NVDA | Jensen Huang: Data Centers Will Become Intelligent Token Production Factories

robot
Abstract generation in progress

NVIDIA CEO Jensen Huang delivered a keynote speech at the annual AI event GTC on Monday, defining the concept of a “Token Factory,” stating that future data centers will no longer be just storage centers but factories that produce intelligent tokens.

Huang announced that computing demand is entering a rapid growth phase, with the forecast for compute power demand from now until 2027 doubling from $500 billion to $1 trillion. He said that market demand for NVIDIA’s latest products is unprecedented, with every major player in the computing field utilizing NVIDIA’s technology for development.

Huang emphasized that through extreme “co-design,” NVIDIA’s per-token cost has reached a world-class, unshakable level, forcing global leaders to manage token output rates as they would manage assets. NVIDIA is much more than just a semiconductor company; the growth story is just beginning, and “we will accelerate everyone’s development.”

NVIDIA also unveiled its next-generation architecture codenamed Vera Rubin, featuring a fully liquid-cooled system design and introducing the Vera CPU optimized for AI tasks. The breakthrough mainly lies in integrating Groq’s deterministic streaming processor technology, enabling “decoupled inference” through Dynamo software. This allows Rubin’s massive memory to handle complex contexts, and with Groq chips rapidly generating tokens, this combination achieves an astonishing 350-fold increase in token generation speed for gigawatt-scale factories.

View Original
This page may contain third-party content, which is provided for information purposes only (not representations/warranties) and should not be considered as an endorsement of its views by Gate, nor as financial or professional advice. See Disclaimer for details.
  • Reward
  • Comment
  • Repost
  • Share
Comment
Add a comment
Add a comment
No comments