The Beginning of AI Scarcity

AI infrastructure is hitting supply chain limits for the first time. Nvidia Blackwell GPU rental rates climbed to $4.08 per hour this week, up 48% from $2.75 two months ago, and CoreWeave raised prices by 20% while extending minimum contract terms from 1 year to 3 years.

OpenAI CFO Sarah Friar said the company is now having to give up on things it wants to do because of a lack of compute. Access to frontier AI is no longer broadly open, and Anthropic has also limited its latest model to only about 40 organizations, with both capacity and security becoming gatekeeping factors.

This shift hits startups harder. The author summarizes this era in five points:

• Relationship Based Selling: Access to cutting-edge models isn't open to every customer; it's allocated first to strategic, high-revenue customers.
• AI to the Highest Bidder: Even with the same model, companies with greater capital and profitability get the advantage.
• Available but Slow: Even with money, fast response times and stable throughput aren't guaranteed.
• Inflationary Commodity: As demand overwhelms fixed supply, prices keep rising.
• Forced Diversification: Developers must spread out across smaller models, on-premise setups, and other providers.

The conclusion is clear: the era of AI abundance is over, and it will likely take years for data center and power infrastructure to catch up.