Jensen Huang: "Token" economics will decide which tech firms live or die — NVIDIA pushes full-stack control from chips to robots

NVIDIA's new thesis: tokens as currency

At the opening of NVIDIA's GTC in San Jose, CEO Jensen Huang argued that the economics of generating AI "tokens" — the discrete units of output from large models — will directly determine companies' revenue and survival in the AI era. He framed a sweeping industrial strategy: to sustain a multi‑trillion‑dollar intelligent economy, compute must be redesigned from a system‑engineering perspective. NVIDIA (英伟达) aims to control five layers — energy, chips, infrastructure, models and applications — and thereby morph from a chip supplier into what Huang called an "AI power" company. Is this a marketing rebrand or a fundamental shift in how digital infrastructure is provisioned? The implications are broader than silicon.

New architecture, software and robots

Huang unveiled the next‑generation Vera CPU + Rubin GPU architecture — dubbed Vera Rubin Ultra — and said the platform can link up to 144 GPUs in a single system with tight hardware‑software vertical integration. He touted liquid cooling and co‑packaged optics as key enablers of much higher energy efficiency; it has been reported that NVIDIA expects the new stack to deliver up to a fivefold improvement in revenue‑per‑watt economics for customers. On the software side NVIDIA introduced NemoClaw, a reference stack for agent‑style computing, and pushed Project GR00T, a generalist model for humanoid robots. The company also celebrated CUDA's 20th anniversary, previewed DLSS 5 for neural rendering, and highlighted partnerships with automakers including BYD (比亚迪), Hyundai and Nissan for Robotaxi readiness.

Market and geopolitical stakes

Huang set a bold target for demand, saying global compute needs will surge — it has been reported that he predicted compute demand could breach $1 trillion by 2027 — and argued that lowering per‑token cost is now the industry's fundamental economics. That vision comes amid persistent geopolitical headwinds: U.S. export controls and trade tensions have already reshaped how advanced AI chips are sold and deployed globally, raising questions about supply‑chain resilience and market access. If token‑level costs truly become the unit of competition, firms that manage energy, packaging, data‑center ops and software together may gain outsized advantage — but can such vertically integrated players thrive when geopolitics slices markets and restricts technology flows?