Epoch scales AI data center research to cover 44% of global compute

Meta: Meta deploys large AI data centers like Prometheus and focuses on internal use for model training, research, and product features such as recommendation systems. It owns a substantial portion of frontier-scale AI compute. Epoch reports strong coverage of Meta's AI compute compared with other hyperscalers. Epoch: Epoch is an AI research organization focused on tracking and analyzing trends in AI compute, chip supply chains, and infrastructure development. Its work includes public databases on AI chip sales, ownership, usage, and data center construction to inform research on AI progress and societal impacts. The current announcement details Epoch's expansion of its AI Data Centers database to provide broader global coverage and detailed facility-level insights. Amazon: Amazon builds and operates data centers featuring its custom Trainium and Inferentia chips while providing cloud compute services to AI developers including Anthropic. It is one of the largest hyperscalers contributing to global AI infrastructure. The news highlights Amazon-linked sites like the New Carlisle facility and notes that much of its public-cloud compute remains outside current Epoch coverage. Google: Google develops custom AI accelerators like TPUs and operates large-scale data centers that support both internal AI efforts such as DeepMind and external cloud customers. It ranks among the major owners of AI compute infrastructure globally. Epoch's database includes several Google data centers, such as those in New Albany and Pryor, as part of its growing coverage of AI facilities. Nvidia: Nvidia designs and manufactures leading AI accelerators, including the widely referenced H100 GPU used as a benchmark for compute capacity measurements. Its chips power a significant portion of AI training and inference workloads across hyperscalers and frontier labs. In this news, Nvidia's H100 serves as the standard unit for Epoch's coverage estimates and data center compute metrics. SpaceXAI: SpaceXAI, also associated with xAI, develops and operates some of the world's largest AI data centers, including Colossus 2 in Memphis. It focuses on frontier model development with owned compute infrastructure. Epoch identifies Colossus 2 as the top-ranked facility by both power capacity and compute in its database. Anthropic: Anthropic is a frontier AI developer that primarily relies on rented cloud compute from providers like Amazon and Google for training and inference. It appears as a user in multiple large data center projects tracked by Epoch. The news references Anthropic-Amazon facilities as examples of covered sites. Microsoft: Microsoft operates major AI data centers such as Fairwater Atlanta and supplies cloud compute to partners including OpenAI. It is a leading hyperscaler with substantial investments in AI hardware and facilities. Epoch covers Microsoft sites and observes that a significant share of its compute, often serving cloud customers, is not yet fully captured in the database. Coverage Growth: Epoch has steadily increased the proportion of global AI compute captured in its data center database by incorporating more facilities of varying sizes and locations. Supply Chain Mapping: Epoch's data center work complements its related projects tracking AI chip sales, ownership, component consumption, and usage by major developers to provide a fuller picture of AI compute distribution. Infrastructure Trends: AI data center construction is expanding beyond the largest frontier sites to include a wider range of scales and additional countries outside the primary US concentration.