OpenAI is about to cut token prices by 40 to 50 percent. Not because serving got cheaper. Because they think Anthropic is about to do the same.

Nick Mugalli has been laying out the pieces for months. This is the one that makes the trade lose steam.

Two weeks ago, compute repriced up 50 percent. Same seller, SpaceX. Two buyers, Google and Anthropic, in the same week. The input that everyone was excited about got more expensive. Now the output, the token, the only thing any of this actually sells, is about to get cut in half.

Input up 50. Output down 50. There is no clever name for that. It is not a land grab. It is not scale economics. It is selling dollars for 80 cents and announcing a sale on the dollar.

## The Price War Nobody Ordered

The logic is defensive. OpenAI cuts prices because they believe Anthropic will cut prices. Anthropic cuts prices because they believe OpenAI will cut prices. Both are right. Neither can afford to be the one who held the line while the other captured market share.

The problem is the timing. This price war is starting at the exact moment when the capital structure supporting it is stretched thinner than it has ever been.

Last week, we published the numbers on AI subscription subsidies. Claude Max runs at negative 900 percent margin at full utilization. ChatGPT Pro runs at negative 1,650 percent. Those numbers assumed current pricing. A 40 percent price cut makes them worse by almost half.

## The Three Bodies Problem

The AI trade has three interconnected markets. Compute supply is concentrated and repricing upward. Token demand is price sensitive and rolling over. The financing that bridges them is circular. Everyone in this trade is everyone else’s customer, lender, and collateral.

Oracle sits at the center of the debt stack. Its backlog of remaining performance obligations sits at $638 billion, all of it underwritten by the labs’ future revenue. If that future revenue shrinks by 40 percent per unit, the math on the backlog changes.

The hyperscaler capex we analyzed earlier this week showed over $1.8 trillion in off-balance-sheet liabilities. Those liabilities assumed a certain trajectory for AI revenue growth. A price war at the top of that pyramid changes the trajectory.

## What Breaks First

Three things happen when token prices drop 40 percent in a market that was already subsidizing consumption by 10x to 70x per user.

First, the reseller layer compresses. Every business that built a markup on top of API pricing wakes up to find their input is now someone’s loss leader. The margin disappears. The wrappers, the middlemen, the inference resellers all get squeezed between falling output prices and rising compute costs.

Second, enterprise procurement shifts. Companies have been quietly downgrading to cheaper, good enough models for months. Usage leaderboards are disappearing. Surprise token bills are showing up in earnings calls. The price cut accelerates the migration. Citadel recently published a note called Tokenomics that reached the same conclusion: AI is no longer gated by what it can do. It is gated by what it costs. Cost curves. Rationing. The boring physics every miracle eventually meets.

Third, the Chinese open models keep sawing through the floor from below. They are already cheaper by a wide margin. Every price cut by OpenAI or Anthropic narrows the gap but does not close it. The commodity pricing floor is set by the lowest cost producer, and that producer is not in Silicon Valley.

## The Balance Sheet Test

The question is not whether this price war happens. It is happening. The question is which balance sheets survive it.

The two or three companies that can lose money longer than everyone else will stay solvent. Everyone in the middle will not. The resellers, the markups, the wrappers, the businesses built on the assumption that API pricing would hold or increase.

When intelligence becomes a commodity, it announces itself the way every commodity does. With a price war. The hype was what AI could do. The reckoning is what it costs.