Meta expands Louisiana AI data center to 5GW, $50B+ investment

Meta's Richland Parish project has quietly become one of the most expensive single data centers ever attempted. When construction began in December 2024, the price tag was around $10 billion. By October 2025, when Meta and Blue Owl Capital formed their joint venture, it had climbed to $27 billion ^[3]. On July 13, 2026, Meta pushed the disclosed figure past $50 billion and lifted planned capacity from 2 gigawatts to 5 gigawatts ^[1]. Louisiana officials now describe the Hyperion campus as one of the largest data centers in history, spanning nearly 10 million square feet across as many as nine buildings ^[2].

But the word 'disclosed' is doing a lot of work here. Bloomberg's reporting, echoed across the financial commentary around the announcement, puts the full lifetime cost closer to $250 billion once the AI chips packed inside the buildings are counted - a figure Meta has pointedly declined to confirm. The distance between the $50 billion Meta will say out loud and the roughly $250 billion the site may ultimately consume is the actual headline. This is a wager large enough that Meta would rather the press release lead with teacher bonuses than with totals.

The more revealing detail is not the number but how Meta is paying for it. Rather than carry the full cost on its balance sheet, Meta sold 80% of the campus to Blue Owl Capital through a $27 billion joint venture, keeping only about 20% for itself and taking a roughly $3 billion one-time distribution while Blue Owl contributed around $7 billion in cash ^[3]. The company then leases the facility back and runs it. In practice, one of the world's most cash-rich firms - Meta earns tens of billions in profit per quarter - structured its flagship AI project so most of the capital, and most of the downside, sits with an outside asset manager and its bond investors.

That structure is what has drawn regulatory fire. Critics argue the arrangement gives Meta an option to exit the project years earlier than a traditional owner would, precisely because it does not hold the majority stake ^[4]. For a company selling the expansion as a multi-decade commitment to Louisiana, the financing quietly says something different: keep the upside, offload the risk, and preserve the freedom to walk.

The financing risk becomes concrete at the power plant. To feed a 5 gigawatt campus, Entergy Louisiana is building a fleet of new combined-cycle natural gas plants, grid-scale batteries, and hundreds of miles of transmission ^[2]. Those gas plants are built to run for roughly three decades. Meta's financing, critics say, could let it step away in a fraction of that time - and more than $500 million in transmission costs fall on all Entergy customers regardless of whether Meta stays ^[4].

That mismatch is the heart of the opposition. Earthjustice attorney Susan Stevens Miller warned that novel financing moves let a trillion-dollar company shift the long-term cost of fossil-fuel infrastructure onto ratepayers, and that regulators have a duty to scrutinize it ^[4]. The Alliance for Affordable Energy called the deal not in the public interest, and the Union of Concerned Scientists accused Meta of operating from the shadows by declining to participate formally at the Public Service Commission. In February 2026, the commission declined to open an investigation anyway ^[4]- a decision that, more than any single number, signals how much leverage a hyperscaler now holds over a state's energy policy.

Here is the contradiction sitting inside the announcement. Only weeks before confirming a 5x expansion of a single campus, Meta was floating the idea that it might rent out excess AI compute to outside customers - a business the company is reportedly building under the banner of a cloud unit that would compete with AWS, Azure, and Google Cloud ^[5]. A firm does not usually 5x a megaproject and simultaneously advertise that it may have too much capacity to use itself.

One reading is confidence: Meta believes demand for AI compute is so deep that any surplus will find a buyer at a premium, echoing the model SpaceX uses for short-term, high-margin capacity. Bloomberg Intelligence has argued that once a company commits well over $100 billion a year to infrastructure, selling what it does not use stops being a side project and becomes a necessity, and estimates such a business could bring in more than $50 billion by 2028 ^[6]. The more skeptical reading is that the willingness to sublease at all is a tell - a company that was certain it would consume every gigawatt would not be workshopping a rental business before the concrete is dry. The market has been split accordingly: the compute-rental report lifted Meta's stock more than 10%, even as the shares have lagged big-tech peers over the past year on capex anxiety.

Meta clearly anticipated the fight. Its announcement was engineered for the constituency that decides whether data centers get built at all - local communities - and led with a headline about teachers and local businesses winning, quadrupled teacher bonuses to $50,000, a $5 million community-college gift, and testimonials from small-business owners ^[1]. With hundreds of cities and states now weighing data center bans, that goodwill is a strategic asset, not a nicety.

Online reception has been far less charitable. On Reddit, the loudest threads treated the subsidy math as the scandal: users repeatedly cited multi-billion-dollar tax abatements flowing to one of the richest companies on earth, and pointed out that the legislation underpinning the incentives requires only a small floor of permanent jobs to keep the benefits. Investor forums were more cynical still, reading the sublease-the-extra framing as a tell about the underlying economics. The gap between Meta's polished community narrative and the skeptical public conversation is itself the risk to watch: the project's social license, more than its engineering, is what remains genuinely contested.