Anthropic is securing dedicated chip supply through deals with semiconductor vendors, with one notable catch: the terms are tied to each vendor's own commercial performance. If a supplier's business weakens, so do the conditions of the deal.
The arrangement is unusual. Standard chip procurement agreements lock in volume and pricing regardless of how the vendor fares. Linking deal terms to vendor health gives Anthropic some protection against getting committed to a partner that can't deliver at scale - but it also means the supply picture could shift if a chip partner hits financial trouble.
The move reflects a core constraint facing every major AI lab right now. Training and serving large language models like Claude requires enormous quantities of specialized compute - primarily GPUs and custom AI accelerators. Companies that rely on spot markets for chips pay unpredictable prices and have no guarantee of availability during high-demand periods. Locked supply agreements let Anthropic plan infrastructure capacity with more confidence.
Claude demand has been climbing. Anthropic launched Claude 3.7 Sonnet in early 2026 with extended thinking capabilities - a mode where the model reasons through problems step by step before giving a final answer - and API usage has grown alongside Claude.ai subscriptions. Without predictable chip supply, that growth eventually runs into a physical ceiling.
The deals don't appear to point to a single vendor, and Anthropic hasn't disclosed the specific suppliers or volume commitments involved. What the structure signals is that Anthropic is trying to diversify its supply chain while keeping flexibility built in - rather than betting entirely on one chip partner at fixed terms for years out.