NVIDIA’s Huang casts AI as a new “token” economy, framing inference as the successor to Bitcoin mining

Huang frames compute as currency

NVIDIA (英伟达) CEO Jensen Huang (黄仁勋) used a keynote in San Jose’s SAP Center to recast AI compute as an emergent token economy — a direct parallel to how Bitcoin (比特币) turned electricity and hardware into monetary rewards. He declared the end of an old computational order and pitched a new metric: Token, not bytes. Short question: who gets to mint value in the AI era — cloud operators, chip makers or whoever owns the fastest inference pipelines?

Mining vs. inference: the analogy and the math

Huang argued that mining and inference are two sides of the same coin: both convert power into tradable outputs. In Bitcoin, miners compete for block rewards; in AI, models compete for “inference rights” and per‑token revenue. It has been reported that token generation from a 1GW data‑center jumped from 22 million to 700 million over two years — a roughly 350x rise — and that global AI compute demand could exceed $1 trillion by 2027. Huang even laid out a 25/25/25/25 compute‑tier blueprint for enterprises, saying he does not “produce tokens” himself but defines the rules and aligns NVIDIA’s product stack to them.

New silicon and software — faster, cheaper tokens

NVIDIA presented a stack pairing its new Vera Rubin prefill stage with Groq’s low‑latency LPU for decoding, coordinated by Dynamo software. NVIDIA said the combination can boost token throughput by up to 35x versus its previous Blackwell platform and can enable some workloads above 1,000 tokens per second. The company and analysts compared the shift to past mining hardware revolutions — CPU → GPU → FPGA → ASIC — arguing that deterministic, single‑purpose accelerators for decoding are the “ASIC moment” for inference.

Market and geopolitical implications

The commercial stakes are clear: whoever can deliver tokens cheapest and fastest will set prices and capture margins. That matters not just for cloud economics but geopolitics. Export controls and trade policy — notably U.S. restrictions on advanced AI chips to China — shape where those compute “factories” can be built and who will control the new rails of value. As Huang reframes compute as currency, the question looms: will data‑center operators become the new banks of the digital age?