Google–Intel cloud deal

Google just made a very old supplier relationship look a lot more strategic. On April 9, Intel and Google announced a multiyear deal that keeps Intel Xeon processors inside Google Cloud and expands joint work on custom infrastructure chips. (intel.com) This is not about the flashy chips that train giant artificial intelligence models. It is about the chips that keep the whole data center moving: the general-purpose processors that run servers and the plumbing chips that handle networking, storage, and security jobs. (intel.com) Google said Intel Xeon 6 processors already power its C4 and N4 cloud instances. Those are the rented virtual machines customers use for everything from ordinary business software to parts of artificial intelligence training and fast-response inference, which is the step where a model answers a real user request. (networkworld.com) The new piece is the custom infrastructure processing unit, which is basically a traffic cop chip for a data center. Intel and Google said they will expand co-development of these application-specific integrated circuit chips so more network and storage work gets pushed off the main processor. (intel.com) That matters because a server running artificial intelligence is not just “doing math.” It is also moving huge amounts of data between chips, checking permissions, and shuttling results back to users, so every task moved off the main processor frees more room for the expensive compute chips to do actual model work. (cloud.google.com) Google is making this bet from a position of unusual strength. Epoch AI estimated that more than 60 percent of global artificial intelligence compute capacity sits with hyperscalers, and its April 2026 analysis said Google held the largest share among the biggest owners by the end of 2025. (epoch.ai) (networkworld.com) What makes Google different is that it did not build that lead by buying only Nvidia chips. Epoch AI said Google’s position was driven heavily by its in-house tensor processing units, which are custom chips built for artificial intelligence workloads, with total capacity roughly equivalent to about 5 million Nvidia H100 graphics processors. (epoch.ai) So this Intel deal is not a retreat from custom hardware. It is Google filling in the rest of the machine around those tensor processing units with a longer-term supply plan for central processing units and support chips, the parts that make a giant artificial intelligence cluster usable and rentable. (intel.com) (cnbc.com) It is also a reminder that “cloud” now means a few companies owning a very large share of the world’s computing muscle. When one provider designs the artificial intelligence chip, buys the server processor, runs the network, and sells the service, customers get speed and scale but also deeper dependence on that provider’s stack. (epoch.ai) (networkworld.com) That dependence shows up in boring places before it shows up in dramatic ones. Internal controls teams need audit evidence, performance records, and security attestations that can travel across vendors, and that gets harder when the hardware, networking layer, and management tools are all tuned to one cloud’s custom design. (cloud.google.com) (networkworld.com) For Intel, the announcement is a foothold inside one of the few companies big enough to shape the next generation of data-center architecture. For Google, it is a way to lock in a multi-generation processor roadmap while it keeps building an artificial intelligence empire that is increasingly custom at the center and highly partnered at the edges. (intel.com) (cnbc.com)