Why token prices are falling

Easy version of the U.S. part of DK's November 4 note. This is not investment advice. Only summary numbers are used.

In one line, AI usage fees are falling fast, while about $500 billion of graphics chips are still waiting to be installed. TS Lombard put more weight on the bearish case that supply could become too large.

A token is a unit an AI model uses to read and write. The token price is the fee for that use. TS Lombard compared AI with railroads, fiber, and the dot-com buildout: the tool becomes common, and the benefit can go to the final user rather than the company that built the pipes. It said data centers with fixed costs, debt, and take-or-pay contracts, promises to pay for capacity whether it is used or not, are the most exposed. A data center is a large building of computers that runs AI calculations. It also said Chinese open models trailing by four to seven months could shrink the lead of frontier labs. A moat is an advantage others cannot easily copy.

BofA called an AI-spending slowdown a large risk to U.S. and world growth through both growth and the wealth effect. The summary said chip makers' pricing power lasts only while chips are scarce, and the biggest risk sits in data centers. The first signal to watch is GPU rental rates. A GPU is a graphics chip used for AI math. If rental rates that held up after July break while token prices fall, the bearish case becomes real. If daily token use rises faster than the price drop, or if Nvidia adds more ecosystem support, the bullish case gains weight.

Three lines to keep. Falling fees plus more equipment can hurt data centers first. Chip pricing power was described as temporary. Rental rates were the confirmation signal in the summary.

The $500 billion and four-to-seven-month figures were not rechecked against the long report. The bearish case is a report's view.