Meta breaks ground on a 1GW data center in Lebanon, Indiana

A single campus hedged between AI and the ad business

The most telling detail in this announcement is not the $10 billion figure or the 1GW target. It is the framing of purpose. Meta describes the Lebanon site as having "the flexibility to handle both AI workloads and our core products," and repeats the point in the body copy.

As AI advances and compute demands continue to grow, gigawatt sites like this one will be critical to advancing the technology that supports our core business as well as our AI ambitions.Montana Labs

That dual-use language matters. A gigawatt of capacity is a large, expensive, long-lived commitment. By declaring it fungible between recommendation and ranking systems that already generate revenue and the AI training and inference workloads that are still proving themselves, Meta gives itself an out: if AI demand shifts, the same power and floor space keep serving the products that pay the bills today.

The build is described mostly through water and power terms

Read the announcement closely and a striking share of the concrete commitments concern water and electricity, not chips or software. Meta pledges to match 100% of the site's energy use with clean energy, pay the full cost of energy and water, and restore 100% of the water it consumes in Lebanon to local watersheds.

The cooling design is specified: a closed-loop, liquid-cooled system that recirculates the same water and uses "zero water for a majority of the year." Liquid cooling is the physical signature of high-density AI hardware, which runs hotter than the servers that powered earlier data center generations. The choice to name it, and to pair it with a closed loop, is a direct answer to the water-consumption scrutiny that has followed large AI facilities.

Two water-restoration projects are attached: an Arable irrigation partnership in the Upper Wabash River Basin projected to restore 200 million gallons a year for ten years, and a Deer Creek stream revitalization. These are specific, bounded commitments rather than aspirational targets, which makes them checkable.

A twenty-year utility-bill subsidy signals how long these sites run

Among the community commitments, one has an unusually long horizon: $1 million each year for 20 years to the Boone REMC Community Fund for direct energy-bill assistance. Meta also commits more than $120 million toward water infrastructure, roads, transmission lines and utility upgrades.

The two-decade energy subsidy is worth pausing on. It implicitly acknowledges that a gigawatt load reshapes a local grid, and it ties Meta's presence to residents' electricity costs for the operating life of the facility. The workforce program through the Boone County Career Collaborative spans three school districts, again framed as a multi-year pipeline rather than a one-time hire event tied to the roughly 300 permanent operational jobs.

What the Lebanon campus tells applied teams about physical constraints

The specific implication of this announcement is that AI capacity is increasingly a negotiation with power grids, water utilities, and county governments before it is a question of models. Meta's second Indiana site is defined by a gigawatt of power, a closed-loop cooling design, watershed restoration, and a 20-year utility commitment — the language of infrastructure, not of AI research.

For teams building on top of frontier compute, the takeaway is grounding: the throughput available to you traces back to sites like this one, where the binding constraints are siting, electricity, and cooling. And Meta's insistence that the same 1GW can serve either AI or its core products is a reminder that even the largest operators are keeping their capacity bets flexible rather than committing all of it to AI alone.