Broadcom deal with OpenAI stalls

Custom AI chips are the story here — not just another stock wobble. OpenAI and Broadcom had been selling a big idea: build OpenAI-designed accelerators at huge scale, lower dependence on Nvidia, and spread them across OpenAI and partner data centers through 2029. But that plan just ran into a very old problem — somebody has to finance the hardware before the savings show up. On Thursday, May 7, Broadcom shares fell after a report said the project had hit an $18 billion financing snag and negotiations had stalled while Broadcom waited to see whether Microsoft would commit to buying part of the chips. (finance.yahoo.com) ### What was this deal supposed to be? The public version of the partnership landed on October 13, 2025. OpenAI and Broadcom said they would collaborate on 10 gigawatts of OpenAI-designed AI accelerators, with Broadcom also supplying the Ethernet-heavy networking around them. (finance.yahoo.com)(openai.com) ### Why did investors care so much? Because custom silicon is one of the few believable ways for giant AI buyers to bend their long-term cost curve. Nvidia still dominates the market for training and inference hardware, but hyperscalers and model labs keep looking for chips tuned to their own workloads. If OpenAI could make its own ac(openai.com)d gain leverage over both pricing and supply. That is why Broadcom stock had jumped when the partnership became public. (openai.com) ### So what actually stalled? The reported problem is financing for the first phase of the buildout. The key detail is not just the $18 billion number. It is the structure around it. Broadcom is reportedly waiting for Microsoft to agree to purchase a portion of the chips before the project can move ahead. Yahoo Finance’s summary of th(openai.com)s demand locked in before it commits to a giant manufacturing and deployment ramp. (finance.yahoo.com) ### Why is Microsoft in the middle? Because Microsoft is still deeply tied to OpenAI’s infrastructure economics. Even as OpenAI pushes for more control over its own compute stack, Microsoft remains a crucial buyer, cloud partner, and financing backstop in the broader OpenAI e(finance.yahoo.com)city. That is a much easier pitch for everyone involved. This is an inference from the reported structure of the talks and the companies’ existing relationship. (finance.yahoo.com) ### Why does $18 billion matter so much? Because it shows how different “designing a chip” is from “deploying a chip fleet.” The hard part is not drawing up silicon. The hard part is paying for wafers, packaging, networking gear, racks, power, and data-center integration befo(finance.yahoo.com)hor customer hesitates, the whole schedule can slip. ### Does this kill the partnership? No — at least nothing public says that. The October 2025 announcement is still out there, and the reported issue sounds like a financing bottleneck, not a technical breakup. But the catch is that giant AI infrastructure plans now have to clear two tests, not one: can the chips work, and can the counterparties stomach the capital intensity? Investors were reminded this week that those are separate questions. (openai.com) ### Why is Broadcom getting singled out? Because Broadcom has become one of the market’s cleanest ways to bet on custom AI silicon beyond Nvidia. That cuts both ways. When a marquee customer deal looks bigger, Broadcom gets rewarded. When the same deal suddenly looks less certain, the stock reacts fast. Thursday’s drop was basically t(openai.com)ble customer program. (finance.yahoo.com) ### What should readers take from this? The AI buildout is still happening. But the money side is starting to bite. The biggest lesson from this Broadcom-OpenAI hiccup is simple — in AI infrastructure, demand headlines are cheap, financed capacity is the real milestone.