Token economics

For most of this year, two lines moved almost in lockstep: the cost of running AI inference (token expenditure) and hyperscaler stock performance. It does makes sense; more AI demand meant more spend, more spend meant more revenue for the cloud giants selling the compute. From late February through May, both climbed together. Token costs nearly doubled, and hyperscalers rode the wave right along with them. AI demand was the story, and the market rewarded it. Then, something shifted. Since early June, the two lines have diverged in the wrong direction, both falling, but the hyperscaler index has cracked harder, down to 86.42 from a peak above 105. Token costs have come off their highs too, but the correlation that held for months is breaking down. Here’s why that matters: when token costs and hyperscaler value rise together, it signals a demand story, everyone wants more AI, prices and revenue climb together. When they fall together, or hyperscalers underperform the cost trend, it starts to look like a margin story, the market questioning whether the economics of serving AI at scale actually pays off once the buildout hype cools. So, a few questions remain: (1) Is falling token cost a sign of efficiency gains (good for margins) or falling demand (bad for revenue)? (2) Are hyperscalers absorbing compute costs to stay competitive, compressing their own margins? (3) Is the market repricing AI infrastructure bets after a year of aggressive capex?

What has indeed changed Token economics used to be a footnote in earnings calls. Now it’s arguably the single most important variable for understanding hyperscaler valuations. Keep an eye as we move into Q3 earnings calls.

Want to know more? You can register for free at Fund@mental.

#iamfundamental #soyfundamental #wealthmanagement #familyoffice #financialadvisor #financialplanning #policymistake #ratecut #stagflation