Token spending — a new productivity yardstick
Box’s CEO said he’s comfortable with engineers 'wasting' some AI tokens during experimentation, and reports note Jensen Huang recently set an explicit benchmark for token usage among top engineers. Those comments frame heavy AI‑tool spend as infrastructure rather than discretionary cost. (businessinsider.com)
Silicon Valley executives are starting to treat AI token spending like electricity: a core input, not a line item to squeeze. (businessinsider.com) On April 11, 2026, Business Insider reported that Box chief executive Aaron Levie said he was comfortable with engineers “wasting” tokens while they test new ideas. Levie said a larger token bill can signal that teams are experimenting rather than standing still. (businessinsider.com) Levie made the remark on a recent episode of the a16z Show, where he argued that early overuse is part of learning how agents and coding tools fit into work. In a separate March post on X, he said token costs will start in engineering, where developers can run multiple agents in parallel or let projects run overnight. (youtube.com, businessinsider.com) A token is the unit artificial intelligence systems count when they read or write text, and companies pay for many tools by the token. More prompts, longer context windows, and more agent runs all push the bill higher. (businessinsider.com, forbes.com) That accounting detail is turning into a management metric. Nvidia chief executive Jensen Huang said on the All-In Podcast in March that if a software engineer or artificial intelligence researcher making $500,000 used only $5,000 in tokens for the year, he would be alarmed. (cnbc.com, rdworldonline.com) Huang put a number on it: he said a $500,000 engineer should consume at least $250,000 worth of tokens annually, or about half their salary. He also said tokens are becoming a recruiting tool in Silicon Valley, not just a usage bill. (cnbc.com, finance.yahoo.com) Other companies are already talking the same way. Forbes reported on March 31 that Meta, Nvidia, and Databricks were rewarding engineers for using more artificial intelligence, with Meta chief technology officer Andrew Bosworth saying there was “No limit.” (forbes.com) The spending is moving beyond software teams. Levie said legal and sales workers could become heavy token users too, and Visa told Business Insider on April 10 that it was burning through nearly 2 trillion AI tokens a month as internal use accelerated. (businessinsider.com, businessinsider.com) The counterargument is cost control. Levie told Fast Company in March that governance and cost could slow enterprise adoption even as agents improve, and Forbes reported on April 10 that users are already hitting token caps and rate limits faster than expected. (fastcompany.com, forbes.com) For now, the signal from executives is blunt: in 2026, some companies are reading a big token bill less as waste than as proof their people are actually using the new tools. (businessinsider.com, cnbc.com)
