Prometheus
AI supercluster campus in New Albany, Ohio
Prometheus is a data centre campus in New Albany, Ohio, operated by Meta for artificial intelligence training. The site hosts hundreds of thousands of GPUs and aims to reach one gigawatt of power capacity at full scale.
- Planned total power capacity of one gigawatt
- Houses hundreds of thousands of GPUs currently
- Uses Nvidia GB200 and GB300 chips
- Trained Meta Muse foundation models
- Features Air-Assisted and Facility Liquid Cooling
What it is
Prometheus is a large-scale artificial intelligence computing facility located in New Albany, Ohio. Operated by Meta, the campus serves as an AI-focused supercluster designed to train foundation models. The site combines traditional data centre structures with rapid deployment buildings to house massive GPU clusters. Meta states that the first phase is already active and supports hundreds of thousands of graphics processing units. The project represents a conversion of existing infrastructure into a specialised AI hub, with plans to expand significantly in capacity and physical footprint.
How it came about
The location originally functioned as a standard Facebook data centre before undergoing conversion. Meta repurposed the existing facilities to create the initial phase of the Prometheus supercluster. This transition allowed the company to deploy AI-specific hardware without building entirely from scratch. The first phase has been used to train Muse foundation models, marking its operational start. Meta continues to expand the site, integrating new hardware generations as they become available for large-scale machine learning tasks.
The building and the halls
The campus consists of multiple interconnected buildings, including traditional data halls and structures built for rapid deployment. These facilities are linked to support a unified computing environment. The design accommodates dense hardware configurations, such as the Ariel variant which pairs GPUs with CPUs in specific rack layouts. Backend networks span distances from two kilometres to over one hundred kilometres to connect building-level fabrics. The infrastructure supports high-bandwidth connections required for distributed AI workloads across the entire site.
Power and cooling
Meta plans for the Prometheus campus to reach a total power capacity of one gigawatt at full scale. The current deployment supports hundreds of thousands of GPUs, requiring substantial electrical infrastructure. Cooling systems include modular Air-Assisted Liquid Cooling and Facility Liquid Cooling options, particularly for newer hardware platforms like the GB300. These liquid solutions help manage the heat generated by high-density computing chips. The energy demand reflects the intensive nature of training large language models and other AI systems.
What runs inside
The primary function of Prometheus is to train artificial intelligence models, specifically Meta’s Muse foundation models. It houses Nvidia GB200 and GB300 chips, along with upcoming Vera Rubin hardware. The system uses complex network topologies to interconnect tens of thousands of GPUs within single buildings. Arista switches and Broadcom silicon provide the necessary bandwidth for these operations. This setup allows researchers and engineers to process vast amounts of data for machine learning applications efficiently.
The place
Situated in New Albany, Ohio, the site benefits from existing infrastructure previously used by Facebook. The suburban location provides space for multiple buildings and extensive network cabling. Ohio offers a strategic position for Meta’s domestic computing needs, allowing for significant expansion. The choice of this town enables the company to convert older facilities while adding new rapid-deployment structures. This approach maximises land use and leverages local power and connectivity resources for large-scale AI operations.
Environment and impact
The conversion to an AI supercluster increases the site's power consumption significantly, with a target of one gigawatt. This scale requires robust cooling solutions to manage thermal output from dense GPU racks. The use of liquid cooling technologies aims to improve efficiency compared to traditional air methods. Local impact includes the continued operation of a major tech facility in the region, supporting Meta’s broader AI development goals. The project highlights the shift in data centre design towards specialised AI hardware.
Network and links
Internal networking relies on Ethernet-based fabrics connecting up to 20,736 GPUs per building. A Backend Aggregation network links these building-level systems over distances exceeding one hundred kilometres. The infrastructure uses Arista 7800-series chassis with Broadcom Jericho3 silicon to handle multi-petabit bandwidth. High-speed NICs provide 800G of input/output per GPU in newer configurations. This topology ensures low latency and high throughput for distributed training tasks across the entire campus.
Questions people ask about Prometheus
Where is the Prometheus data centre located?
The Prometheus campus is located in New Albany, Ohio, in the United States. It was originally a traditional Facebook data centre that has been converted into an AI-focused supercluster.
What hardware does Prometheus use?
The facility houses Nvidia GB200 and GB300 chips, with Vera Rubin hardware arriving soon. It uses specific rack architectures like Ariel for GB200 and Clemente for GB300 deployments.
How large will the Prometheus campus become?
Meta plans to scale the Prometheus campus to a total power capacity of one gigawatt. At full scale, it is expected to comprise roughly one million distributed GPUs across multiple buildings.
What is Prometheus used for?
Prometheus is an AI supercluster used to train foundation models. The first phase has already been used to train Meta’s Muse models, leveraging hundreds of thousands of interconnected GPUs.
Good reading and sources
There is no English Wikipedia article on Prometheus yet, so the facts on this page come from the sources listed above: datacenterdynamics.com/en/news/meta-details-ohio-prometheus-data-cente. Some are the subject's own pages, so its claims about itself are reported as its claims. EveryDatacenter wrote the text in its own words, with AI assistance, and compares it with the sources to make sure no passage is copied.
It may contain mistakes, and figures change over time. Nothing here is investment or buying advice. If something is wrong, tell us. Last refreshed 2026-11-14.