The map of new capacity keeps moving outward
Microsoft has acquired 600 acres of land in Woodward, Iowa, roughly 30 miles northwest of Des Moines, for approximately $37.4 million, a per-acre price that reflects how much developers are now willing to pay for sites with a credible path to power rather than for the land itself. The site is described as a potential gigawatt-scale campus for cloud computing and data storage rather than AI training workloads specifically. The purchase fits a pattern that has become unmistakable this year, with hyperscalers moving capacity decisions into smaller metro areas and genuinely rural counties for one overriding reason: those locations offer available land and, more importantly, a credible path to power that the most saturated markets can no longer promise on a reasonable timeline.
Digital Realty is pursuing the same logic at larger scale in Kokomo, Indiana, a city that until recently had little profile in the data center industry. The company is planning a 476 acre campus with three single-story buildings designed to deliver approximately 1 gigawatt of IT capacity across roughly 3 million square feet of floor space. What stands out most is the timeline attached to that ambition: Digital Realty is targeting completion in 2033, a full seven years from now, for a project only announced this quarter.
A seven-year timeline is the real headline
A 2033 completion date on a project announced in 2026 functions less as a construction estimate in the traditional sense and more as a quiet acknowledgment that securing a gigawatt of reliable power now takes longer than building the buildings that will eventually consume it. Given the capacity shortfalls already showing up in grid operator auctions in markets like PJM, that kind of extended timeline looks set to become the norm rather than the exception for large campuses, regardless of how much capital a developer is willing to commit upfront to speed things along.
Smaller projects are moving measurably faster precisely because they ask less of the surrounding grid at any one time. Serverfarm topped out the first of four planned buildings at its Clarksville, Arkansas site, an 108 megawatt, two-story facility spanning 300,000 square feet, part of a full campus plan that will eventually reach 432 megawatts and 1.2 million square feet once complete. The incremental, phased approach lets Serverfarm bring capacity online in manageable stages rather than waiting years for a single gigawatt-scale interconnection to clear before anything can open.
Why rural and secondary markets are winning the siting race
None of these three sites, Woodward, Kokomo, or Clarksville, were traditional data center hubs before this wave of announcements, and that absence of history is precisely the appeal driving developers there now. Primary markets like Northern Virginia and parts of the PJM footprint are already showing grid strain, as evidenced by capacity auctions clearing at price caps and missing reliability targets outright. Developers are responding by moving to markets with underutilized transmission capacity and local utilities more willing to negotiate large, dedicated interconnections in exchange for the economic development a multi-hundred-million-dollar campus brings to a smaller community.
This shift carries consequences that extend well beyond where the buildings physically sit. Rural and secondary markets typically have smaller existing technical labor pools, less dense fiber infrastructure, and local governments with less experience negotiating large infrastructure deals than their big-metro counterparts have accumulated over decades. Each of those factors adds friction and risk that never shows up in a press release announcing a land purchase, yet enterprise customers relying on these facilities will eventually feel the effects in service reliability, support response times, and overall operational quality.
What this means if you are planning capacity two to five years out
If your infrastructure roadmap assumes new regional capacity will be available on the timelines hyperscalers have historically delivered, revise that assumption downward now rather than discovering the gap mid-project. A gigawatt-scale project announced today with a 2033 completion target represents a seven-year lead time from announcement to availability, and even phased, smaller projects like Serverfarm's are measuring delivery in multi-year increments tied to power availability rather than construction speed alone.
For workloads that are latency-sensitive or tied to specific geographies, start conversations with providers now about which of their announced projects already have secured power versus which remain speculative land positions awaiting an interconnection decision. The gap between a land purchase and an energized, operational facility has widened enough that treating every announcement as equivalent, available capacity will reliably lead to planning errors that surface at the worst possible time, typically when a workload you expected to deploy next year is still waiting on a utility interconnection agreement that has not even been filed.
The broader signal for enterprise strategy
The consistent thread across Woodward, Kokomo, and Clarksville is that power has become the scarce resource shaping data center development, while land and capital, the traditional constraints, are comparatively abundant right now. That reorders the priority list for any CIO evaluating where their workloads will run over the next five years: prioritize providers and regions with demonstrated, secured power delivery over those simply promising the most square footage or the most aggressive stated timeline.
It also strengthens the case for hybrid and multi-region strategies that do not depend on any single gigawatt-scale campus coming online exactly on schedule. Given how far out these timelines now stretch, diversifying capacity commitments across providers and regions functions as basic risk management against delays that look increasingly structural rather than incidental, a different rationale than the usual argument about avoiding vendor lock-in, though it reaches a similar conclusion.
