Google expands an Intel chip tie-up

Google just widened an Intel relationship that most people thought of as ordinary cloud plumbing. On April 9, Intel said Google Cloud is now deploying Intel Xeon 6 processors across workload-optimized instances including C4 and N4, and the two companies are also expanding co-development of custom networking chips called infrastructure processing units. (intel.com) That sounds small until you remember how artificial intelligence systems are built. The graphics processors do the famous math, but central processing units still run the operating system, feed data to the accelerators, coordinate training jobs, and handle many inference tasks that are too small or too irregular for a graphics chip. (intel.com) Google has been making that central processing unit layer more visible inside its cloud. In August 2025, Google Cloud said its C4 virtual machines based on Intel’s sixth-generation Xeon chips were generally available, with claims of up to 30% better general compute performance and up to 60% better machine-learning recommendation performance than the prior generation. (cloud.google.com) Those C4 machines are not just rented servers with a new sticker on them. Google paired the Intel chips with its own Titanium system design and new local solid-state drives, and said the package cut storage access latency by up to 35% on those shapes. (cloud.google.com) The other piece in this deal is the infrastructure processing unit, which is a traffic cop for the data center. Google said earlier C3 machines used Google’s custom Intel infrastructure processing unit to offload networking and security work, and Intel now says that joint chip effort is expanding. (cloud.google.com, intel.com) Why bother adding more Intel when Google already has its own tensor processing units and can rent graphics processors from Nvidia and Advanced Micro Devices? Because giant cloud operators do not want one supplier deciding their costs, their delivery schedule, or the shape of every server they can sell. (intel.com) Intel is leaning hard into that pitch. Its April 2026 statement frames Xeon 6 as the host processor for large-scale training coordination, latency-sensitive inference, and general-purpose computing, which is a way of saying Intel wants to own the parts of artificial intelligence infrastructure that still need a general-purpose brain even when the flashy math runs somewhere else. (intel.com) Google has been building toward this multi-chip setup for two years. At Google Cloud Next 2024, it introduced C4 and N4 as new general-purpose machine families on Intel Xeon, and at the same event it positioned Titanium as the common layer underneath compute, storage, and networking. (cloud.google.com, cloud.google.com) So this is less a surprise partnership than a clearer map of how artificial intelligence data centers are being assembled. One company’s chip does the training math, another company’s chip moves data, and a central processing unit like Xeon 6 keeps the whole machine fed, scheduled, and usable enough that Google can actually sell it as a cloud product. (intel.com, cloud.google.com) For Intel, that matters because it does not need to beat Nvidia everywhere to stay important. If Google keeps buying Xeon chips for the control layer and keeps co-designing infrastructure processing units with Intel, then Intel still gets a seat in the most expensive part of the data-center buildout even as the market shifts toward mixed-architecture systems. (intel.com)