Intel and Google have announced a multiyear collaboration around Google Cloud infrastructure. The news comes at the same time as Intel announcing on X that it has joined Terafab, Elon Musk’s chip project involving Tesla, SpaceX and xAI.
The Google deal: CPUs and beyond
Intel said the companies will align across multiple generations of Intel Xeon processors to improve performance, energy efficiency and total cost of ownership across Google’s infrastructure.
Intel also said Google Cloud continues to deploy Xeon processors across C4 and N4 instances, and Google Cloud’s own documentation shows C4 instances running on sixth- and fifth-generation Intel Xeon processors and N4 on fifth-generation Xeon.
The companies are also expanding co-development of custom ASIC-based infrastructure processing units, or IPUs, which it described as programmable accelerators that offload networking, storage and security functions from host CPUs.
Google has already described earlier Intel-based IPU work in its infrastructure: in a 2024 post on Z3 instances, Google Cloud said the machine family combined fourth-generation Xeon processors with Google’s custom Intel IPU.
Terafab: bigger ambition, fewer details
The Terafab announcement came with fewer published details. Intel said in its X post that its capabilities would help accelerate Terafab’s goal of producing 1 terawatt per year of compute for future advances in AI and robotics. To put the ambition in perspective, Musk said at the project’s launch that current AI compute output is roughly 20 gigawatts per year.
Reuters had reported earlier, on March 22, that Musk said SpaceX and Tesla would build two advanced chip factories in Austin, with one aimed at cars and humanoid robots and the other at AI data centers in space, but that he did not provide a timeline for the project.
What Intel’s own filings say
Those two announcements also arrive against the backdrop of Intel’s own filings. In its 2025 annual report, Intel said it may pause or discontinue Intel 14A and successor leading-edge process technologies if it cannot secure a significant external foundry customer for 14A.
The same filing says Intel has had few external foundry customers to date, ramped Intel 18A into high-volume production in 2025 and is seeking customers for 14A, which it describes as its first node designed from inception for external customers.
It has also reported $307 million in Intel Foundry external revenue for 2025 and a $10.3 billion operating loss for Intel Foundry.
Two announcements, two different registers
The firm’s filing also says the company wants to build both an external foundry business and an ASIC and design services business for external customers. Against that strategy, the two April announcements point to different parts of Intel’s external-customer effort.
The Google deal is a disclosed CPU and IPU infrastructure collaboration. Terafab, by contrast, has so far been described publicly through Intel’s social post and Reuters’ account of it, not through a detailed Intel press release, regulatory filing or published commercial disclosure laying out scope, economics or process-node commitments.