OpenAI and Broadcom Unveil 'Jalapeño,' the ChatGPT Maker's First Custom AI Chip, Built in a Record Nine Months

OpenAI and Broadcom have unveiled Jalapeño, OpenAI's first custom-designed artificial intelligence chip, a milestone that pushes the ChatGPT maker deeper into the fiercely competitive market for the silicon that powers modern AI.

Announced June 24, the processor is what OpenAI calls its first "Intelligence Processor," an accelerator built specifically for inference — the work of running already-trained AI models to answer user queries, as opposed to the training that creates them. It is the opening act of a multi-generation compute platform the two companies say they are building together, with initial deployment targeted for the end of 2026.

Perhaps the most striking claim is the speed of its development. Broadcom and OpenAI say Jalapeño went from initial design to manufacturing tape-out in just nine months, a pace that may rank among the fastest ever achieved for a high-performance, reticle-sized ASIC. Executives credited deep collaboration between OpenAI's engineers and Broadcom's silicon designers — and, notably, the use of OpenAI's own models to accelerate parts of the chip's design and optimization.

The strategic logic is straightforward. OpenAI's compute demands have exploded, and the company has leaned heavily on Nvidia's graphics processors, which are expensive and often in short supply. By designing its own inference chip, OpenAI can tailor the hardware to the way its models actually run, potentially squeezing out more performance for every watt of power consumed. Early testing, the companies said, shows Jalapeño delivering performance per watt "substantially better" than current state-of-the-art hardware.

Broadcom has quietly become one of the biggest winners of the AI boom, positioning itself as the go-to partner for tech giants that want to design custom chips rather than buy off the shelf. The Jalapeño announcement came alongside word that Apple had extended its multibillion-dollar wireless chip supply agreement with Broadcom through 2031, underscoring how central the company has become to the industry's biggest players.

For the broader market, the launch is another sign that the era of one-size-fits-all AI hardware is ending. Google has long designed its own tensor processing units, Amazon and Meta are building custom silicon, and now OpenAI is joining the club. Analysts caution that designing a chip is only half the battle — manufacturing at scale, writing the software to support it and proving reliability in data centers are formidable challenges that have humbled other newcomers.

Still, if Jalapeño performs as promised, it could reshape OpenAI's economics at a moment when the cost of serving hundreds of millions of users is one of the company's largest expenses. The chip is designed for initial deployment by year's end, with successors expected to follow, marking the start of what the companies frame as a long march toward faster, cheaper and more widely available artificial intelligence.