Last reviewed
Jun 2, 2026
Sources
20 citations
Review status
Source-backed
Revision
v1 · 1,670 words
Add missing citations, update stale details, or suggest a clearer explanation.
Prometheus is a roughly 1-gigawatt AI supercluster built by Meta at its data center campus in New Albany, Ohio.[1][2] Announced by chief executive Mark Zuckerberg in July 2025 and expected to come online in 2026, it is intended to be among the first single-site computing facilities to draw more than a gigawatt of power, and Meta describes it as part of a wave of "titan" clusters meant to train the models developed by Meta Superintelligence Labs.[1][3][4] Prometheus is one of two named multi-gigawatt clusters Meta disclosed in 2025, the other being Hyperion in Louisiana, which is planned to scale to 5 gigawatts over several years.[1][3]
Prometheus should not be confused with Project Prometheus, an unrelated "physical AI" startup co-founded by Jeff Bezos in November 2025 that focuses on applying AI to manufacturing and engineering rather than on operating a data center.[5][6]
Prometheus is the internal name for a large AI training cluster that Meta is assembling across multiple data center buildings at its New Albany site, northeast of Columbus.[2][7] The defining characteristic of the project is its electrical scale: Meta expects the facility to be capable of consuming more than one gigawatt of power, a threshold that until the mid-2020s no individual data center had crossed.[1][2] At that scale, a single cluster draws roughly as much electricity as a mid-sized American city.[8]
The cluster was first reported by the semiconductor research firm SemiAnalysis on July 11, 2025, which described a 1-gigawatt cluster called Prometheus in Ohio alongside a larger cluster called Hyperion.[3][9] Zuckerberg publicly confirmed the projects three days later, on July 14, 2025, in a post on his social media accounts.[3][10] SemiAnalysis characterized Prometheus as one of the largest AI training clusters then under construction in the world.[9][11]
Zuckerberg framed Prometheus within a broader strategy of spending heavily on computing infrastructure to pursue artificial superintelligence. In his July 14, 2025 post he wrote that Meta would "invest hundreds of billions of dollars into compute to build superintelligence," and that the first of its large clusters, Prometheus, would come "online in '26."[3][10] He added that a second cluster, Hyperion, would "be able to scale up to 5GW over several years," and that Meta was "building multiple more titan clusters as well," noting that "just one of these covers a significant part of the footprint of Manhattan."[3][10] An accompanying graphic overlaid a planned cluster on a map of New York City to illustrate the comparison.[10]
The announcement followed Meta's reorganization of its AI efforts into Meta Superintelligence Labs, a division that consolidated the company's foundation-model, product, and research teams and recruited senior figures from across the industry, including former Scale AI chief executive Alexandr Wang.[3][12] Prometheus and the other clusters are positioned as the physical infrastructure that will supply this group with what Zuckerberg called "industry-leading levels of compute."[10] The disclosures came amid an industry-wide build-out of frontier training capacity, alongside efforts such as the Stargate Project backed by OpenAI and partners.[11]
Prometheus is located at Meta's New Albany data center campus in central Ohio, with the gigawatt-scale expansion sited at 1500 Beech Road.[2][7] Meta first broke ground in New Albany in 2017 and has described more than $1.5 billion of cumulative investment in the original campus, which the Prometheus expansion substantially enlarges; the company reports the New Albany operation supporting more than 300 operational jobs and around 1,200 skilled-trades workers at peak construction.[13]
| Attribute | Detail |
|---|---|
| Operator | Meta Platforms |
| Location | New Albany, Ohio (1500 Beech Road), near Columbus[2][7] |
| Power capacity | More than 1 gigawatt (1 GW IT power)[1][9] |
| Expected online | 2026[1][2] |
| Configuration | Multiple data center buildings forming one cluster[2][7] |
| Primary purpose | Training Meta Superintelligence Labs models[3][10] |
| First reported | SemiAnalysis, July 11, 2025; confirmed by Zuckerberg July 14, 2025[3][9] |
| Sibling cluster | Hyperion, Richland Parish, Louisiana (scaling toward 5 GW)[1][3] |
