North American colocation vacancy hit a record low of 1 percent in 2026, and 92 percent of the capacity under construction was already pre-leased before a single cabinet went in.
North American colocation vacancy hit a record low of 1 percent in 2026, and 92 percent of the capacity currently under construction was pre-leased before a single cabinet went in, according to JLL's 2026 Global Data Center Market Outlook. Buyers who once planned six to twelve months ahead are now signing 18 to 24 months out. Colocation, the business of renting space, power, and cooling inside someone else's building, was supposed to be the boring middle child of cloud infrastructure. In 2026, it is one of the tightest markets in enterprise technology.
The reason is AI, and specifically the mismatch between how fast GPU demand is growing and how slowly power gets built. Hyperscalers can build their own data centers, but even they cannot always get grid power fast enough, so they lease colocation capacity to bridge the gap. Enterprises deploying AI workloads without the balance sheet for a build to suit campus lease it outright. Neoclouds renting GPU fleets to third parties need somewhere to put the racks. All three buyer types are converging on the same scarce resource at the same time.
The global data center colocation market is projected to grow 18.6 percent year over year to reach 128.6 billion dollars in 2026, up from a 16.8 percent compound annual growth rate between 2021 and 2025, according to a Q2 2026 Databook update from ResearchAndMarkets. The firm expects the market to more than double again, from 108.4 billion dollars in 2025 to roughly 217.0 billion dollars by 2030. In the United States alone, colocation revenue is expected to grow 16.5 percent in 2026 to reach 46.84 billion dollars, on its way to 72.37 billion dollars by 2030, per the same research series.
A separate US focused forecast from ResearchAndMarkets, published in September 2026, put the 2025 US colocation market at 43.71 billion dollars, growing to 85.18 billion dollars by 2031 at an 11.76 percent compound annual rate. The two estimates diverge because they scope the market differently (retail versus retail plus wholesale, and different forecast windows), but they agree on direction: double digit growth sustained for at least five more years, with AI cited as the primary driver in both.
Zoom into the AI specific slice and the growth rate gets steeper. MarketsandMarkets projects the colocation segment of the AI data center market will post the fastest compound annual growth rate of any segment tracked, approximately 32.6 percent, as enterprises outsource AI infrastructure rather than build it themselves. Cushman and Wakefield estimates AI will generate 75 billion dollars in data center demand by 2028, equal to 35 percent of the entire colocation market. Capacity dedicated to AI workloads specifically is expected to nearly triple over the next five to six years, according to a September 2026 ResearchAndMarkets outlook on the US colocation sector.
The public colocation operators are not just forecasting this growth, they are reporting it. Digital Realty closed 2025 with 6.112 billion dollars in full year revenue and 3.34 billion dollars in adjusted EBITDA, and guided 2026 revenue to 6.6 to 6.7 billion dollars. Its fourth quarter of 2025 alone brought in 1.6 billion dollars, up 14 percent year over year, driven by what the company called record leasing across its 0 to 1 megawatt plus interconnection offering. In April 2026 the company signed a 200 megawatt AI lease in Charlotte, one of the largest single colocation commitments on record.
Equinix, which reported roughly 9.9 billion dollars in annual revenue, is playing a different angle on the same trend. Rather than compete purely on megawatts, the company leans on its interconnection business: the software defined network layer that lets AI tenants stitch together multi region inference deployments. In its Q1 2026 results, Equinix said eight of the top ten AI model providers and four of the top five neoclouds, including CoreWeave and Nebius, were expanding on its platform, with Fabric connections tied to AI deals tripling year over year and Fabric revenue up 26 percent. That interconnection moat is a meaningful part of why Equinix sustains a reported 51 percent EBITDA margin, well above the wholesale colocation peers that are mostly renting bare capacity. It is a sharper turn than the one CoreWeave's own revenue miss earlier this year suggested for AI infrastructure spending broadly.
Iron Mountain, the smaller of the three major public colocation operators, held 488 megawatts across 31 facilities in 21 markets at a 96.9 percent occupancy rate as of its Q4 2025 report, with roughly 200 additional megawatts coming online within 18 months and 400 more within 24. Applied Digital, a newer entrant focused on AI specific builds, signed a 7.5 billion dollar hyperscaler deal in April 2026 for its Delta Forge 1 site. Across all four companies, the pattern is consistent: larger commitments, longer terms, with 15 year leases now closer to the norm than the exception, and tenants accepting pre leasing two to three years before a facility even delivers.
