Intel and Google partner on AI systems

Google and Intel are betting that the most important chip in an artificial intelligence data center is not always the one doing the math. On April 9, 2026, the two companies announced a multiyear deal to keep Intel Xeon central processing units at the center of Google’s artificial intelligence and cloud systems while expanding work on Google’s custom infrastructure processing units. (intel.com) A central processing unit is the traffic cop of a server: it moves data, schedules jobs, handles memory, and keeps the rest of the machine fed with work. Intel said its Xeon processors will continue powering Google Cloud across artificial intelligence training, artificial intelligence inference, and general-purpose computing. (intel.com) An infrastructure processing unit is a helper chip that takes networking and input-output chores off the main processor, like moving baggage handling away from the pilot so the plane can keep flying. Google said its custom Intel infrastructure processing unit powers the C3 machine family and offloads parts of the Andromeda virtual network stack for performance, isolation, and security. (cloud.google.com) That division of labor is the point of this deal. Intel described modern artificial intelligence systems as “heterogeneous,” meaning one service now runs on a mix of central processing units, accelerators, memory, storage, and networking chips instead of one giant pool of identical processors. (intel.com) Google has spent years building custom chips for different jobs inside that mix. In April 2025, Google introduced Ironwood, its seventh-generation Tensor Processing Unit, and said it was the company’s first Tensor Processing Unit designed specifically for inference, which is the step where a trained model answers a prompt instead of learning from new data. (blog.google) Intel’s opening in that world is the part around the accelerator, not necessarily the accelerator itself. Intel said the new agreement expands co-development of application-specific integrated circuit infrastructure processing units, while Google’s existing cloud products already pair Intel Xeon with Google-built networking silicon in virtual machines such as C3. (intel.com) (cloud.google.com) Google has also kept shipping fresh Intel-based cloud machines even as the market fixates on graphics processing units. Google Cloud said its C4 virtual machines reached general availability with Intel Xeon 6, targeting workloads such as databases, analytics, gaming, and artificial intelligence that still need strong general-purpose compute around the accelerator layer. (cloud.google.com) The subtext is that cloud companies are trying to squeeze more useful work out of every rack, not just buy more graphics processing units. Google says its artificial intelligence infrastructure is built on its Jupiter data center network and offers both cloud graphics processing units and cloud Tensor Processing Units, which only works smoothly if the central processing units and networking chips keep models, storage, and users connected without bottlenecks. (cloud.google.com) Intel has been pushing exactly that argument in public. In March 2026, the company said Intel Xeon 6 was being used as the host central processing unit in NVIDIA DGX Rubin NVL8 systems, making the case that even the most accelerator-heavy artificial intelligence servers still need a host processor to orchestrate, secure, and scale the box. (intel.com) So this deal is less a surprise alliance than a formal acknowledgment of how giant artificial intelligence systems are actually built in 2026. Google will keep making specialized chips for specialized jobs, and Intel will try to own the layer that ties those jobs together. (intel.com)