OpenAI picks Broadcom to build its first in‑house AI chips in new custom-silicon pact

OpenAI struck a deal with Broadcom to manufacture its first in-house AI processors, deepening the ChatGPT maker’s drive to secure scarce computing power. OpenAI will design the chips while Broadcom develops and deploys them starting in the second half of 2026, with a rollout totaling roughly 10 gigawatts of capacity and completion targeted by 2029. Broadcom shares jumped more than 7% on the news, underscoring investor enthusiasm for custom silicon that could ease dependence on Nvidia’s pricey, supply-constrained accelerators. The pact follows OpenAI’s recent supply agreements with AMD and a planned Nvidia investment package, as the startup pursues a multi-pronged strategy to meet surging AI demand. Financial terms weren’t disclosed, and how OpenAI will fund the program remains unclear. While custom chips have gained traction among Big Tech peers like Google and Amazon, similar efforts at Microsoft and Meta have faced delays, and analysts say Nvidia’s dominance is unlikely to be challenged in the short term. Broadcom’s Ethernet-centric buildout also positions it to challenge Nvidia’s InfiniBand in AI networking.

Read more