OpenAI unveiled 'Jalapeño', its first AI chip to compete with Nvidia

OpenAI announced the launch of Jalapeño, its first artificial intelligence chip designed to enhance the performance of models like ChatGPT and Codex.

The new processor was developed alongside Broadcom and aims to optimize the company's technological infrastructure, improve efficiency, and accelerate AI inference processes.

What is Jalapeño, OpenAI's new artificial intelligence chip

Jalapeño is the first processor from OpenAI developed specifically for inference tasks, that is, to execute and respond to queries made to artificial intelligence models.

The company defines it as an accelerator designed to support the evolution of large language models (LLMs) and to be part of a long-term computing platform focused on advanced AI.

According to OpenAI, the chip was created from scratch based on the processing needs of tools like ChatGPT and Codex.

How the Jalapeño chip was developed

The project took nine months of work, from the initial design stage to the start of production.

Richard Ho, director of OpenAI's Hardware Program, explained that the architecture was optimized around processing cores, memory, networks, and the service patterns used by the most advanced AI models.

According to initial tests, Jalapeño will efficiently handle the company's most important workloads.

The alliance between OpenAI and Broadcom to create AI infrastructure

Broadcom and Celestica participated in the manufacturing and integration of the complete platform, including servers, electronic boards, high-speed networks, and the infrastructure necessary to operate thousands of processors simultaneously.

The companies claim that the system was designed to run both current and future OpenAI models as well as solutions developed by third parties.

Hock Tan, president and CEO of Broadcom, stated that the project represents the beginning of a multigenerational technological roadmap.

What performance does the Jalapeño chip offer

Currently, OpenAI is evaluating Jalapeño's performance using real machine learning workloads on systems like GPT-5.3-Codex-Spark.

Preliminary results show that the processor achieves the expected levels of performance and energy efficiency for its commercial version.

Additionally, the company claims it will offer superior performance per watt compared to the most advanced technologies currently available.

In statements to Reuters, Hock Tan noted that the new design achieves levels comparable to Nvidia's Blackwell chips and the tensor processing units developed by Google.

When will OpenAI's chip start being used

OpenAI indicated that the initial implementation of Jalapeño is expected by the end of this year.

The company hopes that the processor will help optimize response speed, reliability, and operational costs of AI-based services.