OpenAI and Broadcom agree to co-develop and deploy 10GW of custom AI chips to run OpenAI's models over four years; sources: the deal is worth multiple billions
The companies plan to deploy 10 gigawatts of custom AI chips over the next four years — OpenAI and Broadcom are working together …
Wall Street Journal
Context & Ripple Effects
This agreement turns an earlier report of an internally used OpenAI-Broadcom processor into a defined deployment program: the previously reported custom-chip effort was expected to ship in 2026 and be reserved for OpenAI's own use.
OpenAI and Broadcom now have a four-year plan to co-develop and deploy 10GW of custom chips for running OpenAI models, backed by a multibillion-dollar commercial commitment.
Broadcom gains a large, named deployment customer for a custom AI-chip program, while OpenAI secures a purpose-built hardware track for its internal infrastructure.
Second-order effects
The deal gives OpenAI a concrete alternative supply path alongside Nvidia systems; related coverage characterizes the arrangement as a diversification of OpenAI's chip suppliers across training and inference workloads.
A large internal deployment commitment raises the value of workload-specific chip design: suppliers competing for AI infrastructure business must offer more than broadly available accelerators, including design and deployment support.
Third-order effects
If other model developers make comparable commitments, AI hardware procurement could shift from buying standard accelerators toward long-duration, customer-specific capacity programs.
The parallel 10GW commitments point to AI compute being planned more like utility-scale infrastructure, where chip architecture, deployment timing, and sustained capacity are negotiated together.
The trend: Frontier AI developers are building heterogeneous, multi-year compute supply chains that combine general-purpose accelerators with custom silicon for specific workloads.
A few things I've picked up in the past few hours I found interesting: 1/ Racks- These are full racks Broadcom is deploying. That's a big deal. 2/ 5th XPU customer- it is apparent through the interviews that there is a 5th XPU customer. Confirmed. Many thought OpenAI was that.
For all of the people who doubted that these XPUs had legs, what are you saying now? Nvidia/AMD will still grow, but every XPU dollar is less dollars for GPU. There was a choice at least to invest in XPU versus even more GPU. I always said that there were efficiencies with XPUs
Ok so if the mystery 4th Broadcom customer isn't OpenAI then perhaps that $10b customer announce is Apple. Which, is would be more a networking IP play for Apple than full blown compute ASIC. $avgo $aapl
Announcing partnership with @Broadcom to build an OpenAI chip. This deal is on top of the @nvidia and @AMD ones we've announced over the past few weeks, and will allow us to customize performance for specific workloads. The world needs more compute.
We're partnering with Broadcom to deploy 10GW of chips designed by OpenAI. Building our own hardware, in addition to our other partnerships, will help all of us meet the world's growing demand for AI. https://openai.com/...
OpenAI confirms multi-year chip deal with Broadcom after 18 months of quiet collaboration — 10GW of custom AI infrastructure, no equity involved. Chips will be ~30% cheaper than Nvidia's, with OpenAI controlling everything from design to full rack deployment: