OpenAI's First In-House Chip Jalapeño Debuts: 1.7x Per-Watt AI Throughput vs GB300 on DeepSeek R1
Summary
On August 25, OpenAI published a blog post showcasing the performance of its first in-house inference chip, Jalapeño, claiming superior per-watt throughput and latency compared to Nvidia's GB200 and GB300 systems. Developed with Broadcom and manufactured on TSMC's 3nm process, the chip is designed solely for inference and for internal use, aiming to cut GPU procurement costs. Tests show 1.7x per-watt performance over GB300 on DeepSeek R1, though no comparison with Nvidia's latest Vera Rubin was made.
Bubble analysis
This news does not directly address the investment bubble, but OpenAI's in-house chip development to reduce reliance on Nvidia could shift competitive dynamics and capital expenditure flows in AI hardware, indirectly relevant to bubble discussions.
Related signals
linked by shared tags
- OpenAI and Firmus Sign Multi-Year Deal to Purchase Compute from Two Malaysian AI Factories#openai
- OpenAI GPT-6 Astra Model Autonomously Completes 3D Puzzle Game Portal: 24 Hours, Cost $571#openai
- Futures Drop As Iran Hostilities Send Brent To 6 Week High Above $97#openai
- OpenAI Researchers Burn $7,000 a Day on AI Coding, Cost Pressures Spark Industry Reflection#openai
- Huawei Confirms! New Kirin 9050 Pro Announced as First Logic-Folding τ Chip#chip