Celebrating CUDA's 20th Anniversary, Nvidia's Jensen Huang Calls Token the Basic Unit of the AI Era

IT Home, March 17 — The GPU Technology Conference (GTC) opened today (March 17) in San Jose, USA. NVIDIA CEO Jensen Huang explicitly stated in his speech that “tokens” are the fundamental building blocks of the new AI era.

Huang celebrated the 20th anniversary of the CUDA architecture during his speech, emphasizing that CUDA is the “core flywheel” driving accelerated computing, and that this acceleration is surpassing Moore’s Law.

Huang said, “We have spent 20 years building hundreds of millions of GPUs and computing systems running CUDA worldwide. We have penetrated every cloud and every computer company. We serve almost every industry.”

According to a blog post cited by IT Home, Huang stated that NVIDIA is an algorithm platform company supporting platforms like Google Cloud, Azure, and AWS. Its data acceleration libraries, cuDF and cuVS, have been adopted by many leading platforms globally.

When processing structured data, cuDF can speed up data engines by up to 5 times; cuVS significantly accelerates feature extraction for unstructured data. Snap’s CIO said that deploying cuDF reduced the company’s daily data processing costs by 76%, enabling analysis of 10 PB of data within three hours, saving millions of dollars directly.

The conference also highlights open-source project OpenClaw. Through the “Build-a-Claw” activity, attendees can quickly deploy an “always-on” local AI assistant using a DGX Spark workstation or GeForce RTX laptop, without compromising any personal privacy.

Image source: NVIDIA