NVIDIA GPU Rental Price Surges 48% in Two Months, AI Industry Faces Worst Compute Shortage in Five Years

According to 1M AI News monitoring, a survey by The Wall Street Journal reports that the AI industry is currently facing a widespread compute supply crisis, with GPU rental prices skyrocketing, service outages frequent, products being cut, and customer churn occurring simultaneously.

The cloud spot rental price of NVIDIA's latest generation Blackwell chip has risen to $4.08 per hour, up from $2.75 two months ago, representing a 48% increase. The data comes from the compute pricing data provider Ornn's compute price index, which has recently been integrated into the Bloomberg Terminal. J.J. Kardwell, CEO of cloud infrastructure company Vultr, said, "This is the most severe compute shortage I've seen in over five years running this company. Data center construction lead times are too long, and all available power through 2026 has already been spoken for."

One of the most visibly impacted companies is Anthropic. The 90-day normal operation rate of the Claude API as of April 8 is 98.95%, while the industry standard is 99.99%. David Hsu, founder and CEO of the enterprise software development platform Retool, said that he believed Opus 4.6 was the best enterprise-grade model but ultimately switched to OpenAI because "Anthropic has been constantly down." In late March, Anthropic began limiting user token consumption from 5 a.m. to 11 a.m. Pacific Time on weekdays. Behind the frequent outages is explosive growth: annual revenue has grown from $90 billion at the end of 2025 to $140 billion in February and now $300 billion in April, exacerbating the supply gap itself.

OpenAI is also making trade-offs. API token throughput has increased from 60 billion per minute in October of last year to 150 billion by the end of March. CFO Sarah Friar said, "I spend a lot of time scrounging around for the last bit of available compute power; we are making some very painful trade-offs, and some projects have been abandoned due to lack of compute." Previously, OpenAI had already shut down its video generation app, Sora, in part to free up chip resources for programming and enterprise products.

GPU cloud service provider CoreWeave raised prices by over 20% at the end of last year and now requires small and medium clients to sign at least three-year contracts, up from one year. A Bank of America analyst issued a "buy" rating last month, believing that the supply-demand imbalance will persist at least until 2029.