OpenAI Launches Custom AI Chip 'Jalapeño' That Outperforms Nvidia's Blackwell With Up To 4x Better Efficiency
Summary
OpenAI unveils its first custom AI inference chip, 'Jalapeño,' built with Broadcom, delivering up to 4x better performance-per-watt than Nvidia's Blackwell across key benchmarks, with plans to scale production into 2027 to cut compute costs and broaden AI access.
Key Points
- OpenAI unveils its first custom AI inference chip, Jalapeño, developed in partnership with Broadcom, which is already running on working silicon and undergoing testing after being conceived in late 2024.
- Jalapeño delivers 1.5 to 4 times the performance-per-watt compared to leading AI chips, outperforming Nvidia's Blackwell across nearly all benchmark scenarios on both low-latency and high-throughput workloads.
- OpenAI credits its own AI models for dramatically accelerating Jalapeño's development and plans to scale up chip volume into 2027, with the long-term goal of reducing compute costs and making advanced AI more accessible.