The AI Compute Crisis: Navigating the New Scarcity

For the first time since the early 2000s, technology companies are grappling with the constraints of their supply chains, particularly in the realm of AI compute resources. The cost of renting Nvidia's Blackwell GPUs has surged to $4.08 per hour, a 48% increase in just two months, while CoreWeave has raised prices by 20% and extended contract durations. This scarcity is forcing companies like OpenAI to make difficult decisions about which projects to pursue due to limited compute capacity.

Access to cutting-edge AI models is becoming increasingly restricted, with companies like Anthropic limiting their newest models to a select group of organizations. This trend is creating a landscape where only the most financially robust companies can afford the latest AI technologies. The era is defined by five key characteristics: relationship-based selling, where access is granted to strategic customers; AI models going to the highest bidder; slow availability even for those who can pay; inflationary pressures driving up costs; and a forced diversification as developers seek alternatives.

The abundance of AI resources is a thing of the past, and this scarcity is expected to persist for years. Companies must now focus on procurement and margin management to navigate this challenging environment. As energy infrastructure and data center expansions lag behind demand, developers are turning to smaller models and on-premise solutions to bridge the gap.