Intel joins Google, Terafab plans

Intel is making two bets at once: sell more server brains to Google, and help build a factory system ambitious enough to feed Elon Musk’s artificial intelligence companies. Both moves landed within two days in April 2026. (intel.com) (cnbc.com) (techcrunch.com) Google’s side is the cleaner one. Intel and Google said on April 9, 2026 that they signed a multiyear deal to build the next generation of artificial intelligence and cloud infrastructure around Intel Xeon central processing units and Google’s custom infrastructure processing units. (bakersfield.com) (cnbc.com) A central processing unit is the general manager inside a server. A graphics processing unit is the specialist that chews through matrix math, and an infrastructure processing unit is the traffic cop that handles networking, storage, and security so the other chips stay busy. (cloud.google.com 1) (cloud.google.com 2) Google has already been mixing Intel Xeon chips with its own network hardware in cloud products. Its C3 virtual machines use 4th Generation Intel Xeon Scalable processors with Google’s custom Intel infrastructure processing unit, and its newer C4 machines use Intel Xeon 6 processors. (cloud.google.com 1) (cloud.google.com 2) (cloud.google.com 3) That tells you what this deal is really about. Google is not replacing its artificial intelligence accelerators with Intel server chips; it is trying to make the non-accelerator parts of an artificial intelligence data center faster, cheaper, and less power-hungry by pairing general-purpose Xeon chips with custom plumbing. (cnbc.com) (cloud.google.com) The Terafab side is rougher and much bigger. On April 7, 2026, Intel said it would join Terafab, the Musk-backed project with Tesla, SpaceX, and xAI that aims to build extremely large-scale semiconductor production in Texas. (reuters.com) (techcrunch.com) Terafab’s pitch is vertical integration. Instead of designing chips in one place, making memory in another, packaging them in a third, and testing them somewhere else, the project wants to put design, fabrication, memory, advanced packaging, and testing under one roof or one tightly linked campus. (interestingengineering.com) (electrek.co) Advanced packaging is the least glamorous part and maybe the most important. It is the step where separate pieces of silicon and stacks of high-bandwidth memory are wired together into one module, which is why the bottleneck in artificial intelligence hardware is no longer just who can etch the chip, but who can assemble the full package at scale. (electrek.co) (interestingengineering.com) Intel fits both stories for the same reason. It still sells Xeon processors to cloud companies, but it also wants customers for its foundry and packaging business, so a Google server deal and a Terafab manufacturing role both push Intel toward the same goal: being the company that helps build artificial intelligence infrastructure even when it is not selling the headline accelerator chip. (intel.com) (cnbc.com) (techcrunch.com) The simplest way to read this week is that artificial intelligence demand is now so large that the winners are chasing every layer of the stack at once. One deal is about the servers already inside Google’s cloud, and the other is about building enough factories and packaging lines so the next wave of servers can exist at all. (cloud.google.com) (reuters.com) (cnbc.com)