Google–Intel chip deal

The fight inside an artificial intelligence data center is no longer just Nvidia versus everyone else. Google said on April 9, 2026 that it will keep buying multiple generations of Intel Xeon chips while also building custom infrastructure processing units with Intel for its next cloud systems. (intel.com) A central processing unit is the general-purpose brain in a server, the part that handles lots of different jobs even when a flashy accelerator gets the attention. Intel said Google will keep using Xeon processors for artificial intelligence, inference, and general-purpose workloads across Google Cloud. (intel.com) An infrastructure processing unit is a helper chip that takes networking, storage, and security chores off the main processor, like moving the hotel front desk work away from the chef so the kitchen can keep cooking. Intel’s own product material says its infrastructure processing units are built for networking, storage, and other infrastructure workloads. (intel.com) That split matters because artificial intelligence servers spend a lot of time shuttling data, checking permissions, and feeding requests to models, not just doing raw math. Intel and Google said they are expanding co-development of custom application-specific integrated circuit infrastructure processing units to improve efficiency, utilization, and performance at scale. (intel.com) Google is not making this deal because it lacks its own chips. Google already sells Axion, its custom Arm-based processor family, and says those instances can deliver up to 65% better price-performance than other cloud instances for some general-purpose workloads. (cloud.google.com) Google is also already using Intel in live products. Google Cloud said its C4 and N4 machine families were the first from a major cloud provider to use 5th generation Intel Xeon processors, paired with Google Titanium offload hardware. (cloud.google.com) In the new agreement, Intel said the latest Xeon 6 chips are powering Google Cloud C4 and N4 instances, and those systems are aimed at jobs ranging from large-scale artificial intelligence training coordination to low-latency inference. (intel.com) That tells you what cloud companies are optimizing for in 2026. Reuters reported that as the market shifts from training giant models to deploying them widely, demand is rising for general-purpose processors that can keep those services running under heavy everyday workloads. (reuters.com) Intel needs this kind of customer just as much as Google needs optionality. CNBC reported that Google committed to using multiple generations of Intel chips, giving Intel a badly needed proof point in data centers while rivals push Arm-based designs and custom silicon. (cnbc.com) So the headline is not that Google picked Intel instead of its own chips. The headline is that one of the world’s biggest cloud operators decided the winning server is a mixed machine, with a general-purpose processor, a custom traffic-cop chip, and separate artificial intelligence accelerators each doing different parts of the job. (intel.com)