AMD data center sales nearly double to $8.5B

The second source of AI compute just showed up in the numbers.

AMD reported on November 3 third-quarter revenue of $13.4 billion, up 45% from a year earlier. The growth came from the data center.

Data center carried the quarter

Data Center revenue was $8.5 billion.

Up 97% year over year.

AMD credits demand for its EPYC server processors and the launch of its Instinct MI450 Series GPUs. CEO Lisa Su said AMD expects growth to accelerate as it scales production of Helios, its rackscale systems for buying AI compute by the rack rather than by the chip.

The rest of the business was mixed. Client and Gaming revenue fell 8% to $3.7 billion, with gaming alone down 40% to $773 million. Embedded rose 29% to $1.1 billion.

GAAP gross margin was 54%, and 56% on a non-GAAP basis. Diluted earnings per share were $1.46 GAAP and $2.09 non-GAAP.

The guidance is the bigger number

AMD guided fourth-quarter revenue to about $17 billion, plus or minus 3%.

AMD says that midpoint would be roughly 66% growth from a year earlier and 27% growth from the third quarter. That is a company expecting the next three months to look nothing like the last three.

"Demand for our Instinct and EPYC products continued to strengthen," Su said in the release.

Who is running on it

The customer list is the part to read twice.

AMD says Anthropic, Meta, Microsoft, OpenAI and Oracle Cloud Infrastructure are running workloads on the first Helios rackscale systems. HPE will build Helios systems, with Vultr deploying the first of them. The EuroHPC Joint Undertaking picked AMD's Instinct MI430X GPUs for the LUMI-AI supercomputer.

That list is the labs whose models you use every day.

Yes, Nvidia still dominates AI chips, and one strong quarter does not change that. What changes is leverage. A lab with two suppliers negotiates differently than a lab with one, and some of that shows up in what you pay per token.

What it means if you buy compute

If you rent GPUs, AMD capacity on clouds and neoclouds is worth pricing against Nvidia for inference work, especially for models that already run well on AMD's software stack.

If you buy tokens from a lab, you will not see AMD on your invoice. You may see it in the price.

Either way, the software matters as much as the silicon. Most AI code was written for Nvidia's CUDA, so test your own workloads before you assume a cheaper chip is a cheaper result.

What to watch

AMD has set a $17 billion bar for the fourth quarter. Its next report will show whether Helios deployments turned into revenue on schedule.

Watch the fourth quarter. Then watch the per-token prices from the labs on AMD's customer list.