Billionaire Bill Ackman Just Bet on a $700 Billion Hyperscaler AI Spending Wave

An X station from the billionaire capitalist elaborate a bullish thesis that assumes token costs volition go much affordable. If that happens, it tin further warrant precocious superior expenditures and construe into higher profits for tech giants.

Missed Nvidia successful 2009? This Rare Signal Is Flashing Again. In 2009, a "Double Down" awesome flashed for a little-known chipmaker called Nvidia. For the archetypal clip successful years, that aforesaid "Total Conviction" awesome is flashing for a institution 1/100th the size of Nvidia. Continue »

Why token costs substance

AI models similar ChatGPT work substance prompts arsenic tokens. These tokens are besides generated erstwhile users inquire AI models to make images, videos, and different media.

Each token requires immoderate energy to beryllium produced. A unreality institution similar Amazon will bargain Nvidia chips that tin nutrient the tokens. Other AI chips nutrient tokens arsenic well, but Nvidia has the largest marketplace share.

Investors intimately travel however galore tokens tin beryllium generated per watt of energy consumed to find however costs whitethorn spell down successful the future. If the magnitude of tokens generated per watt of energy consumed doubles, past a hyperscaler's token-related costs are chopped successful half.

Ackman is waiting patiently for tech giants to optimize their token costs. The aboriginal stages of the AI roar person been astir maximizing the fig of processed tokens, careless of however galore tokens were being produced for each watt of energy consumed. He believes markets are underestimating this opportunity, and helium has been utilizing the disconnect to physique positions successful hyperscalers.

Hyperscalers tin extremity up with higher margins and amended gross maturation rates

In theory, token optimization would fto hyperscalers run with less Nvidia chips portion serving the aforesaid fig of customers. However, token optimization won't conscionable amended margins. It volition besides boost request for hyperscalers' AI models and unreality platforms.

Ironically, little token costs tin really summation the request for AI chips. The Jevons Paradox finds that accrued ratio tin construe into much customers, truthful investors shouldn't expect AI chipmakers to clang amid little token costs.

The study Ackman shared went connected to accidental that Microsoft processed much than 100 trillion tokens successful a azygous 4th successful 2025, which was a 5x year-over-year improvement. That included 50 trillion tokens successful 1 period alone, showing meaningful month-over-month growth. Alphabet also notified investors that it was processing much than 16 cardinal tokens per infinitesimal successful Q1.