Meta used to run AI spending leaderboards, its own Instagram chief calls it ‘terrible’

During the early wave of the AI boom, companies including Meta were reported to run internal leaderboards tracking and rewarding massive AI token consumption, a trend that became known in Silicon Valley as ‘tokenmaxxing.’ Now, Instagram head Adam Mosseri has flatly called the practice ‘a terrible idea.’

Speaking on a recent episode of ‘Lenny’s Podcast,’ Mosseri said his division managed to bring its AI costs under control simply by cutting wasteful, early-stage experiments. ‘We’ve managed to get the costs reined in a little bit by shutting down the silly things that we were doing,’ he told host Lenny Rachitsky, according to Business Insider, adding, ‘It is not that hard to build a token incinerator.’

Tokens are the basic units of data AI models process, and top providers charge businesses based on consumption. Mosseri now treats them as a standard corporate resource that must be budgeted alongside GPUs, server storage, RAM and payroll, rather than an unlimited experimentation fund.

The comments come as companies globally face surging AI bills, driven by widespread employee adoption and the rise of ‘agentic AI’ tools that demand far more processing power. Uber’s COO, Andrew Macdonald, recently raised similar alarms, questioning whether Uber’s massive AI spending was producing meaningful business results.

Mosseri predicted that within a year or two, ‘the burn rate of a strong engineer might be the same as their salary or their cost of employment,’ and suggested companies will need to place data caps on workers based on their proven ability to deliver returns.

Image: Wikimedia Commons/by TechCrunch