OpenAI unveils Jalapeño: its first custom chip with Broadcom

- June 24, 2026
- Updated: June 25, 2026 at 7:42 AM

OpenAI just introduced its first chip Jalapeño, with Broadcom as the partner on the project. It’s meant for the infrastructure that runs OpenAI’s own models. The chip is an inference-focused ASIC, built for the split-second replies behind tools like ChatGPT, and OpenAI is counting on it to better performance per watt, cut long-term costs, and rely a little less on suppliers like Nvidia.
Jalapeño is squarely aimed at inference, the part that handles the live responses people actually see when they use something like ChatGPT. For OpenAI, the bet is pretty clear: more work done per watt, a cheaper path over time, and less exposure to Nvidia if supply or pricing gets tight.
Reports say Jalapeño went from design to finished hardware in roughly nine months. OpenAI also used its own models to speed up parts of that development process.
Broadcom says Jalapeño can compete with Nvidia’s Blackwell chips and Google’s TPUs. Still, without full specs or outside benchmarks, there’s only so much you can conclude about raw throughput, power efficiency, cost, or how well it holds up when you scale it across real deployments.
For people who follow AI infrastructure, this is the kind of move that matters. It puts OpenAI alongside Google, Amazon, and Meta, companies that are designing hardware around the software they want to run.
There’s nothing here to install or try for yourself. Jalapeño is an internal data-center chip, and as it rolls out, it’ll probably bring even more attention to the energy use, water consumption, and overall footprint that comes with bigger AI clusters.
Choose Softonic on Google and you'll see more of our articles in Top Stories, AI Overviews and AI Mode.
I’m a journalist with more than 30 years of experience in video games and technology. Although my specialty has always been video games, I’ve recently started enjoying exploring the intricacies of project-management tools like Asana, as well as automations with Make.com and N8N.
Latest from Jesús Bosque
You may also like
NewsDiscord’s new Rust client surfaces: 290MB RAM, 1-second launches
Read more
NewsWhite House network appeared as residential proxy exit, researchers say
Read more
NewsAvengers: Doomsday Jeep ad reveals Latveria: first on-screen look at Doom’s MCU home
Read more
NewsApple confirms $7.5 billion bond sale: orders top $30 billion
Read more
NewsHomeland finds new life on streaming: The Game revival is still unconfirmed
Read more
NewsThe Legend of Zelda: Ocarina of Time is out now on Switch 2: change this setting first
Read more