Google pulled the curtain back on its latest creation in San Francisco, the seventh-generation AI chip called Ironwood. Announced at a cloud conference, this chip is all about making AI-powered apps, like OpenAI’s ChatGPT, run faster and smoother. It’s built for what tech folks call inference computing, basically, the heavy lifting needed to spit out quick answers in a chatbot or whip up other responses on the fly.

Google’s been pouring billions into this project for nearly a decade, crafting its tensor processing units (TPUs) as a real competitor to Nvidia’s top-notch AI chips. Unlike Nvidia’s offerings, Google keeps its TPUs in-house; only its own team or cloud service users can tap into them. That’s given Google a leg up in the AI race compared to some of its peers.

A while back, Google split its TPU lineup into two flavors: one beefy version for cooking up massive AI models from the ground up, and a leaner one that skips some of those bells and whistles to cut costs for running AI apps. Now, Ironwood steps in as the star of the show for inference tasks. Amin Vahdat, a Google VP, explained it’s designed to team up in huge groups up to 9,216 chips strong.

This new chip blends the best of those earlier designs and packs more memory, making it a champ at handling AI workloads. “It’s just that the relative importance of inference is going up significantly,” Vahdat said. Compared to last year’s Trillium chip, Ironwood doubles the speed for the same power, which is a big deal for running Google’s Gemini AI models.

Google stayed tight-lipped about whose manufacturing Ironwood. Meanwhile, Alphabet’s stock shot up 9.7% by the end of the day, boosted by news of President Trump scrapping some tariffs out of the blue.