OpenAI is claiming an early performance edge for Jalapeño, its new AI inference chip, in benchmark results reported by The Verge. According to the report, OpenAI said in a Tuesday blog post that Jalapeño can complete AI tasks more efficiently and return responses faster than other AI systems. The company’s claim is specific: on InferenceX, a benchmarking platform for AI inference systems, OpenAI compared Jalapeño with the best results recorded at the time, which The Verge says used Nvidia GB200 or GB300 superchips. OpenAI said Jalapeño produced 1.5 to 1.9 times more AI work per watt across GPT-OSS 120B, DeepSeek R1, and Kimi K2.5 1T than the comparison systems. OpenAI also claimed a latency advantage. The Verge reports that OpenAI said Jalapeño delivered 1.7 to 3.6 times lower end-to-end latency across the same three models. Richard Ho, OpenAI’s hardware vice president, told reporters that Jalapeño offers lower latency and higher throughput together, rather than forcing the usual trade-off between the two. Jalapeño is an application-specific integrated circuit, or ASIC, built in partnership with Broadcom, according to The Verge. OpenAI first introduced the chip in June. Its target workload is inference: running trained models to produce responses, complete tasks, or operate agents. The deployment plan remains limited for now. Ho said OpenAI expects to put Jalapeño into use in “small volumes” by the end of this year, then increase volume in 2027, The Verge reports. OpenAI did not disclose how many chips it expects to deploy next year. OpenAI is also not presenting Jalapeño as a full replacement for its current compute supply chain. Ho said the company’s broader compute strategy still includes partners such as Nvidia, according to The Verge, even as OpenAI continues work on second- and third-generation versions of its own chip. The practical read is that OpenAI is trying to control more of the inference stack without severing its dependence on outside accelerators. The performance figures are OpenAI’s claims from a benchmark comparison, not independently corroborated in this cluster. But the direction is clear: OpenAI is investing in custom silicon for the workloads that matter most once models are trained and used at scale. Who benefits: OpenAI would benefit most if Jalapeño improves inference efficiency in production. Broadcom also benefits from being named as the chip partner. Who's exposed: Nvidia is the comparison point in OpenAI’s benchmark claims, but OpenAI also says it will continue relying on partners such as Nvidia. The exposure is therefore partial, not a stated replacement of Nvidia systems.