Google expands partnership with Intel on AI chips

Google will use multiple generations of Intel CPUs in its AI data centers, expanding a long-running partnership.

Intel said its Xeon 6 chips will handle AI training and inference workloads.

The companies did not disclose financial terms or a timeline for the agreement.

The deal comes as chipmakers push CPUs as a bigger part of AI systems, even as Nvidia remains the dominant supplier of AI accelerators.

Intel also said it is continuing work with Google on the infrastructure processing unit (IPU), while Google still develops its own TPU AI chips and started making its Arm-based Axion CPU in 2024.

🔗 Source: CNBC

🧠 Food for thought

Implications, context, and why it matters.

The partnership backs Intel’s x86, the long-running chip design used in most servers and PCs, while Arm-based processors gain ground in AI servers 1.
A clear challenge comes from Google’s Arm-based Axion CPUs. They reached up to a 60% price-performance gain versus x86 alternatives for customers like ZoomInfo, a business data software company, based on tests during the N4A preview 2.
Google still relies on Xeon CPUs for many data center tasks. Some jobs need backward compatibility with the x86 instruction set or need top single-thread speed 1.
The deal also covers co-developing custom infrastructure processing units (IPUs), chips that offload networking, storage, and security work from the host CPU, to help Xeon compete 1.

The agreement fits a mix-and-match approach in cloud infrastructure. Customers can pick from Intel CPUs, Nvidia GPUs, plus in-house chips like Google’s Axion 1.
Across the industry, performance tuning for a single dominant platform is fading.
Businesses now need software and engineering skills to shift workloads across processor types to manage costs. Tools include vLLM, an open-source software layer for serving large language models, which runs on GPUs and TPUs with minimal code changes 2.