AMD and Intel are teaming up to simplify AI development on x86 CPUs, consolidating their respective optimizations for PyTorch under the open-source oneDNN library. The move aims to give developers a common software path across both companies' processors, eliminating some of the fragmentation involved in optimizing AI workloads for different CPU architectures.
As part of the agreement, AMD will integrate its existing ZenDNN optimizations into Intel's oneDNN, which becomes the default open-source library for x86 acceleration within PyTorch. This continues the companies' recent collaborations, where they most recently partnered to develop the AI Compute Extensions (ACE), a next-gen x86 ISA extension designed to accelerate matrix multiplication, with future integration planned through oneDNN updates.
The timing of this collaboration is hard to ignore. NVIDIA's RTX Spark laptops began arriving last week, bringing Blackwell graphics, a Grace CPU, up to 128GB of unified memory, and critically, NVIDIA's extensive AI software stack to consumer PCs. Meanwhile, Apple's Mac mini and Mac Studio have become a popular choice as compact systems for running local AI models and agents like OpenClaw.
NVIDIA's ambitions extend well beyond its traditional GPU stronghold, too. The company is pushing its Vera CPUs for agentic AI workloads, and AI cloud provider Modal recently announced plans to deploy them alongside its existing AMD and Intel processors. NVIDIA isn't content merely supplying the accelerators; it wants to be a one-stop shop for the computing infrastructure supporting the AI explosion.
These encroachments from Arm-based competitors put AMD and Intel in an interesting position. The firms remain viciously competitive rivals, as AMD's Zen 6-based EPYC 9006 "Venice" processors get ready to do battle against Intel's upcoming Diamond Rapids Xeons. However, either company would prefer x86 remains on top in the server market rather than cede marketshare to Arm-based competitors.
Unifying the software stack won't suddenly make x86 CPUs replacements for AI accelerators, but it could make them easier to deploy and optimize across competing hardware platforms. There's already some progress: oneDNN 3.14 introduces initial AMD Zen optimizations, although they require explicit opt-in configuration.
After years of explosive GPU growth, CPUs are in the spotlight again as increasingly sophisticated AI agents rely on CPUs to handle orchestration, data processing, tool calls, and code execution alongside their own GPU workloads. In a market increasingly shaped by NVIDIA's vertically integrated AI platform, a more unified x86 software ecosystem could help bolster x86's already powerful software support advantage.
A 30-year PC building veteran, Zak is a modern-day Renaissance man who may not be an expert on anything, but knows just a little about nearly everything.