Meta plans two AI supercenters of up to 5 gigawatts

On July 14, 2025, Mark Zuckerberg described Prometheus and Hyperion as part of Meta's large-cluster buildout. The original source supports the documentary core of the event; announced power is future design capacity, not delivered energy or measured useful compute.

The scale is no mere technical detail. One gigawatt equals 1,000 megawatts of electrical power—a level associated with major industrial facilities and, in some cases, a power plant. Meta wants to turn that energy into the computing capacity needed to train and run its next generation of AI models. Document supporting the figure.

Prometheus will arrive in 2026; Hyperion will scale up later

Prometheus will be located in New Albany, Ohio, and Meta expects it to reach 1 GW of capacity in 2026. Zuckerberg has described it as one of the first AI data centers of that scale controlled by a technology company.

The more ambitious project is Hyperion. It will be located in Louisiana, likely in Richland Parish, where Meta had already announced a $10 billion investment to build a data center. The company expects to bring 2 GW of capacity online there by 2030 and eventually expand the complex to 5 GW. Document supporting the figure.

Zuckerberg has said the site will occupy a substantial portion of an area comparable in size to Manhattan. The comparison illustrates the plan’s physical scale, but it is important to distinguish that from the announced power capacity: Hyperion will not have 5 GW when it opens. That is the upper limit of an expansion planned over several years. Document supporting the figure.

The investment also goes beyond any specific figure for these two projects. The CEO has written that Meta will invest hundreds of billions of dollars in computing capacity to build superintelligence, the term the company uses for systems that would vastly exceed human abilities across many tasks. That is an investment commitment, not a finalized budget for Hyperion and Prometheus.

Computing power becomes a competitive advantage

Frontier AI models require enormous clusters of chips, high-speed networks, storage systems and cooling. This combination is commonly called a cluster; when it reaches exceptional scale, companies refer to it as a supercluster.

Meta is not starting from scratch. The company already develops the open Llama model family and operates one of the industry’s largest data center infrastructures. But its announcement follows a campaign to strengthen Meta Superintelligence Labs, its new advanced AI division. In recent weeks, it hired Alexandr Wang, formerly CEO of Scale AI, and Daniel Gross, co-founder of Safe Superintelligence.

The logic is straightforward: hiring top researchers is easier if they can work with the computing capacity they need. OpenAI, Google DeepMind and Anthropic also depend on ever-growing computing resources, while xAI has developed its Colossus supercomputer and OpenAI is advancing Stargate with Oracle and SoftBank.

This race is changing how the strength of an AI company is measured. Publishing a competitive model is no longer enough: companies must be able to train the next generation and serve it to millions of people without computing costs spiraling out of control.

Electricity and water become part of the AI debate

The other side of these announcements is energy. A data center spanning several gigawatts consumes an amount of electricity comparable to that used by a major urban area. It also needs grid infrastructure, available generation capacity and cooling systems that can require large volumes of water.

The pressure is already visible in some U.S. communities. The New York Times reported Monday that a Meta project in Newton County, Georgia, has coincided with water-supply problems in nearby homes. That alone does not prove all data centers will have the same effect, but it does foreshadow the kind of local conflict facilities of this scale can trigger.

The U.S. Department of Energy estimated in late 2024 that data centers could account for between 6.7% and 12% of national electricity consumption in 2028, up from 2.5% in 2022. AI is one reason for that increase, alongside the expansion of cloud computing. Document supporting the figure.

The Trump administration has backed the rollout of AI infrastructure and supports expanding power generation through sources including nuclear, natural gas, geothermal energy and coal. For Meta, securing energy will be as critical as buying chips. For Louisiana and Ohio, the question will be under what conditions that investment arrives: who pays for grid upgrades, where the electricity comes from and how communities’ access to water is protected.

Prometheus will be the first tangible test in 2026. Hyperion represents a larger ambition: a single Meta complex that could grow to a level requiring AI to be planned as a matter of industrial and energy policy, not just software.

Turning the headline into a check

Nameplate power is not useful compute. Between a grid connection and a running model sit substations, cooling, networks, memory, storage, software and equipment availability. announced power is future design capacity, not delivered energy or measured useful compute. An announcement becomes capacity only when every link has a date, owner and measurement.

System comparisons require a fixed workload. Training, tuning and inference use memory, communication and precision differently. A vendor maximum may depend on formats, software or models that do not match real work. To assess how to read a data center as a chain of power, network, cooling, utilization and date, run the same set with versions and consumption recorded.

Utilization reveals the distance between inventory and result. Installed equipment may wait for data, networking or repairs. A useful record tracks available hours, completed work, failures, total energy and delays. It also separates an operating first phase from announced final scale; mixing them turns the future into the present.

What the record must preserve

Physical cost is measured, not inferred from one magnitude. Electricity source and timing, cooling, water, construction, redundancy and service life all matter. An assessment should publish its boundary and avoid vague offsets. That inventory enables comparison and asks what capability is obtained for the resource consumed.

An evidence sheet separates four columns: what the source claims, what it shows, what it did not measure and what would change the conclusion. That discipline prevents an absence from becoming a promise and a condition from vanishing in summary. It also lets the story be updated without rewriting history from a later outcome.

Include a negative case before deciding. Find a situation where the system, rule, transaction or study does not meet the need and record the signal that would require stopping. Selected successes show that something can happen; the negative case reveals the boundary and lowers the cost of discovering it after deployment.

The skill that outlasts the announcement

A valid comparison preserves denominator and axis. It does not pit a point figure against an average, future capacity against installed capacity or a forecast against an observation. When two sources use similar language, reconstruct what they counted and over what period. If those differ, publish them as different measures instead of inventing a ranking.

The record should survive a version change. Keep URL, consultation date, document, configuration and decision. When new evidence appears, add it with its date and explain what it changes. That traceability prevents opposite errors: keeping an expired conclusion or pretending later information was known on the event date.

The transferable skill in this story is how to read a data center as a chain of power, network, cooling, utilization and date. The procedure is short: name the document, preserve the date, fix the axis, find the condition and design a check that can fail. With those steps, a reader need not accept or reject the announcement by intuition; the decision follows a visible chain of evidence.

Before closing, another person should be able to reconstruct the conclusion without knowing the headline. Give them the sources, conditions and negative case, then ask what they would accept and reject. If they need an assumed intent, a figure without a denominator or an undated later fact, the chain still has a gap. That short review catches errors that fluent prose can conceal.