OpenAI partners with Broadcom to make an ideal AI chip (and get distance from Nvidia)

OpenAI and Broadcom have formed a multibillion-dollar partnership to develop OpenAI-designed chips.

Under the deal, OpenAI will design the chips to its own specifications and Broadcom will manage the development and fabrication of the chips, as well as help with their deployment. The companies plan to deploy enough chips to require 10 gigawatts of electrical power beginning in mid-2026, and running through 2029. Broadcom stock jumped almost 10% on the announcement Monday.

The deal marks the second major move by OpenAI to reduce its dependence on Nvidia, which now dominates the AI chip market—the company announced a partnership with chipmaker AMD last week.

The Broadcom partnership is part of a many-pronged effort by OpenAI to dramatically scale up the computing power it uses to train and operate AI models. Within its “Stargate” initiative to add data centers, the company has attracted large investments from Softbank, Oracle, Nvidia, AMD, and MGX. Coreweave, Microsoft, and ARM will play supporting roles. OpenAI also signed a $300 billion deal with Oracle to buy computing power within existing Oracle data centers.

The massive infrastructure investments represent the largest single force buoying up the U.S. stock market.

AI companies and their investors believe that generative AI systems are about to revolutionize the way companies do business, from accounting to engineering to operations. AI may prove to be a transformative technology wave on the order of the internet or mobile (or, some even say, electricity). It might also prove to be something more like the metaverse that has lots of potential but will take time to mature.

A lot of money, and perhaps the health of the U.S. economy, is riding on the answer to that question.