OpenAI Unveils Jalapeño Custom Chip

OpenAI and Broadcom unveiled Jalapeño on Wednesday, the ChatGPT maker's first custom-built inference processor designed specifically for running large language models.

A physical sample of the chip was delivered to OpenAI CEO Sam Altman and President Greg Brockman , marking a major step in OpenAI's strategy to control the full technology stack behind its AI products.

The chip was co-developed from initial design to manufacturing tape-out in just nine months, which OpenAI claims may be the fastest ASIC development cycle ever achieved in high-performance advanced semiconductors.

That speed reflects deep software-hardware co-development with OpenAI's engineering teams, Broadcom's silicon implementation expertise, and the use of OpenAI models to accelerate parts of the design and optimization process.

Why Build a Custom Chip Now?

OpenAI President Greg Brockman told CNBC that OpenAI "cannot get compute fast enough," and Broadcom CEO Hock Tan backed up that assessment, saying compute demand from the company's six customers is "simply insatiable" and will remain elevated through 2028.

Jalapeño is specifically designed for inference, the process of running pre-built AI models in response to user commands.

OpenAI emphasized the chip's low operating cost when running real-time coding models, though more performance-intensive tasks like pre-training will likely still rely on Nvidia hardware.

The chip is an ASIC, which industry experts say is less flexible than Nvidia's GPU, but is also less expensive and can be designed for specific AI tasks.

Early testing shows that Jalapeño will deliver performance per watt substantially better than current state-of-the-art, though OpenAI is still measuring final performance.

The Economics Behind the Move

The financial pressure driving OpenAI's chip strategy is substantial. Last year, keeping ChatGPT servers responsive cost OpenAI $8.4 billion, and with the platform now attracting 900 million weekly users, that operational cost is projected to reach approximately $14 billion this year.

While Nvidia currently commands an estimated 75% profit margin on its high-end processors, OpenAI operates on tighter margins, keeping roughly 33 cents of profit on each dollar generated after accounting for its massive operational expenses. By developing custom silicon optimized for its specific workloads, OpenAI aims to improve those unit economics.

Google and Amazon have both built custom chips to serve a similar purpose, often called "AI accelerators" — silicon designed specifically to speed up machine learning workloads.

Counterpoint Research expects Broadcom to control 60% of the custom AI processor market in 2027.

Gigawatt-Scale Infrastructure Plans

The Jalapeño announcement builds on a broader partnership. In October 2025, OpenAI and Broadcom announced a collaboration for 10 gigawatts of custom AI accelerators, with deployments targeted to start in the second half of 2026 and complete by end of 2029.

Broadcom CEO Hock Tan said the companies are "enabling the deployment of gigawatt-scale data centers with Microsoft and other partners beginning in 2026."

For comparison, a gigawatt is a unit normally used to measure the output of large power plants.

The companies are aiming for initial deployment of the Jalapeño chips by the end of 2026, with Broadcom CEO Tan telling CNBC there would be "small prototype development" in late 2026, ramping up in 2027 and "going full tilt in first half '28."

Vertical Integration Strategy

OpenAI stated it is "not only developing frontier models or building products on top of them; it is designing the infrastructure underneath them: chip architecture, kernels, memory systems, networking, scheduling, deployment systems, and product experience," with each layer optimized around making its models faster, more reliable, and more affordable for users.

OpenAI designed the chip from scratch around its deep understanding of LLM fundamentals, informed by its roadmap of models, kernels, serving systems, and product needs, with partners Broadcom and Celestica helping industrialize the platform through chip implementation, board, rack system integration, high-performance networking, and scalable production systems.

Engineering samples of the Jalapeño chip are running ML workloads in the lab at production target frequency and power, including GPT-5.3-Codex-Spark.

The chip is manufactured by TSMC, the world's leading advanced semiconductor foundry, while server maker Celestica will help build systems around the processor.

What Changed This Week

OpenAI's entry into custom chip design represents a fundamental shift from software company to vertically integrated infrastructure provider. The nine-month development timeline, accelerated by using AI models to design AI hardware, sets a new benchmark for semiconductor development speed. With Broadcom shares climbing following the announcement and deployment beginning later this year, the move signals that leading AI companies view control over silicon as essential to competing at scale.

What to Watch

OpenAI promised a detailed technical report on performance will be presented in the coming months. The key metrics to monitor include actual performance benchmarks against Nvidia's H100 and H200 chips, real-world cost savings per inference operation, and whether OpenAI licenses the chip to external AI firms as some sources have suggested. The AI inference chip market, valued at $20.51 billion in 2026, is projected to reach $36.97 billion by 2030, growing at a 15.9% CAGR. Jalapeño's success or failure in this competitive landscape will determine whether other AI labs accelerate their own custom silicon programs.

Reporting based on coverage from TechCrunch, CNBC, OpenAI, Broadcom, CNN, VentureBeat, Tom's Hardware, Bloomberg, June 24-25, 2026.