The Beginning of Scarcity in AI

GPU rental prices for Nvidia Blackwell chips have surged 48% in just two months, reaching $4.08/hour, while CoreWeave raised prices 20% and extended minimum contracts to three years. OpenAI's CFO has acknowledged making tough trade-offs due to compute constraints, and Anthropic has limited its newest model to roughly 40 organizations. This compute scarcity is ushering in a new era defined by relationship-based selling, pricing that favors well-capitalized companies, slower inference even for paying customers, inflationary GPU costs, and forced diversification toward smaller models or on-premise deployments. The era of abundant AI compute is over, likely for years, as energy infrastructure and data center buildouts lag behind demand.