OpenAI reportedly to mass-produce custom AI chips next year
According to the Financial Times, OpenAI is preparing to mass-produce its own AI chips (so-called “XPUs”) next year in partnership with Broadcom. The move is aimed at providing the massive compute needed for models like GPT and reducing dependence on NVIDIA. Broadcom’s CEO recently announced a large, unnamed client placed about $10 billion in orders — the Times’ sources say that client is OpenAI. OpenAI reportedly plans to use the chips internally rather than sell them.
Background
- Reuters reported in 2023 that OpenAI was exploring building its own chips after GPU shortages affected the API’s speed and reliability.
- OpenAI has reportedly worked with Broadcom and TSMC on chip development; the Times did not confirm whether the TSMC partnership remains active.
- After GPT-5’s release, CEO Sam Altman said OpenAI would prioritize paid users and double its compute fleet over 5 months to meet growing demand.
Why it matters
Building custom chips could help OpenAI avoid GPU shortages, lower costs, and potentially change the competitive landscape currently dominated by NVIDIA — which recently reported a 56% year-over-year revenue increase for Q2.
Sources
What do you think this means for NVIDIA and the wider AI hardware market? Share your thoughts below.