Free tools Windows power users keep installed
One-click scans. No signup required.
Some links on this page are affiliate links: if you buy through them we may earn a commission, at no extra cost to you.
Meta is expanding its Richland Parish, Louisiana, data-center campus to support Hyperion, an AI-training cluster planned to scale to 5 gigawatts (GW) of compute capacity. Meta’s July 2026 announcement puts the project’s regional investment above $50 billion. The 5-GW figure is a capacity target—not proof the site already draws 5 GW of electricity continuously. And Zuckerberg’s Manhattan comparison describes the scale of the infrastructure, not a single Manhattan-sized server building.
What Meta is building in Louisiana
Hyperion is Meta’s name for a large AI-training cluster being developed at the company’s data-center campus in Richland Parish, northeast Louisiana. Meta announced on July 13, 2026, that it was expanding the project to 5 GW of compute capacity and expected investment in the region to exceed $50 billion. Meta’s expansion announcement describes the latest target; it does not mean that all of the planned capacity is already built or operating.
A cluster at this scale is not just a room full of servers. It can include many data halls, accelerator servers, high-speed networking, storage, substations, transformers, cooling systems, backup equipment and the site infrastructure that connects them. The project is best understood as a campus-scale computing system, not one conventional warehouse.
Recommended Free Tools
What “5 GW of compute capacity” does—and does not—tell you
One gigawatt is 1,000 megawatts. Five GW is therefore an exceptionally large capacity figure, but it is not a direct count of GPUs, a measure of model size, or a statement about the campus’s electricity use at every moment.
#1 Best Overall
Several different quantities matter at a data center:
- Compute capacity describes the scale of computing infrastructure the project is designed or planned to support. Meta’s announcement uses this term for Hyperion.
- IT load is the electricity delivered to computing equipment such as servers and networking gear.
- Facility demand also includes cooling, power conversion, lighting and other supporting systems.
- Generation and grid capacity describe the power plants, utility connections and other resources available to supply electricity; they are not the same as actual demand.
- Energy use over time is measured in units such as kilowatt-hours or megawatt-hours. A power-capacity figure alone cannot establish annual consumption.
Meta has not defined the 5-GW figure in the cited announcement as a constant electrical draw. It should not be translated directly into “Hyperion uses 5 GW all day” or into a household-equivalent figure without a clear, matching definition of the capacity being measured.
How the project reached 5 GW
The scale has grown in stages. Meta’s earlier plans for Richland Parish described a project expected to deliver more than 2 GW, while Zuckerberg said in 2025 that Hyperion could eventually scale to 5 GW. Meta’s September 2025 engineering overview said Hyperion was expected to begin coming online in 2028. In July 2026, Meta formally announced the expansion to a 5-GW compute-capacity target and said regional investment would exceed $50 billion. Meta’s engineering overview is the source for the earlier schedule; treat 2028 as an expectation stated in 2025, not a guaranteed completion date. The 2026 announcement does not establish a more precise timetable for the expanded project.
Hyperion is not Prometheus
Meta has another major AI cluster, Prometheus, which the company describes as a planned 1-GW system. It is being assembled across multiple data-center buildings, weatherproof tents and adjacent facilities rather than in one isolated building. Hyperion is the larger cluster planned for Richland Parish.
Rank #2
| Cluster | Announced scale | Arrangement | Schedule context |
|---|---|---|---|
| Prometheus | 1 GW | Multiple buildings, tents and adjacent facilities, according to Meta | Meta has described the cluster in engineering updates. |
| Hyperion | Up to 5 GW of compute capacity | Richland Parish, Louisiana, campus | Expansion announced July 2026; a 2025 Meta article said it was expected to begin coming online in 2028. |
These are distinct projects. The 1-GW Prometheus figure should not be added to Hyperion’s target as though it were part of the Louisiana campus. Meta’s engineering articles explain the Prometheus arrangement and its broader infrastructure plans.
What Zuckerberg’s Manhattan comparison means
Zuckerberg’s Manhattan comparison was reported as a way to convey the scale of the planned infrastructure. It is not a verified architectural measurement showing that server halls cover Manhattan, nor does it establish that one building is that large. The primary Meta sources confirm the cluster’s planned capacity but do not independently reproduce the exact comparison.
A campus supporting a multi-gigawatt cluster can require a vast physical footprint because it must accommodate buildings, substations, power and cooling equipment, roads and construction areas as well as computing hardware. But the Manhattan line is a visualization of scale, not a site plan or a precise land-area calculation. Contemporary reporting covered the comparison and the earlier 5-GW discussion.
Quick wins for a faster PC:
Repair Windows errors before they cause bigger problemsFix Now →Fix the driver behind crashes, sound loss and screen glitchesFind Drivers →Why Meta wants so much computing infrastructure
Meta’s infrastructure supports both training and serving AI models. Training requires large fleets of accelerators to work together; serving, or inference, means running models to answer users and power features at scale. Meta says its ambitions include foundation models, Meta AI and AI features across Facebook, Instagram, WhatsApp and Messenger, as well as work by Meta Superintelligence Labs.
