OpenAI Unveils Custom AI Chip 'Jalapeño,' Claiming Industry-Leading Efficiency Across Major Model Families

OpenAI unveils its first custom AI inference chip, 'Jalapeño,' claiming industry-leading throughput per kilowatt and lower token latency than competing systems across major model families, while simultaneously expanding its data center infrastructure and driving down AI costs through compounding efficiency gains in hardware, software, and model optimization.

• OpenAI reveals its first custom inference chip, Jalapeño, which delivers industry-leading throughput per kilowatt and lower token latency than competing commercial systems across multiple model families including GPT-OSS 120B, DeepSeek R1, and Kimi K2.
• OpenAI is actively managing a broad compute portfolio spanning Microsoft, NVIDIA, AWS, AMD, Broadcom, Cerebras, CoreWeave, Oracle, SB Energy, and SoftBank to optimize capability and cost across different AI workloads, while also developing its own data centers like Project Camellia in Georgia.
• Efficiency gains across models, hardware, and software are driving down costs and expanding what AI can do economically, with GPT-5.6 Sol achieving top coding benchmark scores while using 54% fewer output tokens, reinforcing a compounding advantage where better technology funds further progress.