The GPU Technology Conference (GTC) has commenced in San Jose, USA. At this event, NVIDIA’s CEO, Jensen Huang, posited that “Tokens” are the essential building blocks of the new AI era. He also took the opportunity to celebrate the 20th anniversary of the CUDA architecture, underscoring its role as the “core flywheel” propelling accelerated computing forward, with a pace of development that outstrips Moore’s Law.

Huang explained that NVIDIA operates as an algorithm platform company, supporting a multitude of platforms, including Google Cloud. He highlighted that NVIDIA’s data acceleration libraries, cuDF and cuVS, have been embraced by several leading platforms. Specifically, cuDF dramatically boosts the speed of data engines, while cuVS expedites feature extraction from unstructured data. Following the deployment of cuDF, Snap has witnessed a significant reduction in costs.

The conference will also shine a spotlight on the open-source project OpenClaw, which will empower attendees to swiftly deploy their own localized AI assistants. In NVIDIA’s business landscape, 60% of revenue is derived from hyperscale enterprises, with the remaining 40% coming from other sectors. Huang asserted that NVIDIA offers the lowest cost per token, reinforcing NVIDIA’s position as the “King of Tokens”.