OpenAI and Broadcom announce partnership to design and deploy 10 GW of custom AI accelerators by 2029

OpenAI/OpenAI-Broadcom Custom Accelerators

OpenAI and Broadcom announced a partnership to design and deploy custom AI accelerators optimized for OpenAI models. OpenAI will handle the chip design while Broadcom manages manufacturing, networking, and system integration. The collaboration aims to reduce OpenAI's reliance on off-the-shelf chips, with first systems expected in H2 2026 and full deployment by 2029.

• 10 GW total capacity of custom accelerators optimized for OpenAI models
• OpenAI designs silicon; Broadcom handles manufacturing and integration
• First systems expected H2 2026, full deployment by 2029
• Integrates with Broadcom's Ethernet, PCIe, and optical connectivity solutions; quintuples current OpenAI compute