OpenAI Unveils First Custom AI Chip ‘Jalapeño’ in Bid to Reduce Dependence on Nvidia

OpenAI has unveiled its first custom-built artificial intelligence chip, marking a major milestone in the company’s efforts to strengthen its computing infrastructure and reduce its reliance on Nvidia’s highly sought-after AI processors.

The new chip, named Jalapeño, was developed in collaboration with semiconductor giant Broadcom and is specifically designed to handle AI inference—the process that allows artificial intelligence models to generate responses to user requests. In practical terms, inference is what powers interactions with AI tools such as ChatGPT, enabling them to process information and deliver answers quickly and efficiently.

The announcement highlights OpenAI’s growing ambition to control more of its technology stack as demand for advanced AI systems continues to surge. With AI companies racing to develop increasingly powerful models, access to high-performance computing resources has become one of the industry’s biggest challenges.

OpenAI and other leading AI firms, including Anthropic, have faced increasing pressure to secure enough computing power to support next-generation chatbots, coding assistants, and other AI applications. Building custom chips is seen as a strategic way to lower long-term costs, improve performance, and reduce dependence on third-party suppliers.

According to Broadcom CEO Hock Tan, the Jalapeño chip is capable of delivering performance comparable to Nvidia’s latest Blackwell processors and Google’s Tensor Processing Units (TPUs), both of which are widely used in artificial intelligence workloads.

OpenAI’s Head of Hardware, Richard Ho, said the processor was specifically engineered to work efficiently with large language models (LLMs), the technology that powers many of today’s most advanced AI applications. He expressed confidence that the chip would remain effective for future generations of AI models as the technology continues to evolve.

The company plans to begin deploying Jalapeño by the end of the year, describing it as the first step in a broader multi-generation chip development strategy. OpenAI revealed that the processor is already being tested in its laboratories and has successfully achieved its targeted power and performance benchmarks while running the company’s GPT-5.3-Codex-Spark AI model.

In another key development, Canadian electronics manufacturer Celestica has been selected to build the server systems that will house the new chips. Both the processors and the servers will be used exclusively by OpenAI to support its growing AI operations.

OpenAI also disclosed that its engineering team completed the chip’s design in approximately nine months before sending it to Taiwan Semiconductor Manufacturing Company (TSMC) for production. The company noted that artificial intelligence itself was used to accelerate certain aspects of the chip design process, helping to shorten development timelines.

The move reflects a broader trend within the technology industry, where major players such as Meta, Amazon, and Google have increasingly invested in custom chip development to gain greater control over AI infrastructure. Reports have also suggested that Anthropic is exploring similar plans as competition intensifies in the rapidly expanding AI market.

Despite the growing demand for custom AI processors, Broadcom acknowledged that profitability remains a challenge due to the large amounts of high-bandwidth memory required by advanced AI chips. The company relies on memory suppliers such as SK Hynix and Samsung Electronics to support these products.

As the AI race accelerates, OpenAI’s introduction of Jalapeño signals a significant shift in the company’s strategy. By developing its own hardware, OpenAI is positioning itself to compete more effectively in an industry where computing power has become one of the most valuable resources driving innovation and growth.