To bring capacity online quickly, Meta has used what it calls "rapid deployment structures," widely described in press coverage as tents.[3][14] These are long structures clad in puncture- and weather-resistant fabric over a lightweight aluminum frame, fitted with prefabricated power and cooling modules.[3][14] SemiAnalysis, which first documented the approach, summarized the design priority bluntly: it "isn't about beauty or redundancy," but about "getting compute online fast."[3][9] Notably, the temporary structures forgo backup diesel generators, trading conventional redundancy for speed of deployment.[3][14]
The power strategy for Prometheus relies heavily on behind-the-meter generation built to serve the campus directly rather than drawing the full load from the regional grid. Ohio regulators approved a 200-megawatt natural-gas project, reported as "Socrates South," to supply the site, with two 200-megawatt generation locations planned for an eventual 400 megawatts of on-site capacity, well short of the cluster's roughly 1,000-megawatt requirement.[9][15] To close the remaining gap and meet clean-energy commitments, Meta announced agreements on January 9, 2026 with TerraPower, Oklo, and Vistra; the company said the three deals would support up to 6.6 gigawatts of new and existing clean energy by 2035.[16][17] Under those agreements Oklo is to develop a 1.2-gigawatt power campus in Pike County, Ohio, TerraPower is to build new Natrium nuclear units, and Vistra is to supply more than 2.1 gigawatts from existing nuclear plants.[16][17]
Prometheus is the smaller and earlier of the two named clusters Meta disclosed in 2025. Hyperion, located in Richland Parish, Louisiana, near Holly Ridge, is a separate and far larger project that Zuckerberg said could scale to 5 gigawatts over several years.[1][3][18] Reporting on Hyperion describes an investment on the order of $27 billion, a campus of several thousand acres, multiple new natural-gas plants financed through Entergy Louisiana, and operations ramping later in the decade, with around 2 gigawatts targeted by roughly 2030.[18][19] The two clusters are distinct facilities in different states rather than a single distributed system; Prometheus is the one slated to reach gigawatt scale first.[1][18] Beyond the two named sites, Zuckerberg said Meta was building additional "titan" clusters, each potentially covering a large share of Manhattan's footprint.[3][10]
Prometheus exists to provide training compute for Meta Superintelligence Labs and the company's frontier model program, which includes the open-weight Llama family.[3][10] Concentrating on the order of a gigawatt of computing power in one cluster is meant to support training runs that require large numbers of accelerators interconnected with high bandwidth, the kind of scale Meta argues is necessary to remain competitive with rivals such as OpenAI, Google, and xAI.[9][11] Meta has separately designed its own MTIA inference accelerators, though its largest training clusters have relied predominantly on commercially available GPUs.[20]
If it performs as described, Prometheus would mark a notable step in the scaling of AI infrastructure, being among the first individual data center clusters to cross the one-gigawatt line.[1][2] It also illustrates a shift in how hyperscale operators are willing to build: prioritizing time-to-power through temporary fabric structures and dedicated on-site generation over the multi-year timelines of conventional, fully redundant data centers.[3][14] The naming convention, with Greek mythological figures such as Prometheus, Hyperion, and the reported "Socrates" power project, has become a recognizable shorthand for Meta's 2025-2026 build-out.[3][15]
The scale of Prometheus has drawn local concern in central Ohio over electricity prices, water use, and grid strain. Coverage of the project notes that Prometheus and Hyperion together would consume enough energy to power millions of homes and could pull substantial electricity and water from surrounding communities.[2][8] In Ohio, public pushback has focused on rate cases before the Public Utilities Commission of Ohio involving American Electric Power of Ohio; reporting indicated that an average customer could pay roughly $36 more per month by 2027 across electricity, gas, and water if proposed increases were approved.[2][8] Meta has responded by emphasizing behind-the-meter generation intended to avoid burdening local ratepayers, its clean-energy procurement deals, and a longstanding pledge to restore more water than it consumes globally by 2030.[13][15][16] As of mid-2026 the cluster was still under construction and had not been independently verified at full gigawatt operation, so many specifics rest on company statements and trade reporting rather than completed benchmarks.[2][9]