OpenAI has unveiled new performance results for its custom artificial intelligence chip, Jalapeño, marking a significant step in the company’s push to develop its own computing hardware. The chip is designed specifically for AI inference, the stage in which trained AI models generate responses and perform tasks. OpenAI says early benchmark results show Jalapeño can deliver higher efficiency and lower latency than several Nvidia systems.
Jalapeño Shows Strong Results Against Nvidia
According to OpenAI, Jalapeño demonstrated strong performance in tests involving several large language models. The company reported higher work-per-watt figures and lower response latency compared with Nvidia’s high-end GB200 and GB300 systems. Testing reported by OpenAI showed Jalapeño delivering between 1.5 and 1.9 times more AI work per watt in different workloads, while latency improvements ranged from about 1.7 to 3.6 times depending on the model and system being compared. These results indicate that OpenAI is targeting one of the biggest challenges facing AI companies: providing increasingly sophisticated AI services while controlling the energy and computing resources required to run them.
Sam Altman Calls the Chip Fast
OpenAI CEO Sam Altman highlighted the development publicly, describing the company’s achievement in simple terms as a fast chip. Indian reports have also highlighted OpenAI’s claim that the new hardware can provide substantially greater performance while consuming less power. The announcement comes as demand for AI computing continues to increase, putting pressure on companies to secure large amounts of advanced hardware.
Built With Broadcom
Jalapeño was developed by OpenAI in collaboration with semiconductor company Broadcom. OpenAI previously announced the custom inference processor as part of its strategy to build more of the technology stack supporting its AI models and products. Unlike general-purpose accelerators, Jalapeño has been designed around the requirements of AI inference. This specialized approach allows OpenAI to optimize the chip, memory and networking system for the workloads generated by its models.
What the Chip Means for Nvidia
The development could increase competitive pressure on Nvidia, which currently plays a central role in supplying AI accelerators to technology companies. However, Jalapeño is not an immediate replacement for Nvidia hardware. OpenAI is expected to continue using Nvidia accelerators alongside its own chips because its computing requirements are enormous and extend beyond inference workloads. The early results nevertheless demonstrate why major AI companies are increasingly investing in custom silicon. Specialized chips can potentially improve efficiency, reduce operating costs and give AI companies greater control over their infrastructure.
Jalapeño Deployment Planned for Later This Year
OpenAI plans to begin deploying Jalapeño within its own computing infrastructure by the end of 2026. The company is already working on a second-generation processor, while development of a third generation has also begun. The company developed Jalapeño in collaboration with Broadcom and says artificial intelligence itself played a role in the chip’s development and optimization. OpenAI described the processor as the beginning of a longer-term, multi-generation hardware roadmap.
The results give OpenAI a stronger position in the increasingly competitive AI-chip market. However, the real test will come when Jalapeño is deployed at scale and its performance and cost advantages are measured under large, real-world workloads.