OpenAI's plans to develop a custom artificial intelligence chip with Broadcom have hit a significant financing obstacle, with the two companies locked in tense negotiations over who bears the risk on an initial $18 billion production phase.
The chip, internally called Jalapeno, is a custom inference processor designed by OpenAI to run its AI models more efficiently and cheaply than the general-purpose GPUs it currently buys from Nvidia.
The project is part of a broader 10-gigawatt data centre buildout announced with Broadcom last year and valued at roughly $500 billion over its lifetime.
The immediate problem is straightforward: Broadcom has told OpenAI it will only finance the first phase of chip production, which would consume 1.3 gigawatts of data centre capacity, if Microsoft agrees to purchase roughly 40% of the chips produced.
Under the proposed arrangement, Microsoft would buy the Jalapeno chips, install them in its own data centres and rent the computing capacity back to OpenAI.
Microsoft has not committed to that purchase.
OpenAI executives are privately discussing the risk of signing an agreement without a firm commitment from Redmond, with one internal memo acknowledging that if Microsoft does not follow through, the company would need to find an alternative buyer or financier for billions of dollars' worth of custom silicon.
The two companies did reach a revised vendor financing agreement under which Broadcom will put up more capital upfront, with OpenAI paying back the investment over time with interest, a structure OpenAI executives described internally as a "huge win."
But the unresolved Microsoft question remains the central obstacle.
The project is also running against a deadline. Broadcom needs to alert TSMC, the Taiwanese chip manufacturer that would fabricate the processors, to reserve production capacity in time for the chips to be ready in 2027. Without a conditional agreement soon, the manufacturing slot could be lost and the timeline pushed back further.
For OpenAI, having its own chip is a strategic priority that mirrors the path taken by Amazon, Google and Apple, all of which have developed custom processors to reduce costs and their dependence on Nvidia.
Related reading
- Sony deepens partnership with TSMC on AI image sensors
- Anthropic's Mythos jolts firms, but experts say risk already existed
- Brockman tells OpenAI trial Musk quit the company when he could not have control, and dismissed early ChatGPT…
OpenAI has been working on the project for more than two years, but the financing complexity reveals a tension at the heart of the AI infrastructure boom: even the most valuable private technology company in the world cannot easily finance chip production at this scale without binding commitments from its largest partner.
If the deal collapses, it would be a significant setback for OpenAI's ambitions to control more of its own hardware stack and would leave the company more reliant on Nvidia for the foreseeable future.
