Intel–Google Xeon deal

Intel and Google said on April 9 that they signed a multiyear deal to keep Intel Xeon processors at the center of Google Cloud’s next wave of artificial intelligence infrastructure. (newsroom.intel.com) The companies said Xeon chips will continue to run Google Cloud systems used for artificial intelligence, inference — the step where a model answers a prompt — and general-purpose computing. They also said they will expand joint work on custom infrastructure processing units, or network-and-data-movement chips built as application-specific integrated circuits. (intc.com) A central processing unit is the general manager of a server, handling operating systems, memory, storage and the flow of work between accelerators. An infrastructure processing unit moves networking, security and input-output tasks off the main processor so more of the server can be used for customer workloads. (cloud.google.com) Google and Intel have already been shipping that design in public cloud products: Google introduced its C3 machine family in 2022 with 4th Gen Intel Xeon processors and a custom Intel-Google infrastructure processing unit, then brought Intel Xeon 6 into its C4 virtual machines in 2024. Google said C4 was the first leading hyperscaler offering based on Xeon 6. (cloud.google.com 1) (cloud.google.com 2) That matters in a market where Nvidia graphics processing units dominate the most visible artificial intelligence training systems, but cloud operators still need other chips to feed data, schedule jobs, run storage and serve lower-latency inference. Intel said the Google deal “reinforces the critical role” of central processing units and infrastructure processing units in “heterogeneous” systems that mix several kinds of silicon. (newsroom.intel.com 1) (newsroom.intel.com 2) Google has also been broadening its processor lineup rather than betting on one architecture. Alongside Intel-based instances, Google sells Arm-based Axion central processing units and has been pushing Titanium, its in-house system for networking and storage offload, across newer cloud machines. (cloud.google.com 1) (cloud.google.com 2) Intel, for its part, has been leaning on cloud and artificial intelligence partnerships as it tries to prove Xeon still has a durable role in data centers even when the headline spending is going to accelerators. In the last two months, Intel has also announced Xeon 6 collaborations tied to Nvidia systems, SambaNova inference systems and Cisco edge artificial intelligence gear. (newsroom.intel.com) (newsroom.intel.com) The deal does not replace Google’s own chip efforts or the wider race for graphics processors. It shows Google still expects future artificial intelligence fleets to be built from a mix of Xeon host processors, custom offload chips and accelerators, with each part handling a different job inside the same data center. (intc.com)