OpenAI has entered the custom semiconductor race with the launch of Jalapeño, its first proprietary AI inference processor developed in partnership with Broadcom. The announcement marks a strategic shift for the ChatGPT maker as it seeks greater control over the computing infrastructure powering its artificial intelligence models while reducing long-term dependence on Nvidia's graphics processors.
Designed specifically for AI inference rather than model training, Jalapeño will handle the process of serving responses from large language models such as ChatGPT and Codex. OpenAI said the processor has been built from the ground up to optimise inference workloads, focusing on higher performance per watt, lower latency and improved operational efficiency across hyperscale data centres.
The launch reflects a broader trend across the technology industry, where leading AI companies are increasingly investing in custom silicon to optimise performance, lower infrastructure costs and gain greater independence from external hardware suppliers. Companies including Google, Amazon, Microsoft and Meta have already developed their own AI processors, while Nvidia continues to dominate the AI accelerator market.
According to OpenAI, Jalapeño is its first Intelligence Processor and represents the beginning of a multi-generation hardware roadmap rather than a one-time product. The company plans to deploy the chip across its AI infrastructure during the second half of 2026, complementing rather than replacing its continued use of Nvidia and AMD processors.
Broadcom played a central role in designing the application-specific integrated circuit, while infrastructure specialist Celestica contributed to the supporting rack systems required for deployment. The processor has been engineered specifically around OpenAI's understanding of large language model behaviour, enabling improvements in memory efficiency, networking and overall inference performance.
Unlike conventional graphics processors that support both AI training and inference, Jalapeño has been purpose-built for inference workloads. As AI adoption accelerates globally, inference is expected to account for an increasing share of computing demand as millions of users interact with deployed models every day. Specialised inference chips can improve efficiency while significantly lowering operating costs for AI providers.
OpenAI also revealed that its own AI models assisted engineers during parts of the chip development process, helping accelerate aspects of the design cycle. The processor reportedly reached tape-out in approximately nine months, considerably faster than traditional semiconductor development timelines.
The announcement comes as demand for AI computing infrastructure continues to surge worldwide. Large language models require enormous processing power, making access to high-performance chips one of the industry's biggest competitive advantages. Building proprietary silicon allows companies to tailor hardware to their own workloads while reducing exposure to supply constraints and pricing pressures.
Industry analysts view custom AI chips as an increasingly important strategic asset rather than simply a cost-saving measure. Control over hardware, networking and software enables AI companies to optimise the full technology stack while accelerating innovation across future products and services.
For OpenAI, Jalapeño represents more than a new processor. It signals the company's ambition to become a full-stack AI infrastructure player spanning models, software and hardware. As competition intensifies across the AI ecosystem, ownership of computing infrastructure is emerging as a key differentiator alongside advances in foundation models.
The launch also illustrates how AI innovation is moving beyond software into semiconductor design. As enterprises invest more heavily in generative AI, demand for specialised processors is expected to grow, making custom silicon an increasingly important part of the next phase of artificial intelligence development.