Broadcom has halted production of OpenAI's custom AI chip over funding disagreements. The semiconductor manufacturer demands Microsoft purchase 40 percent of the chips before manufacturing begins. Microsoft has not committed to the deal.
OpenAI manager Sachin Katti labeled the dependency "financially unattractive" in internal communications. The initial production phase alone costs approximately 18 billion dollars.
The impasse reveals tensions within OpenAI's hardware ambitions. OpenAI has pursued custom chips to reduce reliance on Nvidia's GPUs and cut inference costs. Custom silicon typically delivers better economics at scale, but requires massive upfront capital.
Microsoft's hesitation appears rational. The company already invests heavily in OpenAI and commits significant Azure infrastructure to power ChatGPT and other services. A 40 percent commitment to a new chip represents additional capital exposure without guaranteed returns.
Broadcom's demand reflects hard economics. Semiconductor manufacturing requires enormous fixed costs. Broadcom built capacity assuming OpenAI would secure external demand guarantees. Without Microsoft's backing, Broadcom carries unacceptable risk if the chips underperform or OpenAI's needs shift.
This deadlock exposes a structural problem in AI infrastructure. Building custom silicon requires committing billions before demand crystallizes. Nvidia solved this through general-purpose GPUs that multiple customers adopted. OpenAI's bespoke design lacks that diversified customer base.
OpenAI could pursue alternatives. The company could negotiate with other chipmakers, scale back chip ambitions, or accept continued GPU dependency. Intel and Samsung both chase AI chip contracts. However, custom silicon projects typically lose momentum once stalled.
The chip project matters for OpenAI's long-term economics. Custom inference chips could cut costs by 50 percent or more compared to GPUs. Cheaper inference enables broader commercial deployment and higher margins. For Microsoft