Modal secures three-year GPU leases across 25 cloud providers

Modal, an AI infrastructure provider, has shifted its capacity strategy from relying primarily on spot and one-year reservations to securing three-plus-year deals with upfront payments. This change is designed to meet anticipated demand through 2027 and hedge against rising GPU prices and scarcity driven by large buyers such as Anthropic and OpenAI. The company now sources capacity from nearly 25 different clouds, including hyperscalers and neoclouds, and has established a dedicated procurement desk with four full-time employees and a 10-person engineering team to manage bare metal capacity, network issues, and hardware failures. Modal does not own hardware, renting all capacity to maintain capital efficiency, though it remains open to ownership or joint ventures if necessary. The company argues that hardware prices will eventually decline, making long-term fixed-price leases a strategic hedge against price volatility. Modal reports achieving four to five nines of uptime and views its multi-region, aggregated capacity pool as a competitive advantage over the rigid, rectangular capacity blocks offered by competitors. The firm noted that the biggest downside of current GPU scarcity is that almost all capacity is sold as a three-year rectangle starting three to six months in the future, limiting flexibility for customers.