Nvidia previously weighed a massive chip partnership with OpenAI, but OpenAI is now building its own chips, according to new reporting

OpenAI is moving to a more hardware-controlled approach for running its most important models, a shift that stands in contrast to earlier discussions that, according to recent reporting, involved Nvidia at very large scale.

The new reporting says Nvidia once explored a chip deal with OpenAI valued at up to $100 billion. It also says that deal did not proceed, and that OpenAI has instead moved toward designing its own AI chips from scratch.

In parallel with that change, OpenAI and Broadcom have unveiled a custom AI chip called Jalapeo, described as built specifically to power ChatGPT. The announcement indicates that OpenAI is treating hardware design as a core capability rather than a procurement issue.

For Nvidia, the development reinforces how competitive the AI chip market has become. Even with strong demand for general-purpose accelerators, the biggest customers can push toward custom chips to reduce costs, tailor performance, and improve control over supply and system design.

Jalapeo’s introduction also highlights the growing role of specialized partnerships. Rather than relying solely on a single vendor’s standard platforms, OpenAI’s approach, as described in the reporting, combines its model requirements with chip design and manufacturing support from another hardware heavyweight, Broadcom.

What is not clear from the reporting, and what has not been disclosed in the information provided here, is why the $100 billion exploration with Nvidia ended. The details of the earlier talks, including scope, timelines, and what performance targets or commercial terms were considered, were not described.

Company and sector context: custom AI silicon has become a strategic differentiator across the industry. For model developers, owning more of the compute stack can help manage latency and throughput needs for large-scale inference and training, while potentially improving unit economics as deployments expand.

Looking ahead, investors and industry watchers are likely to focus on whether OpenAI’s custom-chip roadmap reduces its reliance on Nvidia’s hardware over time, and whether Nvidia can defend its ecosystem position through new platforms, partnerships, or performance improvements. Another key question will be how quickly custom chips like Jalapeo move from announcement to broad deployment and measurable impact.