OpenAI and Broadcom announce strategic collaboration to deploy 10 gigawatts of OpenAI-designed AI accelerators

Source: OpenAI Date: 2025-10-13 URL: https://openai.com/index/openai-and-broadcom-announce-strategic-collaboration

Summary

OpenAI and Broadcom announced a strategic collaboration in October 2025 to deploy 10 gigawatts of OpenAI-designed custom AI accelerators — chips designed by OpenAI and fabricated through Broadcom’s ASIC and networking ecosystem. This represented OpenAI’s entry into custom silicon design, following Google’s TPU model and Meta’s MTIA approach. Broadcom, as the dominant custom ASIC networking and chip design partner for hyperscalers, was the logical manufacturing and ecosystem partner for OpenAI’s custom chip ambitions.

Implications

OpenAI as a silicon company. Designing custom AI accelerators is a multi-year, multi-billion dollar commitment that reflects OpenAI’s intent to control its own hardware destiny rather than remain indefinitely dependent on Nvidia pricing and allocation. The 10GW figure is aspirational — it sets a target scale for the custom silicon program over the 2026-2030 window.

Thread: Stargate infrastructure and hardware strategy. Pairs directly with the AMD partnership (same month) as part of a multi-vendor compute strategy. Custom silicon (Broadcom), commodity GPU (AMD), and existing Nvidia capacity create three distinct supply chains. This mirrors Amazon (Trainium/Inferentia), Google (TPU), and Meta (MTIA) patterns.

Watch: How far along OpenAI’s custom chip design was at announcement — early tapeout, volume production, or still in architecture definition — and whether the Broadcom partnership delivered functional silicon within the 2026 window.