The world of artificial intelligence is experiencing a significant technological shift. OpenAI has officially unveiled its own Jalapeño processor — a specialized solution designed to accelerate inference, that is, the direct execution of requests to neural networks. This event marks the company's transition from using third-party hardware to building its own hardware base, which will allow it to fully control the performance and cost of its services.

Secret Project and Strategic Partnership

The development of the microchip was carried out under strict secrecy. During the presentation, a symbolic silicon wafer with the new chip was handed over to OpenAI CEO Sam Altman and President Greg Brockman. This ceremony was conducted by leaders of the partner company Broadcom — Hock Tan and Charlie Kawwas.

Collaboration with Broadcom will serve as the foundation for massive changes in OpenAI's infrastructure. The official partnership, which involves the deployment of gigawatt-scale data centers jointly with Microsoft and other partners, is scheduled to launch by the end of 2026.

Architecture Born from Practice

It is important to note that Jalapeño is not just an adapted graphics processor. It is a platform built from scratch specifically for the inference tasks of modern large language models (LLMs). The chip's architecture is based on the daily experience of operating such giant services as ChatGPT, Codex, and commercial APIs.

The main advantage of the new product is the combination of high computing power with minimal latency, which is critical for interactive products. Thanks to the optimization of data transfer within the architecture, the actual utilization of Jalapeño's power has approached the theoretical peak, which was previously unattainable for standard solutions.

Record Development Speeds

The process of creating Jalapeño — from the first drawings to the final release for production (tape-out) — took only nine months. According to developers, this is one of the fastest development cycles for specialized microchips (ASICs) in the history of the industry.

Such speeds were achieved thanks to the deep synergy of OpenAI teams and Broadcom's expertise. A unique factor was the use of OpenAI's own active AI models to automate the design and optimize the chip's topology. The company has effectively closed the technological loop: neural networks serving users today help create hardware for future, even more powerful models.

Currently, engineering samples of the chip are already successfully undergoing testing in laboratories at target operating frequencies, running promising models, including GPT‑5.3‑Codex‑Spark.

Ecosystem and Democratization of AI

Jalapeño will be the first generation of a computing platform that will scale in the coming years. To implement this infrastructure, OpenAI has attracted key partners:

  • Broadcom is responsible for the realization of silicon wafers and high-performance networking solutions, including the Tomahawk network chip.
  • Celestica ensures the integration of boards, racks, and the creation of scalable production systems for server installations.

The main goal of the project is the democratization of access to artificial intelligence technologies. The transition to its own hardware base will reduce the cost of computations. This will directly affect the response speed of ChatGPT, allow Codex agents to perform more complex multi-step tasks without delays, and make commercial API products more accessible to developers, researchers, and small businesses.