The fourth expansion at one address
Meta announced its fourth expansion of the Eagle Mountain data center campus near Salt Lake City, Utah, adding new buildings equipped with closed-loop liquid cooling systems backed by dry cooling technology for periods of peak thermal load. The campus originally broke ground in 2018 and went operational in 2021, and each expansion round since has added capacity to the same footprint rather than establishing a new site elsewhere in the region or the country. Total investment at the campus now exceeds 3 billion dollars, a figure that has climbed steadily with each successive round.
The company's public statement framed the expansion in workforce terms as much as capacity terms: more than 1,200 workers at peak construction, and more than 300 permanent jobs once the new buildings come online. That framing is not incidental. Data center operators increasingly lead announcements with job and investment figures because local permitting and community goodwill, not power availability alone, have become the binding constraint on how fast a hyperscaler can add capacity at an existing site.
Why Meta keeps returning to the same address
Expanding an existing, already-permitted campus is meaningfully faster than opening a new site. The interconnection agreements, water rights, zoning approvals, and community relationships are already established, which collapses a timeline that can otherwise stretch years into a process measured in months. For a hyperscaler racing to bring AI training and inference capacity online as fast as possible, that speed advantage outweighs the diversification benefits of spreading new capacity across multiple regions.
The tradeoff is concentration. Four expansion rounds at one campus means Eagle Mountain now represents an outsized share of Meta's Utah and broader Mountain West footprint, all drawing power from the same regional grid and competing for the same local labor and water resources with each new round. A single-site strategy compounds efficiently until it runs into a shared constraint, whether that is grid capacity, water availability, or a local political shift against further expansion, at which point the entire multi-billion dollar investment sits exposed to one jurisdiction's decisions.
The cooling technology choice is a signal too
The decision to build with closed-loop liquid cooling backed by dry cooling, rather than the evaporative cooling systems that consume far more water, reflects the same water-stress pressure showing up across the industry, including the Texas reporting crackdown playing out this same week. Closed-loop systems recirculate coolant rather than continuously drawing and evaporating fresh water, cutting operational water consumption dramatically at the cost of somewhat higher electricity use for mechanical cooling.
That tradeoff, water savings for higher power draw, is becoming the default engineering choice for new hyperscaler capacity in the Mountain West and other water-stressed regions, and it is a useful signal for enterprise buyers evaluating a provider's environmental risk exposure. A facility built on evaporative cooling in a drought-prone state carries meaningfully more regulatory and community risk over its operating life than one built on closed-loop systems, even when both meet current permitting requirements on paper.
What single-site concentration means for capacity planning
For enterprises whose AI or cloud workloads run adjacent to, or compete for capacity with, a hyperscaler's concentrated regional footprint, this pattern carries a practical implication: available capacity, power interconnection queue position, and even skilled labor pricing in that region will increasingly reflect one dominant tenant's expansion schedule rather than a diversified market of competing developers. Businesses trying to build or lease capacity near Salt Lake City are effectively negotiating in a market Meta's expansion pace substantially shapes.
That is not unique to Meta or Utah. The same dynamic plays out around Microsoft's concentration in central Ohio, Google's in Iowa and Oklahoma, and Amazon's in Northern Virginia, where each hyperscaler's repeated expansion of proven sites has similarly reshaped local interconnection queues, land prices, and skilled trades availability for everyone else trying to build nearby. Enterprise buyers evaluating a region for their own capacity should map which hyperscaler already dominates local grid interconnection queues before assuming a competitive, multi-vendor market exists there, since that dominant tenant effectively sets the price and pace for everyone downstream.
The resilience question this raises
Cloud region labels obscure physical concentration. A cloud provider's marketing might describe multiple availability zones within a region as independently resilient, but if those zones draw power from the same regional grid, share the same water source, or sit within the same jurisdiction's regulatory reach, they share more correlated risk than the marketing implies. Meta's repeated expansion of a single physical campus is a clean illustration of how capacity growth and geographic diversification are not the same thing, even when the headline investment figures keep climbing.
CIOs running business continuity and disaster recovery planning against cloud or AI infrastructure providers should ask a more specific question than which region hosts their workloads: which physical campus, which grid interconnection, and which water source. A provider whose capacity growth concentrates repeatedly at one site, however impressive the investment total looks, carries a different risk profile than one spreading equivalent capacity across genuinely independent locations, and that difference rarely shows up until a regional outage, drought, or grid curtailment event forces the question.
The takeaway for vendor risk reviews
Add physical site concentration to the standard vendor risk review alongside financial stability and security posture. Ask hyperscaler and colocation partners directly what share of their capacity in a given region sits on a single campus or grid interconnection, and build contractual or architectural mitigations, such as workload distribution requirements across genuinely separate physical sites, where that concentration is high. This is a five-minute question in a vendor review meeting, and most procurement teams still are not asking it, which means most enterprises still do not actually know the answer for their own critical workloads.
Meta's Eagle Mountain pattern is not a red flag on its own. Repeated expansion of a proven site is a rational, speed-optimized strategy that most hyperscalers pursue to some degree, and it is exactly what a well-run infrastructure organization should do when a site has already cleared permitting and secured community support. It is, however, a reminder that headline capacity and investment figures say very little about geographic and regulatory diversification, and that the two numbers enterprise buyers actually need, physical site concentration and shared infrastructure dependency, rarely appear in the press release at all.
