Broadcom unveils a custom chip for OpenAI as it challenges Nvidia's dominance

By Britney Nguyen

The custom chip is designed to support the ChatGPT maker's future AI models and products

OpenAI unveiled its first chip on Wednesday.

Broadcom's stock hasn't seen the same level of investor appreciation as smaller chip plays this year, but the company showed on Wednesday that it's very much still in the game when it comes to artificial-intelligence semiconductors.

The company unveiled its first custom chip for OpenAI on Wednesday, calling it the first of multiple generations focused on running future AI models.

The chip, called Jalapeno, is the ChatGPT maker's first "intelligence processor." It was designed by the AI startup using its "deep understanding" of large language models to support inference, according to an OpenAI statement. The chip is part of an effort between the two companies "to make advanced AI faster, more reliable, and more accessible to more people," OpenAI added. Electronics manufacturer Celestica (CLS) partnered on boards, racks and systems for the chip.

Initial deployment of Jalapeno is expected by the end of this year, the AI startup said.

Shares of Broadcom (AVGO) edged about 0.5% higher on Wednesday, while Celestica's stock climbed more than 3%. Despite being up 11% so far this year, Broadcom's stock is the second-worst performer in the PHLX Semiconductor Index SOX in 2026 to date.

OpenAI said the chip is part of its effort to have a "full-stack advantage" beyond its AI models and products. That means being innovative when it comes to chips, memory systems and networking. Still, the company uses chips from Nvidia and Advanced Micro Devices (AMD) as well.

Broadcom designs custom chips, known as application-specific integrated circuits, alongside its customers. Alphabet (GOOGL) (GOOG) is its most notable partner, and the two companies have co-developed Google's tensor processing units for more than a decade. They introduced an inference-focused processor as part of the eighth generation of the chip in April.

While Nvidia (NVDA) has dominated the market for AI chips in the era of training, its leadership will be tested as the industry shifts to inference, or the process of running AI models. The rise in interest in ASICs has prompted debate on Wall Street over the advantages custom chips could have over merchant graphics processing units from Nvidia and AMD in the era of inference.

While GPUs can support broader AI needs, custom chips are designed to support specific tasks that companies may need - and at a lower cost.

The new AI accelerator is optimized to meet the demands of current and future AI models, OpenAI said, adding that engineering samples are now running machine learning workloads, including its GPT-5.3-Codex-Spark text-only coding model. Although it's still in testing, Jalapeno is showing to be more energy-efficient than the current leading-edge chips, the company said.

OpenAI said its chip took nine months from co-development to tape-out, or the final phase of the process when a design is sent for manufacturing.

Don't miss: Micron grew by $1 trillion this year. These two wild cards will determine its trajectory.

-Britney Nguyen

This content was created by MarketWatch, which is operated by Dow Jones & Co. MarketWatch is published independently from Dow Jones Newswires and The Wall Street Journal.

(END) Dow Jones Newswires

06-24-26 1719ET