At this scale, buying accelerators is only one part of the challenge. A model may be spread across many racks or buildings, so the system must move data quickly between processors, feed and store training data, distribute power, remove heat, schedule workloads and recover when equipment or network components fail. Meta has described clusters containing 24,576 GPUs in earlier infrastructure work, and its engineering material emphasizes the role of networking, storage, power delivery and thermal management. Meta’s GenAI infrastructure overview discusses those system-level demands.
AI hardware also produces concentrated heat. Denser accelerator racks increase the need for specialized cooling—including liquid cooling for some future hardware—and robust power delivery. A failure in one part of a distributed system can interrupt work across a much larger cluster, making monitoring, failure analysis and job recovery essential. Meta’s Prometheus engineering article discusses reliability at several levels, including power distribution and data halls.
Building more of its own infrastructure can give Meta greater control over hardware, networking and deployment schedules, and may reduce exposure to shortages of rented capacity. It also commits the company to enormous capital spending and long lead times. Even companies building private facilities can continue to rely on outside cloud capacity as new sites come online: Meta has also announced agreements involving AMD accelerator supply and CoreWeave capacity. Those deals are separate from Hyperion’s 5-GW figure and do not map directly to a specific campus.
PC Slower Than It Used to Be?
A free scan shows the junk files, broken settings and background clutter dragging Windows down - then fixes them in one click.Free scan · Windows 10 & 11Outdated Drivers Are Slowing You Down
One free scan finds every outdated or missing driver and matches the right update for your exact hardware.Free scan · exact hardware matchPower supply, clean energy and the ratepayer question
Meta says its Louisiana power arrangement with Entergy will support seven natural-gas-fired plants, three grid-scale batteries, nuclear uprates and purchased power. It also says it will help fund up to 2.5 GW of clean and renewable generation and match the site’s energy use with clean and renewable energy. These are company-described plans and commitments, not evidence that every unit of electricity reaching the site at every hour will be renewable.
The distinction matters. Renewable-energy matching and new renewable generation can be part of a company’s procurement strategy, while natural gas may provide dispatchable generation and reliability. The physical mix supplying a grid at a particular time is a separate question. The disclosed natural-gas plans also mean the project should not be described as “carbon-free” on the basis of renewable matching alone.
Meta says it will pay the full costs of the energy, water and related infrastructure used by the site and says its agreement will save Entergy Louisiana customers about $2 billion, in addition to $650 million from a prior agreement. Those are Meta’s claims; a reader should not treat projected savings or protections from rate impacts as independently established without the relevant utility and regulatory documentation. The company’s economic-impact announcement provides its account of the customer-savings claim.
Meta also cites water-infrastructure investment and water-restoration projects, but the cited materials do not provide a verified site water-consumption figure. Without a reliable consumption estimate and a clear accounting boundary, it is not possible to assess the campus’s full water demand from those commitments alone.
What Meta says the project means for local jobs and investment
Meta estimates that the project will support more than 7,500 jobs at peak construction and about 1,000 operational roles. It says Louisiana businesses have received more than $1.6 billion in contracts and that it is investing more than $1 billion in local road, water and wastewater improvements. The company has also described support for Louisiana Delta Community College, scholarships for Richland Parish graduates, schools and local organizations.
These figures help explain the local economic case, but they are Meta-reported estimates and commitments, not an independent assessment of net jobs, tax revenue or long-term community costs. Construction work is not the same as permanent employment, and spending on infrastructure does not by itself show how costs and benefits will be distributed among residents, businesses and public agencies.
What could go wrong
A planned cluster is a long-lived infrastructure bet, and several risks can alter its timetable or economics:
- Grid constraints and delays: transmission, generation, transformers and interconnections can take years to build, even when a company has committed to a project.
- Cost and hardware obsolescence: accelerator technology changes quickly, while data halls, power systems and cooling infrastructure take much longer to plan and construct.
- Energy and environmental trade-offs: a large power requirement can complicate emissions, reliability, water planning and electricity affordability. Matching claims do not settle those questions by themselves.
- Demand uncertainty: AI use may grow, but efficiency gains, changing model designs or business priorities could change how much capacity is needed.
- Execution risk: more compute does not guarantee better models, successful products or a particular financial return for Meta.
Those risks do not make the announcement unreal. They are reasons to distinguish a multi-year buildout target from an operating facility and to watch for updates on construction, power delivery, utilization and independently documented local impacts.
Do these 3 things before closing this tab:
1Repair Windows errors before they cause bigger problems2Scan for outdated or missing drivers - takes under a minute3Clear out junk files and repair common Windows errorsQuick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
