On August 25, OpenAI unveiled Jalapeno, its first custom AI chip. It delivers up to 13.4 petaflops of 4-bit compute and 232 gigabytes of memory linked at 15.4 terabytes per second. OpenAI says the chip cuts latency versus Nvidia's GB300 by up to 3.6 times.
OpenAI fully unveiled its first custom AI accelerator, called Jalapeno, on August 25. The chip delivers up to 13.4 petaflops of 4-bit compute and accesses 232 gigabytes of memory linked at 15.4 terabytes per second. According to OpenAI, Jalapeno cuts end-to-end latency by up to 3.6 times compared with Nvidia's GB300, the chip the company currently relies on, while drawing less power. The chip was built with Broadcom as manufacturing partner, with OpenAI handling system design and Broadcom the physical implementation. Jalapeno moved from first architecture concept to first silicon in under 20 months, with only nine months between the first logic code and manufacturing handoff.
What stands out is the path there: OpenAI used its own large language models to speed up the design process itself. Hardware vice president Richard Ho says the models give engineers extra capability without taking over final decisions. The small team, averaging fewer than 100 people, could explore far more design paths in parallel than typical development cycles allow. For companies building chips or software, the case shows how generative AI is reaching into highly complex engineering work and can shorten development timelines. For OpenAI itself, the custom chip also reduces reliance on Nvidia as sole supplier of compute power.
It remains open whether the lab-measured gains in latency and power draw will hold up once Jalapeno enters full production use in OpenAI's inference fleet. It is also unclear how fast the broader trend toward AI-assisted chip design will accelerate in coming years, and how other chipmakers will respond. What is clear so far is that Jalapeno's development time was short, and OpenAI is already working on second and third generation designs.
What this means for decision-makers
- Track how AI-assisted chip design affects development timelines across the industry.
- Assess supplier risk if major customers begin using custom chips instead of Nvidia GPUs.
- Follow benchmark results for Jalapeno once the chip enters full production use at OpenAI.
This story was produced automatically from the source named above and checked by software before publication. The image is symbolic and shows neither the event nor a real person. How this paper is made