The gold rush of artificial intelligence is rapidly draining the one resource developers cannot do without: computing power. This sharp shortage of capacity is causing frustration among expert users, forcing companies to cancel products and causing reliability issues.
In recent months, demand has exploded for so-called “agent” AI, autonomous tools that perform tasks independently. Companies are struggling to secure the computing power needed to serve a growing customer base that is significantly increasing AI usage.
“Everyone talks about oil, but I think the world is especially short on tokens,” said Ben Pouladian, engineer and tech investor. […] Historically, price increases have been one of the few ways to address a supply crisis, but such a move could be dangerous for leading companies engaged in fierce competition to acquire users.
Prices for hourly GPU rentals, the microchips used to train and run models, have soared since autumn. Anthropic, maker of the Claude chatbot, has been hit by frequent outages and has begun rationing computing supply during peak hours. OpenAI has dropped the Sora video generation app, partly to free up resources for programming and enterprise products based on a new model.
Token usage in OpenAI’s APIs rose from six billion per minute in October to 15 billion by the end of March. Sarah Friar, CFO of OpenAI, admitted: “We are making very difficult trade-offs on projects we are not pursuing because we don’t have enough compute.”
Toward the end of last year, CoreWeave, one of the largest AI cloud companies, raised prices by over 20%. […] According to the Ornn Compute Price index, renting one of Nvidia’s most advanced Blackwell chips now costs $4.08 per hour, a 48% increase compared to just two months ago.
“There is a massive capacity crisis, unlike anything I’ve seen in over five years of business,” said JJ Kardwell, CEO of Vultr. “The question is: why don’t we just install more equipment? Delivery times are too long. Data center construction times are biblical, and available power through 2026 has already been fully booked.”
Due to poor reliability, some enterprise customers are migrating to other providers. David Hsu, CEO of Retool, explained he switched to OpenAI models because Anthropic “keeps falling down continuously.” Core internet service reliability is usually measured in “nines” (99.99% uptime is the standard), but Anthropic’s API recorded 98.95% uptime over the last 90 days.
These disruptions occur while Anthropic is experiencing explosive growth: its annual revenue run rate rose from $9 billion at the end of 2025 to $30 billion in April 2026. At the end of March, the company suddenly limited the number of tokens usable during peak hours, sparking user complaints on social media. Boris Cherny, head of Claude Code, responded: “Capacity is a resource we manage carefully and we are prioritizing customers who use our products and APIs.”
