OpenAI Launches Jalapeño Chip as Nvidia Debuts Vera Rubin

OpenAI launched its first custom artificial intelligence processor, Jalapeño, to reduce its reliance on Nvidia for inference tasks. Developed in collaboration with Broadcom Inc, the chip uses HBM4 memory to execute user queries with high speed and operational efficiency. Independent research from SemiAnalysis indicates that Jalapeño outperforms Nvidia's Blackwell architecture in performance per watt, though analysts note that Nvidia's upcoming Rubin platform provides a more accurate comparison.

In response to the shift toward custom silicon by hyperscalers, Nvidia is expanding its strategy beyond GPUs to focus on data orchestration. The company is rolling out the Vera Rubin architecture, which integrates the Rubin GPU with the Vera CPU, the Groq 3 LPX inference accelerator, and specialized networking racks. The Vera CPU is designed to reduce bottlenecks by accelerating data movement to the GPU, with Nvidia reporting up to a 3x improvement in these operations.

OpenAI intends to deploy Jalapeño across its infrastructure by the end of the year and is already developing subsequent generations. This move aligns OpenAI with other tech giants like Google, Amazon, Microsoft, and Meta, who are deploying proprietary silicon to challenge Nvidia's dominance, although Nvidia remains the primary provider for large-scale model training.