Nvidia Needed Groq After All

Greetings from the San Jose Convention Center, where Nvidia is throwing its annual AI conference. The AI craze is palpable here. 

Humanoid robots offer fist-bumps to conferencegoers, and attendees are buying sweaters with Nvidia CEO Jensen Huang’s face on them. Nebius, a cloud provider that rents out Nvidia graphics processing units, took over a bar and served guests server-themed cocktails, such as a GPU Spritz and Ryes of the Machines. 

As expected, Nvidia announced its plan to add technology from AI chip startup Groq to its graphics processing unit chip systems, specifically to handle specialized AI inference tasks like coding. (We reported on Sunday about some hidden details about the new system.)

It’s a tacit but major admission by Nvidia that its GPUs, on their own, aren’t necessarily good enough for some of today’s most important AI computing tasks. You wouldn’t have gleaned that from Huang’s somewhat dismissive comments about Groq in January!