Anthropic’s Claude Code pricing fiasco is what it looks like when a company is squeezed at three ends. Anthropic quietly removed Claude Code from the $20 Pro plan, making it exclusive to the $100 and $200 Max tiers. Their Head of Growth framed it as a small test on 2% of new signups (which didn’t match what users were seeing). Within hours they reversed it.

What interests me is what the test reveals about the bind Anthropic is in. It was an attempt to fix unit economics: heavy users on flat-fee plans consume vastly more than the plans recover, and the Head of Growth, Amol Avasare, said as much on X – plans weren’t built for current usage patterns.

That’s one real pressure. But it’s not the only one. They’re squeezed three ways at once.

The first squeeze is unit economics. Someone running Claude Code all day on a $20 subscription costs far more to serve than they pay. Either prices go up or costs come down. However raising prices risks making them uncompetitive against OpenAI and Google, who are already taking advantage of this moment.

The second squeeze is compute. Claude has been below 99% uptime for a quarter. They are clearly struggling with the huge increase in demand they’re experiencing. A year ago the product was mostly chat. Today a significant share of usage is coding agents running for hours. Demand shape changing faster than provisioning can keep up.

So why not do what Gmail and Bluesky did and gate new signups? Match supply to demand, protect the experience for existing users, generate some FOMO and desirability in the process, and buy time to sort the rest out.

That brings us to the third squeeze. Anthropic’s valuation, like that of every frontier AI lab, rests on growth trajectory rather than current profitability. However dressed up, limiting signups reads as a capacity wall, and from there it’s a short step to growth slowing and the IPO narrative wobbling.

The best approach for managing the compute squeeze is ruled out by the growth squeeze, which means infrastructure strain has to be absorbed through rate limits and outages instead, upsetting all your existing users in the process.

It also means heavy users keep arriving, which continues to make the unit economics worse, which is how you end up running silent pricing tests on Tuesday afternoons.