Meta Commits $50 Billion to a Louisiana Data Center and Hires From AWS

--- headline: "Meta Commits $50 Billion to a Louisiana Data Center and Hires From AWS" category: business story_number: "04" slug: meta-louisiana-data-center-aws-hire date: 2026-07-18 ---

In the space of a single week, Meta has redrawn the map of its ambitions in poured concrete and poached talent. The company confirmed that its Hyperion data center supercluster in rural Richland Parish, Louisiana, will now cost more than $50 billion, nearly double the $27 billion figure disclosed last October, and a person familiar with the plans told Bloomberg the full build-out could ultimately exceed $250 billion. Days later came the hire that gives the numbers a strategy: Dave Brown, one of the most senior infrastructure executives at Amazon Web Services, is leaving to help run Meta's data-center expansion.

Together, the two moves signal that Meta no longer sees itself as merely a builder of models and feeds. It intends to become an owner of compute at a scale that rivals the cloud giants, and possibly a seller of it.

The Hyperion bet

Hyperion is the physical center of gravity. Meta announced it is expanding the Richland Parish campus to 5 gigawatts of compute capacity, up from an original design of 2 gigawatts, across a footprint that will span roughly 10 million square feet. The company says the site should reach 2 gigawatts by 2030, though it has offered no timeline for when the full 5-gigawatt build will be finished.

Delivering that much power to a corner of northeast Louisiana is its own megaproject. As part of the arrangement, Meta agreed to finance power generation and transmission with utility Entergy Louisiana, including seven combined-cycle natural gas plants, grid-scale battery storage at three sites, and roughly 240 miles of new high-voltage transmission lines. The build has been aided by generous state tax incentives, a subsidy that has, by local accounts, split the surrounding community over water, land, and who ultimately benefits.

Hyperion is only one node in a far larger plan. Meta has said it will spend between $125 billion and $145 billion on AI infrastructure this year, and Chief Executive Mark Zuckerberg has committed the company to roughly doubling its compute to 14 gigawatts by 2027. Against that backdrop, a single project that could top $250 billion is not an outlier. It is the template.

The AWS hire

The Brown hire is the tell that Meta is thinking beyond its own workloads. According to a July 15, 2026 memo from AWS Chief Executive Matt Garman, Brown, the senior vice president of AWS Compute, AI, and Platform and a member of Amazon CEO Andy Jassy's advisory group, is departing at the end of July after nearly two decades helping build the world's largest cloud business. He is expected to join Meta in the coming weeks, reporting to the company's head of infrastructure. Dave Treadwell will succeed him at AWS.

Brown's mandate at Meta reportedly centers on two things: building data centers at massive scale, and constructing something internally referred to as Meta Compute. That second item is the one that matters. Meta Compute represents a potential pivot toward renting out AI infrastructure to outside customers, the exact business model Brown spent his career refining at the company that invented it.

The recruitment is pointed. Meta is not hiring a networking specialist or a chip designer; it is hiring the person who ran compute at AWS to help it challenge AWS.

Analysis: from model lab to landlord

The strategic logic ties directly to a report that surfaced the same week. Anthropic is in very preliminary talks to lease roughly $10 billion in computing power from Meta, according to CNBC. If that deal happens, it would place a leading frontier lab in the position of renting capacity from a direct competitor, and it would give Meta Compute an anchor customer before the business formally exists.

That is the shape of the emerging AI economy. Compute is the binding constraint, and any company with spare capacity can sell it, even to rivals it is racing against on models. Meta has committed more capital to AI infrastructure this year than almost any consumer-technology company in history, and building at 14-gigawatt scale all but guarantees it will have excess capacity to monetize during the stretches when its own products cannot absorb it. Hiring a Dave Brown is how a company turns a cost center into a product line.

The risk is equally plain. Owning physical infrastructure is now arguably the most defensible position in AI, because even fierce rivals become customers when they cannot build fast enough themselves. But defensibility is not the same as profitability. Gigawatt-scale campuses are won on electricity, permitting, and land rather than on model quality, and Louisiana offers all three on favorable terms. Spending $250 billion is the easy part. Earning a return on it, either through dramatically better products or a genuinely competitive cloud business, is the hard part, and it will take years to know whether Zuckerberg's bet pays off.

There is also a credibility question buried in the numbers. A project whose cost estimate roughly doubled in nine months, from $27 billion to more than $50 billion, and whose full scope is now pegged near $250 billion by people close to it, is moving faster than any published plan. Investors have so far rewarded the ambition. They will eventually ask about the return.

What to watch next

Three things will tell us whether this is a durable strategy or an expensive detour. First, whether the Anthropic compute talks convert into a signed deal, which would validate Meta Compute as a real business rather than an internal aspiration. Second, how quickly Brown reshapes Meta's infrastructure organization, and whether more AWS and cloud veterans follow him out the door. Third, the physical milestones in Richland Parish: the 2-gigawatt target for 2030, the Entergy gas plants and transmission lines, and whether local resistance over subsidies and resources slows the timeline. If Meta hits its marks, it will have quietly become one of the largest compute landlords on the planet. If it stumbles, it will have built the most expensive infrastructure bet in consumer tech on ground that is still contested.