October 13, 2025 — OpenAI and Broadcom announced a strategic collaboration to deploy up to 10 gigawatts of OpenAI-designed AI accelerators, with racks expected to begin deployment in the second half of 2026.
What happened
OpenAI will design the accelerators and networking systems, developed and manufactured in partnership with Broadcom, using Broadcom Ethernet connectivity rather than proprietary interconnects. The companies said they had been working together for 18 months before going public with the plan.
Why it matters
The deal is one of roughly 33 gigawatts of compute commitments OpenAI announced across partnerships with Nvidia, Oracle, AMD and Broadcom in a three-week span in late 2025, illustrating the scale of custom-silicon investment now underway among frontier AI labs seeking to control more of their own hardware roadmap.
Security and infrastructure impact
Custom AI silicon programs of this scale add new vendors and supply chains for security teams to vet, particularly around firmware integrity and hardware supply-chain assurance for chips destined for critical AI infrastructure.
Leave a Reply