OpenAI unveils first custom AI chip, ‘Jalapeno’

OpenAI has introduced its first custom-designed AI chip, Jalapeno, developed with Broadcom to enhance inference performance, improve compute efficiency, and support the growing demand for advanced AI models powering ChatGPT and future applications.

OpenAI has taken a significant step toward strengthening its artificial intelligence infrastructure with the launch of its first custom AI chip, named Jalapeno. Developed in collaboration with Broadcom, the new processor is designed specifically for AI inference workloads and reflects OpenAI’s growing focus on building key elements of its technology stack in-house.

The announcement marks OpenAI’s formal entry into the AI silicon segment, an area that has become increasingly important as demand for generative AI services continues to rise. The Jalapeno chips will be manufactured by Broadcom and deployed to support inference operations—the process through which trained AI models generate responses and interact with users across platforms such as ChatGPT and other AI-powered services.

According to OpenAI, the chip has been engineered from the ground up with modern large language models (LLMs) in mind. Unlike many accelerators adapted from existing computing architectures, Jalapeno was created specifically to address the requirements of large-scale AI inference. The design draws on insights gained from OpenAI’s experience developing and operating products including ChatGPT, Codex, APIs, and future AI agents.

Built for large-scale AI deployments

OpenAI said its development teams worked closely with technology partners Broadcom and Celestica to bring the platform from concept to production. The collaboration covered chip implementation, networking architecture, system integration, rack-level deployment, and manufacturing readiness.

The company noted that engineering samples of Jalapeno are already running machine-learning workloads in testing environments while meeting targeted performance and power-efficiency benchmarks. OpenAI added that the chip is capable of supporting both current and upcoming generations of AI models, including workloads associated with GPT-5.3 and Codex Spark.

OpenAI President Greg Brockman emphasized the strategic value of developing proprietary hardware, stating that greater control over the technology stack can help the company deliver AI capabilities more efficiently and expand access to advanced intelligence.

Focus on performance and cost efficiency

The new processor is intended to combine the high throughput associated with leading AI accelerators with the low-latency performance required for interactive AI applications. This balance is expected to be particularly valuable as organizations increasingly deploy conversational AI and agent-based systems that require rapid response times at scale.

Industry observers view the launch as part of a broader trend among AI developers seeking greater control over computing infrastructure amid rising demand for specialized hardware. For OpenAI, Jalapeno represents an effort to optimize costs, improve performance, and reduce dependence on third-party processors while supporting the next generation of AI services.

The company said efficiency gains achieved through the new chip will be reinvested into future infrastructure and model development, helping accelerate the delivery of more capable and accessible AI systems.