A high-profile plan to invest up to $100 billion in OpenAI’s AI infrastructure was unveiled in September 2025 as a letter of intent, not a binding contract.
Five months later, no binding agreement has closed. Nvidia CEO Jensen Huang has publicly walked back the figure, saying the $100 billion was never a commitment and that Nvidia would invest incrementally.
Reuters reported that OpenAI has quietly explored alternatives to Nvidia chips since last year, citing sources who described concerns about inference speed on some Nvidia GPUs, particularly in Codex, OpenAI’s code-generation model.
OpenAI CEO Sam Altman attempted to ease investor concerns, praising Nvidia as a top partner and confirming a long-term role as a major customer.
The initial plan described 10 gigawatts of Nvidia systems for OpenAI, a scale Huang said would match Nvidia’s total yearly GPU shipments. Yet the LOI remained non-binding, and Huang emphasized that investment would proceed “one step at a time.”
OpenAI has since been linked with alternative hardware partners, including Cerebras and Groq, and has pursued broader compute partnerships with AMD and Broadcom as it hedges against dependence on Nvidia for future AI workloads.
