Google expands Intel tie-up

Google just made a very un-Google AI move: instead of betting on one kind of chip, it signed up for multiple future generations of Intel processors for Google Cloud and expanded joint work on another class of data-center chip at the same time. That means the company building its own Tensor Processing Units is also locking in outside silicon for years ahead. (cnbc.com) (newsroom.intel.com) The first part of the deal is ordinary on the surface: Google Cloud will keep deploying Intel Xeon server processors, including the new Xeon 6 line, across artificial intelligence training, artificial intelligence inference, and general-purpose computing. Google already uses Intel in cloud products such as C4 and N4 instances, so this is an expansion of a long-running supplier relationship, not a brand-new one. (newsroom.intel.com) (cloud.google.com) The second part is the more revealing one: Intel and Google said they are expanding co-development of custom infrastructure processing units. An infrastructure processing unit is the traffic cop chip in a data center, handling networking, storage, and security jobs so the main processor can spend more time on customer workloads. (newsroom.intel.com) That sounds niche until you picture a warehouse where the forklifts keep stopping to do paperwork. If the infrastructure processing unit takes the paperwork away from the central processor, the same rack can push more useful work without adding another full server. (newsroom.intel.com) This is happening because the artificial intelligence market is shifting from training giant models to running them all day for users. Reuters reported that this deployment phase is reviving demand for general-purpose central processors, and CNBC quoted Nvidia’s head of artificial intelligence infrastructure saying central processors are becoming a bottleneck for newer agent-style workloads. (usnews.com) (cnbc.com) That helps explain why Google is widening its chip mix instead of replacing one chip family with another. In March, Google Cloud was still highlighting new Nvidia-based virtual machines at Nvidia’s GTC 2026 event, while this week it said Intel Xeon and custom infrastructure processing units will remain part of the same cloud stack. (virtualizationreview.com) (newsroom.intel.com) Google has also spent years building its own Tensor Processing Units, and it sells Arm-based Axion central processors in Google Cloud, so adding a deeper Intel commitment points to a portfolio strategy rather than a winner-take-all strategy. In plain English: Google wants shelves stocked with different tools for different jobs, not one giant box of identical hammers. (cnbc.com) (cloud.google.com) For Intel, this is one of the clearest signs yet that it still has a place in the artificial intelligence build-out even after Nvidia dominated the first wave. Intel’s announcement leaned hard on “heterogeneous” systems, which is industry shorthand for mixing specialized chips with general-purpose chips in one coordinated machine. (newsroom.intel.com) (cnbc.com) Wall Street heard that message immediately. Intel shares rose nearly 5% on April 9, 2026, after the announcement, while Alphabet closed only slightly higher, which is a clue about who needed the credibility boost more. (cnbc.com) The real takeaway is not that Google picked Intel over Nvidia, or Intel over its own chips. It is that the biggest cloud companies are building artificial intelligence systems like airports now: one chip hauls the passengers, another handles the baggage, and the whole place fails if the ground crew is too slow. (newsroom.intel.com) (virtualizationreview.com)