The shapes of AI partnerships.

Hyperscaler invests in AI lab; AI lab commits massive compute purchase; sometimes paired with chip co-development

A hyperscaler (AWS, Microsoft Azure, Google Cloud, Oracle Cloud) makes a strategic investment in an AI lab, structured around a parallel commitment by the AI lab to purchase massive compute capacity from the hyperscaler over multi-year terms. The investment funds the compute; the compute purchase justifies the investment; the relationship is structurally circular and increasingly common at the frontier-lab scale (Anthropic-AWS $8B, Microsoft-OpenAI $250B Azure commitment in the 2025 restructure, OpenAI-Oracle 4.5 GW Stargate, Anthropic-Google up-to-1M TPUs).

Initial deal grants hyperscaler exclusivity in exchange for investment + capacity; AI lab later restructures to multi-home as compute demand outgrows one provider

An AI lab and a hyperscaler enter into a formation deal that grants the hyperscaler exclusivity or strong preference (API exclusivity, capacity exclusivity, distribution exclusivity, model rights). Over time — typically 18-36 months at frontier scale — the AI lab''s compute demand or strategic posture outgrows what one hyperscaler can supply, and the relationship restructures to allow multi-homing. The original hyperscaler typically loses preferential terms, retains some preferred status, and absorbs the friction cost of the restructure. Microsoft-OpenAI 2019-2025 is the canonical case (Azure API exclusivity, ROFR on compute, $13B+ investment; restructured 2025 with Microsoft losing ROFR and OpenAI committing $250B to Azure but free to multi-home). Anthropic-AWS-then-Google follows the same arc compressed (AWS as primary then Anthropic adding Google TPUs at scale).

Third-party model embedded into an enterprise's commercial product

An enterprise OEMs an AI-enabled product into its commercial offering, where the underlying AI capability is provided by a third-party model provider (Anthropic, OpenAI, Google, a fine-tuned open-weights model, etc.). The most common AI partnership pattern in 2026 because few enterprises train foundation models themselves; most embed models from frontier labs into their own products.