Intel expands AI partnerships

Intel just made two bets at once: one on cheaper artificial intelligence computing, and one on locking down the plumbing that runs it. On April 9, Intel and Google said they were extending a multiyear infrastructure partnership around Intel Xeon processors and custom infrastructure chips. (intel.com) The Google deal is not about flashy chatbots. It is about the warehouse-sized computers behind them, where central processing units handle general work and specialized chips take over the repetitive traffic control. (intel.com) Intel said its Xeon processors will keep powering Google Cloud systems for artificial intelligence inference, cloud services, and general-purpose workloads. Google and Intel also said they will expand co-development of custom application-specific infrastructure processing units, which are chips built for one narrow job instead of many. (intel.com) Google has already been mixing Intel central processing units with its own custom networking hardware in cloud products. Google Cloud says its C3 virtual machines use fourth-generation Intel Xeon chips with Google’s custom Intel infrastructure processing unit, and its C4 machines use Intel Xeon 6. (cloud.google.com 1) (cloud.google.com 2) That split matters because artificial intelligence systems do not run on one kind of chip anymore. Google’s own seventh-generation Tensor Processing Unit, called Ironwood, was introduced for high-volume inference, while Xeon stays in the stack for the broader computing jobs around it. (blog.google) (intel.com) Intel’s second move came through Anthropic’s new Project Glasswing, announced on April 7. Anthropic said the project is meant to secure critical software for the artificial intelligence era, and it launched with partners including Amazon Web Services, Apple, Broadcom, Cisco, CrowdStrike, Google, Microsoft, NVIDIA, Palo Alto Networks, and the Linux Foundation. (anthropic.com) Project Glasswing is basically a neighborhood watch for the software that keeps banks, clouds, and networks running. Anthropic says the group will give defenders early help from its Claude Mythos Preview model so they can find and fix weaknesses before attackers do. (anthropic.com) Intel’s role adds the chip layer to a project that was already heavy on cloud and security companies. That is why Intel described the effort as securing artificial intelligence infrastructure from silicon to software, meaning from the physical processor up through the code sitting on top of it. (anthropic.com) (intel.com) Put those two announcements together and the picture is pretty clear. The big artificial intelligence buildout is turning into a stack of long-term alliances, where one company supplies general compute, another supplies custom accelerators, and a wider club tries to keep the whole system from becoming a security mess. (intel.com) (anthropic.com) That is also a reminder that the real competition is no longer just model against model. It is cloud against cloud, chip stack against chip stack, and security posture against security posture, with Intel trying to stay essential even as the largest cloud companies keep designing more of their own hardware. (intel.com) (blog.google)