OpenAI secures dedicated chip supply through a $10B hardware agreement with Broadcom.
OpenAI signs a $10 billion agreement with Broadcom to develop custom artificial intelligence chips, aiming to ease ongoing GPU shortages that limit the rollout of new ChatGPT models.
Broadcom confirms the deal during its earnings call, citing a one-time order from a new AI customer. Sources identify the customer as OpenAI. The order, scheduled for delivery by mid-2026, includes AI server racks powered by Broadcom’s semiconductors.
In February 2025, OpenAI CEO Sam Altman posted on X that the deployment of GPT-4.5 is delayed due to limited access to GPUs. OpenAI primarily relies on Nvidia’s GPUs, which remain in high demand across the AI sector. As demand outpaces supply, compute infrastructure becomes a constraint for large-scale model training.
To address this, OpenAI partners with Broadcom to co-develop application-specific chips. Broadcom’s XPUs target high-performance training tasks and are intended to supplement, not replace, Nvidia GPUs.
The move aligns with OpenAI’s broader strategy to expand compute access. The company sets a long-term cloud agreement with Oracle reportedly worth over $30 billion per year, along with a smaller compute deal with Google. It is also building its own data centre project, Stargate, which faces delays.
Broadcom’s shares rose 11% following the announcement. The order is expected to materially impact the company’s 2026 revenue. Broadcom beats Q3 earnings expectations and projects 24% year-on-year revenue growth for the next quarter.
The company continues its pivot from mobile chips to AI infrastructure. Its Jericho networking chip, introduced in August, connects data centres across long distances to support distributed AI workloads. Broadcom’s valuation climbs significantly, placing it among the largest publicly listed firms globally.
