The most dangerous words in AI right now are "Pro Max exhausted in 1.5 hours."
Anthropic quietly downgraded its cache TTL from one hour to five minutes on March 6th. It sounds technical. It isn't. What it means: the company's best customers—the ones paying for the expensive tier—are burning through their monthly allowance faster than anyone expected, because the system is now refreshing itself more often and charging them each time. A developer reported hitting their quota ceiling on a Pro Max plan in 90 minutes. The response wasn't "we'll fix this"—it was silence, then a buried GitHub thread.
This matters because it reveals something nobody talks about: the companies winning the AI race right now aren't winning because their models are better. They're winning because they've found the sweet spot between capability and scarcity. Make it powerful enough that developers can't live without it. Make it expensive enough that you don't have to worry about load. Make it just broken enough that people blame themselves for not being smart enough to use it.
Markets haven't priced this in yet. AI companies are still valued on the assumption that their APIs will scale gracefully and capture network effects the way cloud infrastructure did in 2010. But that's not what's happening. What's happening is bottleneck management disguised as product development.
Coinbase filed an 8-K on April 7th. Google filed an 8-K on the same day—the same day insiders at Google also filed Form 4s. The timing is almost certainly innocent. But it's worth asking: what does it mean when material events cluster like this during a week when geopolitical risk is supposedly rising and oil prices aren't moving?
It means nobody actually believes oil prices are going to move. Or the Strait blockade matters. Or anything.
What they do believe in is the bottleneck. The quota. The thing you can't get enough of. Anthropic's quota exhaustion isn't a bug. It's a business model finding its shape. And when a developer has to choose between upgrading to a more expensive tier or finding a competitor who won't throttle them—that's when you'll see the real market stress.
There's a broader pattern here that should alarm you: every crisis this week (blockade, refinery shutdown, visa chaos) has met the market with indifference. But every constraint that's artificial—quota walls, TTL downgrades, insider filing clusters—gets treated as signal. We're not pricing in real-world friction anymore. We're pricing in artificial friction, because that's where the money is being made.
The paradox: the more fragile supply chains become, the more investors chase the companies that are deliberately creating artificial scarcity. It's the opposite of capitalism. It's feudalism with APIs.
Watch for complaints about quota exhaustion to go mainstream in the next 72 hours. When developers start migrating away from Claude because of cost friction, the market will finally move. But it won't move on geopolitical risk.
It'll move on the constraint it can actually price.