OpenAI's Jalapeño Chip Challenges Nvidia, Optimizes Energy Use

OpenAI has unveiled its proprietary AI chip, codenamed Jalapeño, asserting that it surpasses Nvidia's processors in both performance and energy efficiency. This development signals OpenAI's strategic move to reduce its substantial computing costs and lessen its reliance on third-party chip manufacturers like Nvidia.

According to the company's statement, Jalapeño demonstrates superior speed and energy efficiency in executing various open-source models compared to benchmark Nvidia systems. The chip is specifically engineered for "inference," the process of using a trained AI model to generate responses, rather than for training new models. Consequently, OpenAI will continue to require accelerators from Nvidia and other suppliers to meet its escalating computational demands.

OpenAI reports that Jalapeño offers 1.5 to 1.9 times more AI work per watt and between 1.7 and 3.6 times lower end-to-end latency. The company intends to deploy a limited quantity of Jalapeño-equipped systems by the end of this year, with plans to significantly scale up capacity throughout 2027. Furthermore, OpenAI is already developing a second generation of the chip and has initiated work on a third.

Developing in-house processors has become a prevalent strategy among major AI companies, including Google with its TPUs, and Amazon and Microsoft, to manage costs and secure necessary computing power. Anthropic is also reportedly working on its own processor design team. OpenAI, however, does not plan to sell Jalapeño to other companies, stating that it needs all internally produced capacity to meet the growing demand for its services.