Construction spending is outrunning even the AI hype cycle
New data from ConstructConnect shows US data center construction starts hit $22.3 billion in June 2026, the second-highest monthly total ever recorded, trailing only January's $25.5 billion. Year-to-date starts through June reached $81.5 billion, already surpassing the full 2025 calendar year total of $72.5 billion with six months still to go. For context, 2024's full-year total was roughly $27 billion, meaning 2026 year-to-date spending already runs at roughly three times that pace, a rate of acceleration that has surprised even analysts who have tracked this market closely for years and modeled far more gradual growth curves.
This is not simply a story of more projects breaking ground somewhere in the country. It reflects a fundamental repricing of what it costs to build the infrastructure underpinning the AI buildout, and that repricing eventually shows up in what enterprises pay for cloud and AI compute, regardless of how GPU or memory prices move independently of everything happening on the construction side of the industry. Buyers who focus exclusively on chip pricing when forecasting their cloud costs are missing a large and growing share of what actually drives the bill.
The cost curve is bending sharply upward
The most striking figure in the report is not the volume of starts but the cost per square foot. Median construction costs for data centers built year-to-date in 2026 hit $570 per square foot, up 38 percent from 2025's median of $415. Average costs climbed even more sharply, up 67 percent to $859 per square foot from $514 the prior year. Over a five-year period, median costs have compounded at 17 percent annually and average costs at 22 percent annually, a pace that outstrips general construction inflation by a wide margin.
That gap between median and average growth rates points to a bifurcated market: a smaller number of extremely expensive, high-density AI-optimized facilities are pulling the average well above the typical project, while a larger base of more conventional data centers sees somewhat more moderate cost inflation. Enterprises negotiating colocation or build-to-suit deals should expect pricing to reflect whichever tier of facility they are actually buying into, and should ask providers explicitly which cost benchmark applies to their specific project.
Regional concentration is intensifying
North Carolina, Indiana, Illinois, and Michigan each recorded more than $10 billion in year-to-date data center construction starts, with Virginia and Texas close behind despite the moratorium constraints Texas has recently imposed on new grid connections. This concentration in a handful of states, rather than broader geographic distribution, reflects how tightly power availability and existing grid interconnection now gate where developers can realistically build new capacity at scale, even when land and local tax incentives look attractive elsewhere in the country.
Looking forward, 85 projects are sitting in late-stage preconstruction with planned year-end start dates representing $78.2 billion in additional spending, suggesting the current pace is not a temporary spike but the leading edge of sustained, elevated construction activity through the remainder of 2026 and into 2027. Enterprises planning multi-year capacity needs should assume this concentration persists rather than expecting costs to ease as more states enter the market, since the states currently winning projects hold structural power advantages that newer entrants cannot quickly replicate.
Bigger facilities, fewer of them
The average data center under construction has more than doubled in size since 2022, with Q2 2026 projects averaging nearly 700,000 square feet. Developers are consolidating capacity into fewer, larger facilities to capture economies of scale on cooling, power distribution, and land acquisition costs that do not scale linearly with facility size, a rational response to rising per-square-foot construction costs across nearly every region tracked in the report and a trend that shows no sign of reversing.
This consolidation trend has a direct implication for enterprises evaluating colocation versus hyperscaler cloud strategies. Fewer, larger facilities mean fewer available sites for smaller enterprise colocation deals, since developers increasingly prioritize anchor tenants capable of committing to hundreds of megawatts over customers seeking a few racks or a small dedicated suite within a shared facility. Mid-sized enterprises without hundreds of megawatts of demand should expect fewer colocation options and less negotiating leverage over time.
What rising construction costs mean for cloud pricing
Every dollar added to the cost of building a data center eventually flows through to what hyperscalers and colocation providers charge their customers, whether through direct rack and power pricing or indirectly through cloud compute rates charged across an entire regional footprint spanning many separate facilities. A 38 to 67 percent jump in construction costs over a single year is a meaningful input cost increase that vendors cannot fully absorb through efficiency gains alone, no matter how aggressively they optimize operations elsewhere in the business.
CIOs and CFOs negotiating multi-year cloud or colocation contracts should treat this data as leverage-relevant context: if a provider cites rising infrastructure costs as justification for price increases, the ConstructConnect figures confirm those costs are real and industry-wide, not vendor-specific padding. The right response is not accepting price increases passively but negotiating longer-term capacity commitments now, before construction costs climb further and lock in even higher baseline pricing for the next contract cycle. Locking in current rates for three to five years looks increasingly like the more disciplined move given where this cost curve is headed.
