TL;DR: As major AI labs compete for customers, a big bad margin problem has reared its ugly head: More users means they need even more (expensive) compute. This strain is sharpest at Anthropic—users are left frustrated by stricter usage limits and more outages, with the frequent downtime pushing some enterprise clients to competitors. What happened: As AI companies’ customer bases grow, power users are also devouring compute for intensive tasks like agentic AI, and it’s creating a supply crunch. It’s a problem we’ve seen before as new tech infrastructure gets built out, whether it's railroads or the internet boom: Demand bubbles up quicker than the plumbing needed to run it, and price hikes are the first (or only) available lever, per the Wall Street Journal. The CEO of a cloud infrastructure firm told the WSJ that “the power that’s available through 2026 is already all spoken for.” There’s also a cost to getting ahead of yourself. As Anthropic CEO Dario Amodei put it, it’s better to lose customers in the short term than bankrupt the company because you overestimated demand. Everyone’s rationing: With such compute constraints, AI firms are tightening their belts. OpenAI recently scrapped its Sora video model in part to free up precious compute, and last week it launched a new $100-per-month ChatGPT Pro tier (its priciest option is $200). OpenAI also set new daily limits for Plus users to prevent long marathon sessions. Anthropic now meters tokens during peak weekday hours—8am to 2pm ET—and Google limited free access to Gemini 3 after a demand surge last November. Claude’s catch-22: The demand vs. compute tension is sharpest at Anthropic. It’s gained a lot more users this year, with its annual revenue run rate more than tripling from $9 billion late last year to $30 billion this month. But it’s also a much smaller company than OpenAI, which appears more willing to spend the big bucks to grab as much compute as possible. OpenAI also claims that integrated ads inside its chatbot (an option Anthropic has so far ruled out) will bring $100 billion in revenue by 2030. AI diehards hit a wall: Power users are now facing limits after just a few prompts, which has forced them to reorient their workdays around peak hours and reset windows. AI companies, meanwhile, continue to stack increasingly granular subscription tiers, ranging from free to over $200 a month, in an effort to reach the sweet spot of growing users without melting servers or margins. Bottom line: OpenAI and Anthropic expect to spend nearly $65 billion combined in 2026 to train and run their models, per the WSJ. Both are eyeing IPOs that may come later this year, and right now, the math on their margins looks a little dicey. —WK | 
