Google expands Intel pact

Most artificial intelligence jobs in the cloud do not run on just one kind of chip. A graphics processor handles the heavy math, but a central processing unit still feeds data, manages memory, and runs the rest of the application around it. (intel.com) That is why Google and Intel just expanded a multiyear deal announced on April 9, 2026. Intel said its Xeon central processing units will keep powering Google Cloud across artificial intelligence, inference, and general-purpose workloads. (intel.com) Inference is the part where a trained model answers your prompt in real time, and it has become the expensive part of the artificial intelligence boom. Google said its Ironwood Tensor Processing Unit was built specifically for high-volume, low-latency inference, which shows how much cloud demand has shifted from training to serving answers fast. (blog.google) Intel is not just supplying off-the-shelf server chips here. The two companies said they are expanding co-development of custom application-specific infrastructure processing units, which are support chips that move networking, storage, and security work off the main processor so more of the system can stay focused on customer workloads. (intel.com) Google has already been building that kind of custom plumbing with Intel for years. In 2022, Google Cloud introduced its C3 machine series with a custom Intel infrastructure processing unit that helped offload networking and storage tasks from the server’s main processors. (cloud.google.com) The new pact also keeps Google Cloud on multiple generations of Xeon chips instead of treating the central processing unit as a commodity part. Google’s C4 virtual machines first launched on fifth-generation Xeon processors and later reached general availability on sixth-generation Xeon 6, giving customers newer memory bandwidth and processor options for databases, analytics, and artificial intelligence jobs. (cloud.google.com 1) (cloud.google.com 2) Energy use is the other half of the story. Intel said the expanded partnership is aimed at improving efficiency, utilization, and performance at scale, while Google’s own cloud guidance tells customers to cut artificial intelligence power use by choosing specialized hardware, reducing wasted resources, and placing workloads in lower-carbon regions. (intel.com) (cloud.google.com) This also shows Google hedging its artificial intelligence infrastructure instead of betting on one chip family. Google is pushing its own Tensor Processing Units like Ironwood for inference, while still deepening its Intel relationship for central processing units and custom infrastructure chips inside the same data centers. (blog.google) (intel.com) For Intel, that matters because cloud companies have been designing more of their own hardware and buying more graphics processors from rivals. A multiyear role inside Google Cloud gives Xeon a place in the artificial intelligence stack even as the flashiest workloads move to specialized accelerators. (intel.com) For Google, the payoff is tighter control over the whole machine, from the chip that answers the prompt to the chip that moves the data packet. In cloud computing, shaving a little power and a little delay off millions of requests is how a hardware partnership turns into a business strategy. (intel.com)