The bargaining power in AI data centers is shifting. According to a September 7 report by The Information, as cloud providers like Microsoft and Google rush to deploy NVIDIA servers, data center operators such as CoreWeave are gaining leverage, with contract terms starting to favor the supply side.

The report indicates that data center operators are securing more advantageous terms in Service Level Agreements (SLAs) and payment conditions, with extreme clauses being softened. Concurrently, chip manufacturers like NVIDIA and AMD sometimes offer credit guarantees for data center projects for up to 15 years to ensure smooth deployment of their chips, with NVIDIA being particularly aggressive. Furthermore, electricity costs are highly volatile, and the cost of building a 1-gigawatt data center has more than doubled in the past few years. NVIDIA CEO Jensen Huang stated at an Equinix customer conference that hashrate is fragmenting, and the AI world is moving towards decentralization and high distribution. NVIDIA, Equinix, and Together AI have jointly launched a plan to deploy NVIDIA hardware in Equinix data centers to provide open-source AI model inference services for small and medium-sized enterprises.