OpenAI announced its first custom-designed AI inference chip, Jalapeño, developed alongside Broadcom to optimize the performance and cost of running its real-time AI models.
- Custom AI chip developed jointly by OpenAI and Broadcom
- Focused on improving inference workload efficiency
- Early tests show superior performance per watt versus current GPUs
What happened
OpenAI officially revealed Jalapeño, its inaugural custom-built processor made in partnership with Broadcom. This chip was engineered specifically for inference workloads, the stage where pre-trained AI models respond to user requests. The collaboration was publicly confirmed in October 2025, although speculation about OpenAI designing its own silicon had circulated for some time.
OpenAI reported that early evaluations of Jalapeño demonstrate notably better performance per watt than existing state-of-the-art hardware solutions, especially in real-time coding model applications. The chip’s development was assisted by OpenAI’s own AI models, underscoring the integration of AI expertise in both software and hardware innovation.
Why it matters
This launch represents OpenAI’s deeper vertical integration, addressing key infrastructure components beyond AI model design. Historically reliant on Nvidia GPUs for heavy compute tasks, OpenAI’s move towards custom silicon can reduce costs and increase speed at the inference level—a critical factor for scaling AI services economically.
Similar efforts by tech giants like Google and Amazon have demonstrated the value of specialized AI accelerators in speeding up machine learning workloads. OpenAI’s focus on inference aligns with broader industry trends emphasizing operational efficiency, given that pre-training models still typically require high-powered GPUs. Optimizing inference cost structures could substantially impact AI business models going forward.
What to watch next
The ongoing testing phase will reveal how Jalapeño performs under large-scale production conditions, including its deployment efficiency and compatibility with OpenAI’s expanding suite of AI-powered products like Codex. Observers will also be looking at how cost savings from this chip affect OpenAI’s overall competitive positioning in the cloud AI market.
In the longer term, OpenAI’s chip development signals a potential trend of AI companies designing tailored hardware to control the entire AI stack—from silicon to software to product delivery. Industry watchers should monitor whether OpenAI scales up this hardware initiative, possibly extending to training accelerators or new infrastructure layers.