OpenAI’s Jalapeño chip is built for fast inference at scale, benchmarks show
OpenAI unveiled its Jalapeño inference chip at the Hot Chips conference on August 25, reporting it delivers up to 1.9 times more performance per watt than Nvidia Blackwell systems. The custom hardware reached 85,448 tokens per second per kilowatt in initial tests. While the benchmarks show the chip outperforms Nvidia in speed and power efficiency, OpenAI stated it will continue purchasing hardware from Nvidia despite the development of its own semiconductors.
Listen to Live Briefing
Real-time synthesized voice briefing · Live Feeds Desk
- ✓ OpenAI presented the Jalapeño inference chip at the Hot Chips conference on August 25.
- ✓ The Jalapeño chip delivers up to 1.9 times more performance per watt than Nvidia Blackwell systems.
- ✓ Initial benchmark tests showed the chip reached 85,448 tokens per second per kilowatt.
- ✓ OpenAI developed the chip in partnership with Broadcom.
What changed
OpenAI published specific benchmark data showing the Jalapeño chip reaches 1.9x the efficiency per watt of Nvidia Blackwell.
Live updates
-
OpenAI Jalapeño chip hits 1.9x efficiency of Nvidia Blackwell in benchmarks
OpenAI unveiled its Jalapeño inference chip at the Hot Chips conference on August 25, reporting it delivers up to 1.9 times more performance per watt than Nvidia Blackwell systems. The custom hardware reached 85,448 tokens per second per kilowatt in initial tests. While the benchmarks show the chip outperforms Nvidia in speed and power efficiency, OpenAI stated it will continue purchasing hardware from Nvidia despite the development of its own semiconductors.
Why it matters
OpenAI partnered with Broadcom to create the Jalapeño chip to reduce reliance on Nvidia's market dominance. Custom silicon allows the company to optimize inference at scale, lowering costs and energy consumption for AI workloads.
What is confirmed
- OpenAI presented the Jalapeño inference chip at the Hot Chips conference on August 25.
- The Jalapeño chip delivers up to 1.9 times more performance per watt than Nvidia Blackwell systems.
- Initial benchmark tests showed the chip reached 85,448 tokens per second per kilowatt.
- OpenAI developed the chip in partnership with Broadcom.
Still unconfirmed
- OpenAI intends to continue buying chips from Nvidia.
What to watch next
- Release of third-party independent benchmarks for Jalapeño
- Details on the volume of Jalapeño chip production and deployment scale
confidence 95%Sources used for this update (5)
- 247wallst.com — OpenAI’s Custom Chip Embarrasses Nvidia, While Company Vows to Keep Buying From It
- www.techrepublic.com — OpenAI’s Jalapeño Benchmark Promises Faster, Cheaper AI
- www.martincid.com — OpenAI’s Jalapeño chip runs inference at 1.9x Nvidia Blackwell’s efficiency per watt
- quantumzeitgeist.com — OpenAI’s Jalapeño chip boosts AI speed and cuts power use
- www.tekedia.com — OpenAI’s Jalapeño Puts Nvidia’s AI Chip Dominance Under Pressure
-
OpenAI's Jalapeño chip outperforms Nvidia processors in AI inference benchmarks
OpenAI's custom Jalapeño chip, developed with Broadcom, has shown industry-leading speed and efficiency in AI inference benchmarks, outperforming Nvidia processors. The chip is designed for fast inference at scale. Early benchmark results indicate improved performance and power efficiency compared to existing chips. OpenAI claims the Jalapeño chip can outperform Nvidia's processors in tests.
Why it matters
The development of custom AI chips is crucial for companies like OpenAI that rely heavily on AI computations. By reducing reliance on third-party hardware, OpenAI aims to optimize performance and costs. The Jalapeño chip's performance could impact the AI hardware market, potentially challenging Nvidia's dominance.
What is confirmed
- OpenAI's Jalapeño chip is built for fast inference at scale.
- Jalapeño's first results show industry-leading speed and efficiency in AI inference.
- OpenAI claims its Jalapeño chip can outperform Nvidia processors in tests.
Still unconfirmed
- OpenAI Jalapeño chip is better than Nvidia Blackwell
What to watch next
- Late-2026 deployment of Jalapeño chips
- Further benchmark results and comparisons with other AI chips
- Impact on Nvidia's market share and AI hardware market dynamics
confidence 85%Sources used for this update (10)
- TechCrunch — OpenAI’s Jalapeño chip is built for fast inference at scale, benchmarks show
- OpenAI — Jalapeño’s first results show industry-leading speed and efficiency in AI inference
- SemiAnalysis — OpenAI Jalapeño: Better Than Nvidia Blackwell
- Yahoo Finance — OpenAI Claims Its New Chips Can Outperform Nvidia Processors in Tests
- CNBC — OpenAI says its Broadcom custom chip is a winner. What does that mean for Nvidia?
- CNBC — Tuesday's big stock stories: What’s likely to move the market in the next trading session
- Axios — OpenAI says its Jalapeño chip bests Nvidia, others
- techgenyz.com — OpenAI Jalapeño Chip Challenges NVIDIA With Faster AI Inference and Lower Power Use
- digitalmarketreports.com — OpenAI Says Jalapeño Outperforms Current Inference Chips in Early Benchmarks
- www.cxodigitalpulse.com — OpenAI Publishes First Jalapeño Chip Benchmarks Ahead of Late-2026 Deployment
Community Sentiment: How do you assess this situation?
Voice your perspective · Real-time aggregated sentiment from the Live Feeds community