AI compute crunch drives Nvidia GPU rental costs up 48%

The artificial intelligence gold rush is hitting a wall, as a severe shortage of computing power drives rental costs for Nvidia's latest GPUs up 48% in just two months and forces major AI labs to curtail services.

"I do spend a lot of time trying to find any last-minute compute available," Sarah Friar, OpenAI’s chief financial officer, said in a recent public interview. "We’re making some very tough trades at the moment on things we’re not pursuing because we don’t have enough compute."

The capacity crunch has seen spot-market rental prices for an Nvidia Blackwell GPU jump from $2.75 to $4.08 per hour in two months, according to data provider Ornn. This has forced cloud provider CoreWeave to raise prices by over 20% and extend contract terms. Meanwhile, Anthropic's Claude API uptime fell to a low of 98.32% in March, causing some enterprise clients to switch providers.

This bottleneck threatens to slow the deployment of powerful new AI tools just as user adoption explodes, impacting the massive investments pouring into the sector and potentially derailing productivity gains that companies have been chasing. The issue pits AI companies in a ferocious competition for limited resources, where the ability to secure computing capacity is becoming a primary determinant of success.

Anthropic Strains Under Demand

The strain is particularly visible at Anthropic, maker of the popular Claude chatbot. The company has experienced frequent outages, with its Claude API uptime dropping from 99.82% in October to a low of 98.32% in March before a slight recovery. The instability has pushed some customers to competitors. David Hsu, CEO of software platform Retool, said he switched his company's AI agent tool from Anthropic's model to one from OpenAI, stating, "Anthropic has just been going down all the time."

In response to the demand, which saw Anthropic's annual revenue run rate reportedly double to $30 billion in just two months this year, the company began limiting user activity during peak hours.

OpenAI Sacrifices Products for Compute

OpenAI, a key competitor to Anthropic, is facing similar pressures. The company's token consumption on its API platform surged from six billion per minute in October to 15 billion in late March. The intense demand for resources led OpenAI to scrap its hyped Sora video-generation app, in part to reallocate computing power to more lucrative enterprise and coding products, according to The Wall Street Journal.

The scarcity is creating a seller's market for infrastructure providers. CoreWeave, a major AI cloud company, has not only increased prices by more than 20% but is also requiring smaller customers to commit to three-year contracts, up from one year previously. "The lead times are too long," said J.J. Kardwell, CEO of cloud infrastructure firm Vultr. "Data center build times are long, the power that’s available through 2026 is already all spoken for."

This article is for informational purposes only and does not constitute investment advice.