OpenAI is stepping deeper into AI hardware with its first custom chip, Jalapeño, developed with Broadcom. The company says the processor could improve AI inference speed, reduce latency and deliver more computing performance per watt.
OpenAI is expanding its ambitions beyond artificial intelligence models and into the hardware that powers them. The company has developed a custom AI chip called Jalapeño, marking a significant step towards gaining greater control over the infrastructure behind its AI services.
Developed in collaboration with Broadcom, the chip has been designed specifically for AI inference—the process through which trained models respond to user requests.
OpenAI says Jalapeño can handle more AI workloads while consuming less power and reducing response times. The company tested the chip using models including GPT-OSS 120B, DeepSeek R1 and Kimi K2.5 1T.
What Is OpenAI’s Jalapeño AI Chip?
Jalapeño is a custom processor built around the specific requirements of OpenAI’s AI models.
The company says the main objective was to combine high performance with low latency, allowing AI systems to process requests more efficiently without significantly increasing response times.
This is particularly important as AI services handle increasingly large numbers of queries. For systems processing millions of requests, improvements in energy efficiency and response speed can have a substantial impact on operating costs and overall performance.
Why Is Inference So Important?
The key focus of Jalapeño is AI inference.
In simple terms, inference happens when a trained AI model takes a user’s prompt or request and generates an output. Every time someone asks ChatGPT a question, requests an image or uses an AI-powered coding tool, the underlying model is performing inference.
Unlike chips primarily designed for training AI models, Jalapeño has been developed with the requirements of inference workloads in mind.
OpenAI worked with Broadcom to develop the processor, while Celestica contributed to the boards, rack systems and production hardware.
The company says its approach involved designing several parts of the infrastructure together rather than treating the chip as an isolated component.
“By co-designing the chip, software, memory, networking and serving systems around our models, we can improve performance and efficiency across the entire stack,” OpenAI said.
OpenAI Claims Major Performance Improvements
According to OpenAI’s testing, Jalapeño delivered 1.5 to 1.9 times more AI work per watt than the comparison systems across the three models tested.
The company also reported 1.7 to 3.6 times lower end-to-end latency.
For highly interactive workloads, OpenAI said the performance advantage could reach up to 4.1 times.
If these improvements translate effectively to large-scale deployments, the technology could allow OpenAI to handle more AI requests while consuming less electricity and delivering faster responses.
Is OpenAI Moving Away From Nvidia?
The development of Jalapeño highlights OpenAI’s effort to become less dependent on off-the-shelf AI hardware and gain greater control over its computing infrastructure.
However, the chip should not be viewed as an immediate replacement for Nvidia hardware.
OpenAI continues to work with multiple hardware partners and plans to use Jalapeño alongside other computing solutions.
The company expects to deploy the new chip in small volumes by the end of 2026, with broader deployment planned during 2027.
OpenAI Wants Control of the Entire AI Hardware Stack
The significance of Jalapeño goes beyond the chip itself. OpenAI is attempting to optimise the complete infrastructure used to operate its AI models.
That includes:
- AI processors
- Software
- Memory
- Networking
- Rack and serving systems
- Model-specific optimisation
By designing these components together, OpenAI believes it can improve efficiency and performance across the entire computing stack.
The company also said the chip progressed from its initial design to manufacturing tape-out in just nine months. OpenAI’s own AI models were reportedly used to assist engineers during parts of the design and optimisation process.
Jalapeño Won’t End OpenAI’s Nvidia Dependence Yet
Despite the ambitious performance claims, OpenAI’s strategy remains a gradual one.
Jalapeño will initially be deployed at relatively small scale, and Nvidia is expected to remain an important part of OpenAI’s broader computing infrastructure.
The company is also planning additional generations of its custom hardware, suggesting that Jalapeño is more likely to be the beginning of a long-term chip strategy than a one-off project.
What OpenAI’s Custom Chip Means for the AI Industry
The move into custom silicon could give OpenAI greater flexibility over how its AI models are operated and scaled. It could also help the company target specific workloads instead of relying entirely on general-purpose AI accelerators.
For Nvidia, the development represents another sign that major AI companies are increasingly interested in developing their own hardware.
Still, OpenAI’s chip ambitions are at an early stage. Jalapeño is expected to complement—not immediately replace—Nvidia and other hardware platforms.
If the company’s performance and efficiency claims hold up during large-scale deployment, however, Jalapeño could become an important part of OpenAI’s strategy to build a more vertically integrated AI infrastructure.