According to Reuters, OpenAI has chosen Broadcom to build its first proprietary AI processor, planning the rollout of 10 gigawatts (~ the computing equivalent to powering millions of homes) by the second half of 2026. Under this agreement, Broadcom will handle development and deployment, integrating with OpenAI’s design specifications.

This expands on OpenAI’s existing multi-supplier strategy: earlier deals with AMD and Nvidia secured compute capacity and hardware supply. The new partnership with Broadcom is intended to give OpenAI more control over the hardware stack and reduce dependency on external vendors.

Market reaction was strong, Broadcom shares jumped over 10% on the announcement. Analysts note that while many face high obstacles in chip design and scaling, OpenAI’s access to capital and existing compute demand gives it a better shot than typical startups.

Key Highlights

• OpenAI partners with Broadcom to build its custom AI processor.

• Target deployment: 10 gigawatts of compute capacity by 2026.

• Broadcom to handle design & manufacture aligned with OpenAI’s specs.

• Broadcom stock surged ~10% following the news.

• Move strengthens OpenAI’s strategy to control more of its hardware infrastructure.

Why This Matters

1. Vertical integration push: Custom hardware can yield performance, cost, and differentiation advantages.

2. Reducing supply risk: By diversifying its chip partnerships, OpenAI gains more resilience in a constrained hardware market.

3. High stakes in scaling: Deploying 10 GW level compute requires massive infrastructure, power, and design capability.

4. Competitive signal: This elevates the arms race among AI firms—not just over models, but over who controls the compute substrate.

Source
Reuters – Full Article