OpenAI and Broadcom unveiled Jalapeño, a custom inference processor designed in a nine-month development cycle to optimize LLM inference at scale.

OpenAI and Broadcom announced their first custom AI chip on Wednesday, completing a nine-month development sprint that marks the ChatGPT maker's entry into silicon design and represents its most direct challenge yet to Nvidia's dominance in AI computing.

Called Jalapeño, the chip is an application-specific integrated circuit (ASIC) built exclusively for inference—the process of running trained AI models in response to user prompts. It is not designed for training. This distinction is deliberate: OpenAI's $30 billion investment in Nvidia from February 2026, combined with an agreement to deploy 10 gigawatts of Nvidia's Vera Rubin platform, keeps the GPU giant central to its training infrastructure.

The partnership was publicly announced in October 2025. Eight months later, physical chips are moving into data centers.

OpenAI President Greg Brockman told CNBC's David Faber that the company's own AI models significantly accelerated the design process. "The degree to which our models have been able to accelerate it was very surprising to us," he said. OpenAI employed prior-generation models rather than its cutting-edge GPT-5.5, compressing what typically spans several years into a nine-month cycle through a software-hardware feedback loop.

Early testing shows Jalapeño delivers "performance per watt substantially better than current state-of-the-art," according to the company. Brockman wrote on X that "perf per watt looking incredible." OpenAI has already tested GPT-5.3-Codex-Spark running on the chip in a production workload, albeit in a test environment.

Broadcom manufactures the silicon and contributes Tomahawk networking technology. Celestica handles board, rack, and system integration.

OpenAI plans to roll out Jalapeño across active data centers by year-end, with additional generations to follow. The economics are urgent. In 2025, OpenAI generated $13.07 billion in revenue against $34 billion in spending, posting an operating loss of nearly $20.92 billion. Research and development costs driven by compute infrastructure consumed $19.18 billion—roughly 56% of total spending—while the company paid Microsoft over $10.59 billion for R&D and compute alone.

Jalapeño represents OpenAI's attempt to bend this cost curve, and the timing aligns with a heavily anticipated initial public offering.

Broadcom CEO Hock Tan told CNBC that compute demand from the company's six customers is "simply insatiable." "It's just much more than we can address," he said, "and this is not just '26, not '27, we're seeing that same and even elevated demand in '28 as well."

Broadcom shares rose more than 1% Wednesday. The stock is up 10% in 2026 and has multiplied nearly sevenfold since the end of 2022.

Jalapeño joins a growing wave of custom AI silicon from Google's TPUs, Amazon's Trainium, Microsoft's Maia 200, and Meta's MTIA lineup. OpenAI is the latest software company to conclude that controlling hardware is no longer optional—it is the only path to sustainable economics.

"By designing more of the stack ourselves, we can serve more intelligence with greater efficiency and keep pushing advanced AI toward broader access," Brockman said in a statement.