OpenAI has launched Jalapeño, its inaugural in-house AI chip, developed entirely by its own engineering team with collaboration from Broadcom and Celestica. This new accelerator is specifically designed for inference tasks rather than for training artificial intelligence models. It has been optimized not just for existing OpenAI models but also for future neural networks that the company intends to develop. Broadcom's CEO, Hock Tan, asserts that Jalapeño competes effectively with processors from major players like Nvidia and Google. However, detailed specifications and independent evaluations against competitors' chips are not yet available, as the hardware is still undergoing testing within OpenAI's facilities. OpenAI aims to evolve into a vertically integrated entity, managing both its AI products and the underlying infrastructure. This strategy is expected to lead to quicker and more cost-effective services for end users, a feat that only Google has successfully accomplished thus far. The company anticipates that Jalapeño chips will be operational by the end of this year, with the current agreement between OpenAI and Broadcom indicating a goal of achieving 10 gigawatts of computing capacity. As for the chip's potential impact on the market, some industry observers believe Nvidia may need to be concerned, while others view the move as primarily a marketing strategy.
Informational material. 18+.
