How OpenAI Built the Jalapeño Custom Inference Chip [Model Behavior]
I'm Nina Park. Welcome to Model Behavior. Today is June 24th, 2026. We are examining a significant shift in the AI infrastructure landscape as OpenAI transitions from being primarily a software and research organization to a custom hardware designer. On this program, Model Behavior looks at how complex AI systems are built, deployed, and operated within professional environments. We focus on the engineering realities behind the headlines to understand the true impact on the industry. <br/><i>acting_description:</i> Professional, steady, leading <i>speed:</i> 1.0 <i>trailing_silence:</i> 0.3 I'm Thatcher Collins. Nina, the major news this morning is the official unveiling of Jalapeño. This is the first home-grown silicon chip from OpenAI, and it is designed specifically for inference tasks. They are positioning this as an Intelligence Processor, a piece of hardware optimized to handle the heavy volume of day-to-day user queries rather than the massive computational load required to train new models. According to reporting from The Next Web, OpenAI has already moved past the prototype stage and is currently running laboratory samples for a specialized model identified as GPT-five.three-Codex-Spark. <br/><i>acting_description:</i> Engaged, responsive, grounded <i>speed:</i> 0.98 <i>trailing_silence:</i> 0.4 The broader hardware strategy appears to rely heavily on strategic collaboration. To bring Jalapeño to market, OpenAI partnered with Broadcom for the essential connectivity and networking components, and with Celestica for the production of the boards and server racks. Thatcher, what stands out is that OpenAI is not attempting to build every component from scratch. They focused on the core logic and architecture, but they are leaning on the Broadcom Tomahawk switching silicon to ensure the entire system can function at a massive data center scale. <br/><i>acting_description:</i> Clear, authoritative, confident <i>speed:</i> 1.0 <i>trailing_silence:</i> 0.3 The timeline is what really catches my attention, Nina. They claim to have moved from the initial design phase to a final manufacturing tape-out in only nine months. In the semiconductor world, that is an incredibly aggressive development cycle. They are attributing this accelerated pace to a self-reinforcing loop where their own existing AI models assisted engineers in the design and verification process. If AI can genuinely help human engineers design better chips faster, it significantly lowers the capital barrier to entry for custom silicon across the entire technology sector. <br/><i>acting_description:</i> Questioning, sharp, inquisitive <i>speed:</i> 0.96 <i>trailing_silence:</i> 0.4 We should also consider the financial and strategic motives behind this move. The hardware lead at OpenAI, Richard Ho, suggests this initiative provides them with full stack control. Essentially, they want to reduce their position as a captive customer of Nvidia. By owning the model architecture, the software layers, and now the physical chip underneath, they can tune the entire stack for maximum energy efficiency. Early internal testing suggests that the performance per watt is substantially higher than current state-of-the-art GPUs, although a full technical report is still several months away. <br/><i>acting_description:</i> Measured, steady, analytical <i>speed:</i> 1.0 <i>trailing_silence:</i> 0.3 It is important to maintain some skepticism, Nina. Vendor-provided benchmarks at a product launch always deserve a measured look. While Jalapeño might be highly efficient for running models, the training of those models remains the domain of Nvidia hardware. OpenAI has been transparent about the fact that Nvidia remains a critical partner for training their largest, most complex systems. This is not a total divorce from the existing supply chain