The numbers behind the boom
ConstructConnect's latest data center report puts June 2026 construction starts at 22.3 billion dollars, the second highest monthly total the firm has recorded, trailing only January's 25.5 billion dollars. Year to date spending through June hit 81.5 billion dollars, already well above all of 2025's 72.5 billion dollar total and more than three times 2024's roughly 27 billion dollars. The pipeline behind those numbers is not thinning either: 85 projects representing 78.2 billion dollars sit in preconstruction, with 23 new groundbreakings in June alone and 116 for the year.
The geography of the buildout has shifted from the traditional hubs. North Carolina, Indiana, Illinois, and Michigan have each passed 10 billion dollars in year to date starts, with Virginia and Texas close behind. The preconstruction pipeline points further afield still, led by Texas, North Carolina, Virginia, Arkansas, and Utah. Developers are chasing power availability and permitting speed more than proximity to existing tech hubs, and that migration is reshaping which local governments are suddenly negotiating with hyperscaler capital for the first time.
Where the cost inflation is coming from
The more consequential number in the report is the cost per square foot, which matters more over time than the topline spend figure that grabs the headline. Average construction cost is up 67 percent year over year and median cost is up 38 percent, a five year compounded annual increase of 22 percent on the average basis and 17 percent on the median. Michael Guckes, ConstructConnect's chief economist, points to a related structural shift: the average data center has more than doubled in size since 2022, now sitting near 700,000 square feet, which changes the engineering, electrical, and mechanical complexity of a typical project rather than simply scaling it up.
Larger facilities require denser power delivery, more sophisticated cooling, and structural specifications that a 2022 era data center design never anticipated. Add persistent shortages of transformers, switchgear, and skilled electrical labor, all chasing the same finite supply chain that utilities and industrial projects are also drawing from, and the cost curve stops looking like inflation and starts looking like a genuine capacity constraint on how fast new supply can come online at any price.
This is a pricing problem, not a developer problem
Enterprise buyers tend to treat construction cost inflation as someone else's balance sheet issue, relevant to REITs and hyperscaler capex guidance but not to their own procurement conversations. That is a mistake. Colocation leases, reserved instance pricing, and long term committed use discounts are all ultimately underwritten against the cost of building and operating the capacity behind them. A 38 to 67 percent jump in build cost gets amortized into the pricing of whatever gets built next, typically with a lag of twelve to eighteen months as new capacity comes online and old fixed rate contracts roll off, and that lag is the only reason today's pricing does not already reflect it in full.
That lag is the window that matters for anyone negotiating a multi year cloud or colocation commitment right now. Contracts signed today are being priced against a cost base that is materially cheaper than what a developer breaking ground this quarter will actually pay to deliver the same square footage in 2027 or 2028. Buyers who lock in longer commitments now are effectively purchasing insurance against a cost curve that shows no sign of flattening.
The build versus buy calculus just moved
For enterprises still weighing owned or leased data center capacity against hyperscaler cloud for specific workloads, rising construction costs push the math further toward cloud and colocation for anything that is not a genuinely long duration, steady state workload. Building your own facility now means absorbing 2026's inflated cost basis directly, with no ability to spread it across a hyperscaler's diversified customer base the way cloud pricing does. That calculus was already difficult to justify outside of regulatory or latency edge cases; it is harder still today.
It does mean the opposite of what many boards assume, however: cloud is not becoming permanently cheaper. It is becoming a mechanism for spreading rapidly rising infrastructure costs across a large customer base rather than a way to avoid those costs altogether. CTOs should expect renewal cycles over the next eighteen months to carry price increases tied directly to this construction cost data, and should be modeling that now rather than treating the next vendor renewal as a surprise negotiation.
What to do with this before the next renewal
Finance and infrastructure teams should pull current cloud and colocation contracts and check how much runway remains before renewal, then model a 20 to 40 percent step up in unit costs against that timeline, informed by the median and average increases in this report. Where multi year lock ins are available at current pricing, they are worth evaluating seriously even at a premium to month to month rates, given the direction of the underlying cost curve. Vendors with the newest, most efficient facilities are the ones best positioned to absorb some of this cost increase rather than passing it through in full, which makes facility age and efficiency a legitimate question to raise directly in the next renewal conversation rather than an afterthought left to the account team.
The broader lesson is that infrastructure cost is now a genuine strategic variable again, after nearly a decade where cloud pricing mostly trended down or flat. Treating construction cost data as background noise for real estate analysts, rather than an input to procurement strategy, is no longer a defensible position for a technology leader managing a material cloud spend. The finance teams that built their cloud forecasting models on a decade of falling unit costs need to rebuild those models around a different assumption, and the data center construction numbers behind this report are the clearest early signal of exactly how much that assumption is about to change.
