Overview
- Huang said a $500,000 software engineer should consume at least $250,000 in AI tokens annually and indicated he would be alarmed by far lower usage, framing token spend as a performance signal.
- He proposed giving engineers token budgets worth roughly half of base pay and disclosed Nvidia is allocating tokens for thousands of engineers the way it issues laptops.
- Supporters such as Box CEO Aaron Levie forecast token budgeting will spread beyond engineering into roles like legal and sales as autonomous agents run continuously and consume more compute.
- Skeptics warn of runaway costs and uncertain returns, with Chamath Palihapitiya citing a startup’s decision to halt use of an AI coding tool after token and inference bills surged.
- Analysis of Nvidia’s GTC reveals a coordinated bet on agentic, continuous workloads replacing prompt‑based usage, tying new hardware and software to a projected $1 trillion in demand through 2027.