OpenAI and Broadcom Unveil First Joint AI Chip, Jalapeño
The new accelerator, designed specifically for large language model inference, promises superior performance per watt with initial data centre deployment scheduled for late 2026.
OpenAI and Broadcom have unveiled the design for Jalapeño, their first jointly developed AI accelerator, marking a significant milestone in the tech giant’s transition from software developer to hardware manufacturer. The chip is engineered specifically for large language model (LLM) inference, the process of running pre-built AI models to generate responses. This development represents OpenAI’s initial foray into manufacturing the custom silicon it relies on to train and operate its ChatGPT models.
The collaboration, which was formally announced in October 2025, has moved at a rapid pace. OpenAI stated that Jalapeño was co-developed from initial design to manufacturing tape-out in just nine months. The company attributed this speed to deep software-hardware co-development between its engineering teams and Broadcom’s silicon implementation expertise, noting that OpenAI utilised its own AI models to accelerate parts of the design and optimization process.
In a blog post detailing the launch, OpenAI described Jalapeño as its “first Intelligence Processor,” architected around its vision for the future of LLM inference. The company claims the chip offers “performance per watt substantially better than current state-of-the-art” technology. However, OpenAI hedged these early assertions, confirming that final performance testing is still underway. A more thorough technical report detailing the chip’s performance specifications is expected to be released in the coming months.
The strategic move to build custom silicon aligns with a broader industry trend, with competitors such as Google and Amazon also developing custom “AI accelerators” to reduce their dependence on Nvidia GPUs. By controlling more components across the LLM process, OpenAI aims to increase business efficiency and expand the effective use of its ChatGPT tools. The partnership intends to establish a “multi-generation compute platform” designed to make advanced AI faster, more reliable, and more accessible.
Initial deployment of the Jalapeño chip in data centres is scheduled for late 2026. While the exact magnitude of its superior performance per watt has not yet been independently verified or quantified, the launch signals a decisive step for OpenAI in securing its computational infrastructure and reducing operating costs associated with third-party hardware.