Colocation's bottleneck is not real estate, it is electricity. JLL's 2026 outlook reports that average grid connection wait times in primary data center markets now exceed four years, and that power availability, not floor space, has become the defining competitive variable in site selection. In Northern Virginia, the largest colocation market in the world by installed capacity, available powered shell space had fallen below 5 percent as of early 2025, and the PJM Interconnection queue that covers the region runs backlogs into 2030 in some areas.
That scarcity is redirecting capital toward secondary markets. According to the September 2026 ResearchAndMarkets outlook, the Southeastern United States, led by Virginia, Georgia, and North Carolina, attracted more than 15.04 billion dollars in colocation investment in 2025 and is projected to add roughly 13,628 megawatts of new power capacity between 2026 and 2031. The Southwest, led by Texas and Arizona, drew 14.95 billion dollars over the same period, partly because ERCOT offers faster interconnection timelines than PJM, even though it carries its own grid reliability risks. The Midwest trailed at 6.07 billion dollars, and even the New York and New Jersey market, a mature financial services hub, has seen vacancy fall below 6 percent.
The same power constraint is playing out globally. In the Asia Pacific region, JLL projects data center capacity will expand from 32 gigawatts to 57 gigawatts by 2030, a 12 percent compound annual growth rate, with colocation growing faster than the region overall at 19 percent while on premises capacity actually shrinks by 6 percent as enterprises continue migrating to cloud and colocation providers. Nearly 80 percent of colocation capacity currently under construction worldwide is already pre leased, pushing buyers to lock in space two to four years ahead of delivery in tier 1 markets like Ashburn, Silicon Valley, Phoenix, and Singapore, where usable capacity is close to fully absorbed.
The dollar figures behind this demand are large enough to explain the leasing pace. Citigroup has forecast more than 2.8 trillion dollars in AI related infrastructure investment through 2029, and estimates hyperscalers alone will spend about 490 billion dollars on AI infrastructure by the end of 2026. IDC has previously estimated that roughly 20 percent of global data center capacity is already dedicated to AI workloads, a share that is only rising as inference traffic, not just training runs, becomes the dominant driver of compute demand. Inference is a different kind of tenant than training: it needs to sit close to users for latency, which means GPU capacity increasingly needs to be geographically distributed rather than concentrated in one campus, and that distribution requirement favors colocation providers with points of presence in many metros over a single hyperscale build.
This is not purely a real estate story for DataStorage.com readers. Every colocation lease is also a storage decision. GPU capacity is scarce enough that enterprises are routinely running workloads across AWS today, a neocloud tomorrow, and a different colocation facility next quarter depending on where capacity clears first. Every one of those moves drags data along with it, and egress fees on large datasets can erase the savings that motivated the move in the first place. That is the same architecture problem raised on the DataStorage.com Podcast, where Massed Compute founder and CTO Sunny Smith argued that storage has to sit physically adjacent to GPU capacity with high east west bandwidth, because moving a petabyte of data between facilities is prohibitively expensive and slow. As colocation becomes the default home for AI compute rather than a stopgap, storage adjacency and egress free architecture stop being a nice to have and become a prerequisite for keeping AI workloads portable across providers, a point explored further in Why Storage Is the Anchor of the AI Infrastructure Stack.
For infrastructure teams evaluating colocation as part of an AI deployment plan, three things follow directly from the data above. First, lock in capacity earlier than instinct suggests. With 92 percent of North American capacity under construction already pre leased and primary markets running at roughly 1 percent vacancy, waiting for a better price is increasingly a bet against physics, not negotiation. Second, weight power delivery timelines as heavily as price per kilowatt when comparing sites, since a facility quoting a lower rate in a market with a four year interconnection queue is not actually available on any timeline that matters for a 2026 or 2027 deployment. Third, treat data portability as a design requirement from day one, not a migration project for later, because the same power scarcity pushing colocation demand up is also pushing buyers toward multi provider strategies, and a storage layer with egress fees attached makes every future move more expensive than it needs to be.
| Market segment | 2025 size | 2026 projected | Growth driver |
|---|---|---|---|
| Global colocation | $108.4B | $128.6B, up 18.6% | AI and GPU demand, hyperscaler build out |
| US colocation, Databook series | $40.21B | $46.84B, up 16.5% | AI infrastructure, hybrid cloud adoption |
| US colocation, Outlook series | $43.71B | On pace for $85.18B by 2031 | AI, hyperscale expansion, high density compute |
| AI specific colocation segment | New segment | Fastest at 32.6% CAGR | Enterprise outsourcing of AI infrastructure |
The question buyers asked in 2022 was whether a colocation provider had space. The question in 2026 is whether it has power, and when it can deliver.
Compare AWS, Google Cloud, Azure, and alternatives like Backblaze B2 Discover how much you could save in seconds
