CFO Jean Hu noted the data center AI business performed "better than our expectation when we got into the Q3 quarter." Non-GAAP earnings per share grew 74% year-over-year to $2.09, reflecting what Hu called "significant operating leverage" as data center scale increases.

What Is Driving the Numbers

The data center segment posted operating income of $2.9 billion, up 168% year-over-year, with operating margin expanding to 34% from 25% a year ago. Two products anchored the growth: the AMD Instinct MI450 Series GPUs and fifth-generation EPYC server processors.

Agentic AI, which involves AI systems that autonomously execute multi-step tasks, is creating a new demand pattern for AMD hardware. GPUs handle model inference while CPUs run the agent workflows that orchestrate those calls. AMD says large cloud providers including Amazon and Microsoft are deploying high-core-count EPYC chips specifically for agent execution and orchestration layers.

AMD has reached over 41% of server CPU revenue share, according to reporting on the earnings results, with EPYC gaining traction in premium data center sockets tied to AI infrastructure.

New Hardware for Agentic Workloads

AMD launched two distinct rack-scale systems targeting AI infrastructure. The first, AMD Helios, is built with 72 Instinct MI455X GPUs and 18 sixth-generation EPYC Venice CPUs per rack. A single Helios rack delivers up to 2.9 exaflops peak FP4 performance, 31 terabytes of HBM4 memory, and 1.7 petabytes per second of memory bandwidth. AMD says Helios delivers up to 30% more inference tokens per dollar than competing systems.

The second, introduced on October 12, is AMD Helion, an open blueprint for agentic AI rack infrastructure. Helion is based on an Open Compute Project spec-design and supports CPU-only configurations with up to 32,768 cores for agent execution at hyperscale, as well as combined CPU and GPU configurations for enterprise deployments. It includes an integrated token router that directs agent model calls across both frontier and open-source models.

Helion rack configurations are planned to be available in November 2026, according to AMD's SEC filing for the quarter.

Other Business Segments

Client revenue reached $2.9 billion, up 7% year-over-year on Ryzen mobile processor sales. The gaming segment pulled total client and gaming revenue to $3.7 billion, down 8% year-over-year as console-related revenue declined. Embedded revenue grew 29% to $1.1 billion.

Company-wide GAAP operating income rose 75% to $3.9 billion, with operating margin widening to 29%, up more than 500 basis points from the prior year period. Net income on a GAAP basis was $2.4 billion, or $1.46 diluted EPS, as detailed in the Q3 earnings call transcript.

What Comes Next

Su said AMD expects data center growth to accelerate in Q4 as Helios production scales and Venice CPU output increases. The company did not issue specific Q4 revenue guidance in the materials reviewed, but Hu characterized the fourth-quarter outlook as reflecting "accelerating revenue and earnings growth."

AMD's bet is that agentic AI infrastructure requires both compute types at scale, and that EPYC's position in hyperscale data centers positions the company to capture CPU spend that previously went elsewhere.