Meta's El Paso Data Center Grew From $1.5B to $10B in Five Months

A groundbreaking that was rewritten in place

The same page carries two versions of the same project. The October 15, 2025 groundbreaking described a facility with an initial investment of more than $1.5 billion, roughly 100 operational jobs, and 1,800 construction workers onsite at peak. It was described as able to "scale to 1GW."

The March 26, 2026 update on that same page tells a materially different story: the investment climbed to more than $10 billion, operational jobs to more than 300, and peak construction to more than 4,000. The 1GW figure is no longer a ceiling the site "can scale to" but a size the data center "will now officially grow to."

That is a roughly sixfold-to-sevenfold increase in stated capital, tripled operational headcount, and doubled construction workforce, all within five months of breaking ground. The speed of the revision is itself a data point about how volatile AI infrastructure planning currently is.

Designing for hardware Meta doesn't have yet

Meta is explicit that it does not know exactly what it is building for. The announcement states that AI "and its inference and training needs, is still evolving," and that "different AI configurations will require different approaches to hardware and network systems designs."

The practical response is a facility engineered for flexibility. Meta says the El Paso design has "systems that can support both the traditional servers of today and future generations of AI-enabled hardware." In other words, the building is a hedge against not knowing which accelerators, racks, or cooling densities will win.

AI, and its inference and training needs, is still evolving, so our design needs to balance what we know today with what we might know in the future.Montana Labs

This is worth noting for anyone modeling AI economics: the physical plant is being provisioned as an option on future hardware, not a fixed spec. The investment jump partly reflects that a 1GW site sized for next-generation AI hardware costs far more than an initial phase built around today's servers.

Water and grid claims that carry real numbers

El Paso is a water-constrained desert city, and Meta's environmental commitments are unusually concrete here. The company plans a closed-loop, liquid-cooled system that "recirculates the same water so that there is no operational water use for a majority of the year," and commits to restoring 200% of the water the data center consumes to local watersheds.

Specific partnerships back the claim: DigDeep will bring clean running water to more than 100 El Paso homes; the Bonneville Environmental Foundation project targets irrigation technology for local farmers; and a $25,000 AguaCares grant is aimed at helping roughly 100 families with water bills. A $500,000 grant funds a workforce program with El Paso Public Schools.

On power, Meta says it has projects under contract adding more than 5,000 megawatts of clean energy to the Texas grid, and that it paid for new transmission lines and substations to connect the site through El Paso Electric. That last detail matters: a single site now requires purpose-built grid infrastructure, not just an interconnection request.

What a septupled budget signals for anyone planning around Meta's platform

For teams building on Meta's models and products, the El Paso revisions are a signal about supply. Meta ties this facility directly to delivering "top-tier AI models" and to consumer products it names: smart glasses, AI assistants, live translations, and video editing tools. This is the compute that those features run on.

The willingness to revise a groundbreaking-stage project from $1.5 billion to more than $10 billion, mid-construction, suggests Meta expects demand for training and inference capacity to keep outrunning its own recent forecasts. That has a flip side worth watching: capacity built as an option on uncertain future hardware is capacity whose per-unit economics won't be clear until the hardware ships.

The specific implication of El Paso is that Meta is now willing to lock in decade-scale local commitments — water restoration to 2030, dedicated grid infrastructure, LEED Gold buildings, community grants — as the price of siting AI compute at gigawatt scale. Those obligations are not optional add-ons; they are becoming a fixed cost of building where the power and the workforce are.