Intel–Google AI pact

Google just made a very public bet that artificial intelligence data centers still need more than giant graphics chips. On April 9, Intel said Google Cloud will keep using Intel Xeon processors for cloud workloads and expand joint work on custom infrastructure processing units, which are helper chips that move data, networking, and storage work off the main processor. (intel.com) A central processing unit is the general-purpose brain in a server. It handles the messy coordination work around artificial intelligence systems, including scheduling jobs, feeding data to accelerators, and running ordinary cloud software that customers still need beside model training. (intel.com) An infrastructure processing unit is more like a traffic cop bolted onto the server. Google has already used custom Intel infrastructure processing units in its C3 virtual machines so the main chips spend less time on networking and security overhead and more time running customer code. (cloud.google.com) That split matters because artificial intelligence has two different jobs. Training is the expensive part where a model learns from huge piles of data, while inference is the cheaper but nonstop part where the finished model answers prompts for millions of users. (cloud.google.com) Google already has its own Tensor Processing Units for the heaviest artificial intelligence work. Its latest Ironwood generation was introduced as a custom chip for large-scale training and high-volume inference, with Google saying it delivers more than four times the performance per chip of the prior generation Tensor Processing Unit v6e. (blog.google) So this Intel deal is not Google replacing its homegrown chips. It is Google filling in the rest of the machine room with Intel parts that keep cloud systems balanced, which is why Intel’s release talks about “heterogeneous” systems built from different chip types doing different jobs. (intel.com) Intel said Google Cloud is already deploying Intel Xeon 6 processors in C4 and N4 instances. Intel’s own description of C4 says those virtual machines can reach 4.2 gigahertz on the fastest cores and add native half-precision support, which is useful for some artificial intelligence and high-performance computing workloads. (intel.com) (community.intel.com) The timing matters because Google Cloud is also pushing Arm-based Axion central processing units and Google-made Tensor Processing Units at the same time. In November 2025, Google said Ironwood and Axion were both part of its artificial intelligence cloud push, which means Intel is winning a place in a data center that is no longer built around one supplier. (cloud.google.com) Intel needs that foothold. CNBC reported on April 9 that Google committed to multiple generations of Intel chips, a useful endorsement for a company trying to prove it still matters in artificial intelligence infrastructure after Nvidia became the default name for accelerator hardware. (cnbc.com) Google needs it too, for a different reason. Cloud providers sell reliability and price as much as raw speed, and custom infrastructure processing units plus standard server processors can lower bottlenecks in the parts of artificial intelligence systems that customers notice as lag, congestion, or higher bills. (intel.com) (cloud.google.com) What Intel and Google announced is less a moonshot chip reveal than a plumbing deal for the artificial intelligence era. The companies are saying the future cloud stack will mix Google Tensor Processing Units, Intel Xeon central processing units, and custom infrastructure processing units in the same buildings, with each chip handling the part of the job it is best at. (intel.com)