OpenAI Deploys First Model on Non-Nvidia Hardware
In a significant shift in its infrastructure strategy, OpenAI has reportedly begun deploying a new AI model on hardware not powered by Nvidia chips. The move is seen as an effort to diversify its hardware supply chain and mitigate risks associated with relying on a single vendor. OpenAI is said to be collaborating with alternative chipmakers to address costs and supply constraints.
- The first OpenAI model to run on non-Nvidia hardware is GPT-5.3-Codex-Spark, a specialized version of its coding assistant, which is being deployed on chips from Cerebras Systems. This model is designed for high-throughput, low-latency tasks like real-time code editing and testing, and is being offered as a research preview to ChatGPT Pro subscribers. - This move is part of a broader strategy to diversify OpenAI's hardware suppliers, which includes a deal to deploy 6 gigawatts of AMD GPUs and an agreement to use Google's Tensor Processing Units (TPUs) to help lower the cost of inference. OpenAI has also partnered with Broadcom to develop custom AI accelerators and networking components. - While diversifying, OpenAI is also developing its own in-house AI chips. The company is reportedly finalizing its first design and plans to send it to Taiwan Semiconductor Manufacturing Co. (TSMC) for fabrication, with the goal of mass production in 2026. - The push for alternatives is driven by the high cost and supply constraints of high-end GPUs. A single Nvidia H100 GPU, a workhorse for AI training, can cost around $40,000, and the total cost to train a large language model can run into the hundreds of thousands or millions of dollars. - This diversification strategy isn't unique to OpenAI. Microsoft, a major OpenAI partner, is developing its own "Maia" line of AI accelerator chips. Similarly, Google has been developing its own Tensor Processing Units (TPUs) for nearly a decade, and Meta is also exploring the use of Google's TPUs. - Despite exploring alternatives, OpenAI has stated its partnership with Nvidia remains "foundational," with Nvidia's hardware continuing to be the core for training its most powerful models. CEO Sam Altman has publicly stated that Nvidia makes "the best AI chips in the world" and that OpenAI hopes to be a "gigantic customer for a very long time." - The focus on hardware extends to a grander vision from CEO Sam Altman, who is reportedly seeking to raise trillions of dollars to dramatically increase the world's chip manufacturing capacity to avoid future supply bottlenecks for advanced AI. He has stated that "infrastructure is destiny," believing that chip development and data center capacity will determine which countries lead in AI.
