WORLD

OpenAI unveils first in-house AI chip 'Jalapeño,' says it rivals Nvidia's Blackwell and Google's TPU

by
Jung Mok-hee
Published : June 25, 2026 - 07:10:49
    • Copy Completed!

View Korean Original

From design to tape-out in 9 months — the fastest custom chip development cycle in history — as OpenAI moves to reduce its dependence on Nvidia; chip uses memory from Samsung Electronics and SK Hynix

The OpenAI logo [Reuters]
The OpenAI logo [Reuters]

OpenAI has entered the AI chip market in earnest, unveiling a custom AI semiconductor it has been developing since last year.

OpenAI and Broadcom jointly revealed the AI chip "Jalapeño" on Wednesday and said they plan to deploy it in actual data centers later this year.

Broadcom CEO Hock Tan and President Charlie Kawas handed a Jalapeño prototype directly to OpenAI CEO Sam Altman and President Greg Brockman that day.

According to the two companies, final performance evaluation of Jalapeño is still ongoing, but early tests show it outperforms current state-of-the-art AI chips in performance per watt.

The chip went from initial design to the tape-out stage — when the design is handed off to a foundry for production — in just nine months. The two companies said that is the fastest development cycle ever recorded for a custom application-specific integrated circuit (ASIC).

The rapid development was made possible by using OpenAI's own AI models throughout the design and optimization process.

Tan told Reuters that Jalapeño "is on par with Nvidia's Blackwell chip and Google's tensor processing units (TPUs)."

TSMC in Taiwan will mass-produce the chip. Tan added that Samsung Electronics and SK Hynix are supplying memory chips to Broadcom for the product.

The two companies have also mapped out a roadmap for future chips, Bloomberg reported. The next version is slated for 2028, with new chips to follow annually thereafter.

Tan said that while Jalapeño is focused on inference, future chips may target other areas.

"The world is moving toward a compute-based economy," Brockman said. "Jalapeño will make compute more abundant, enabling faster, more reliable and more affordable AI for individuals and businesses."

OpenAI, widely regarded as Nvidia's single largest customer for graphics processing units (GPUs), has now developed its own inference chip — putting Nvidia under sharper competitive pressure in the AI accelerator market.

Google led the way with its TPUs, and OpenAI has now followed with its own chip, while Anthropic is also exploring chip development. All three leading AI model companies are moving to reduce their reliance on Nvidia.

Nvidia's share price fell more than 1 percent from the previous close as of 1:50 p.m. Eastern time Wednesday, trading in the $197 range during the session.


mokiya@heraldcorp.com
This content was produced with the assistance of AI translation services.

MOST READ