Data center capacity is extremely limited, with most tenants securing space today in small, fragmented blocks and contracting for 2028 deliveries, according to JLL’s 2026 North America Data Center Midyear report, released Tuesday. 

Demand continues to exceed expectations, with North America roughly doubling net new demand in the first half of 2026 to 25GW,  the report says. Vacancy has remained at 1% for the third consecutive year, despite an unprecedented amount of construction, reflecting the demand for digital services and AI adoption. 

The challenge to procure and deliver new data center capacity is disrupting development pipelines and increasing reliance on existing, stabilized data centers, according to Morningstar. 

Because existing assets already boast contractually reserved capacity, they are largely protected from grid constraints and new interconnection moratoriums, the firm said in a commentary released Aug. 5. 

“Once operational, data centers are substantially protected from grid constraints and new interconnection moratoriums through binding utility agreements that fix capacity rights, allocate infrastructure risk, and defined service priority,” Morningstar said. 

“Facilities that are 10, 15 [or] 20 years old still have in place power and fiber,” Sean Farney, vice president of data center strategy at JLL, told Facilities Dive. “They still have a mechanical plant that has many years of useful life.” As a result, hyperscale, colocation and even enterprise organizations are updating older data centers to support AI computing requirements, he said. 

Although not subject to the same grid constraints facing new developments, existing facilities must deal with power scarcity issues in other ways. Depending on their contractual commitments, existing facilities could still face minimum-load obligations and previously negotiated service terms, according to Morningstar. 

Another attraction is that existing data center clusters create a “self-reinforcing concentration effect” as operators seek proximity to dense fiber routes, cloud availability zones, carrier hotels and deep pools of skilled labor in major metropolitan areas, Morningstar said.

These dynamics are supporting “robust operating performance for stabilized assets,” Morningstar said. Data center rents have increased approximately 9% year over year since 2020, according to JLL. Larger-scale deployments over 20 megawatts are now achieving average rents of approximately $141 per kilowatt hour, excluding energy costs, “reflecting a premium for high-density, hyperscaler-ready capacity,” Morningstar said. 

Despite strong contractual protections, these facilities are not wholly protected from ongoing power scarcity issues, according to Morningstar. “This insulation is not absolute,” the firm said, noting that while these structures improve cost certainty and ESG alignment, they generally “do not displace the utility’s role in physical delivery.” 

Most of the time, local utilities or system operators retain control over transmission, balancing and delivery, meaning that physical supply of power remains tied to grid conditions, it said. For example, as electricity demand from data centers outpaces infrastructure expansion, utilities and regulators are increasingly incorporating curtailment provisions, demand response obligations and load-shedding mechanisms into large-load frameworks that could force operators to reduce or disconnect their power use during grid emergencies.

Morningstar cites the U.S. Department of Energy in January authorizing PJM Interconnection to divert power away from data centers and large industrial factories to residential households during extreme cold snaps to avoid blackouts as well as rules enacted in Texas that require data centers with loads over 75MW to participate in mandatory demand management programs. 

“As a result, stabilized data centers should be viewed as contractually secured but systemically contingent assets — protected from routine grid constraints under existing agreements, yet still exposed to term and tail-risk scenarios involving emergency curtailment, regulatory intervention, or structural changes to electricity market design,” Morningstar said.

As a result of these factors, data center landlords are capturing large rent increases on renewals: up 70% on average since 2020, according to JLL’s midyear report. Most leases signed today carry annual escalations of 3% or more with no concessions, with early-stage tenants continuing to pay a premium, it says. 

Enterprises work to bring data center capabilities on-site

Artificial intelligence is driving more than 50% of new data center capacity, but even markets with attractive energy prices may still be “commercially unviable if substations are full, transmission upgrades are backlogged, interconnection studies are slow or utility operators cannot assure adequate generation capacity during peak usage,” Morningstar said.

While existing data centers may not be able to support liquid cooling that is required to support AI GPU storage compute, they can be upgraded or retrofitted to support this new demand, according to Farney. “This is happening all over the industry at hyperscale [and] co-location and enterprise facilities,” he said. 

In addition to pure-play AI companies, enterprise users are also managing hybrid portfolios of on-premise, colocation and cloud infrastructure, with individual site requirements at a much lower threshold of 500 kW to 3 MW, per JLL’s report. 

“Some enterprises are beginning to retrofit their on-premise data centers, which were largely made empty over the last 10 years, moving compute loads to the cloud,” Farney said. “This means they have some stranded power, which is perfect for building out a test and AI environment,” Farney said. 

These projects also offer the opportunity to reduce long-term spend versus leveraging the cloud for AI initiatives while bringing AI infrastructure close for low latency and having it accessible for on-site company staff, Farney said. 

“Running AI inference in other AI applications in the cloud off-site can be expensive over time. It can also be limited by latency if those requirements exist, so enterprises are working to build their own on-prem AI inference lab environments in … formerly abandoned on-premise data centers,” Farney said. 

The primary challenges in these circumstances are retrofitting mechanical and electrical infrastructure, according to Farney. But if systems have been maintained properly and were adequately sized with high-quality gear, then retrofits are not challenging, he said. 

“Yet even this group is struggling to secure capacity to secure business growth,” JLL says in its report.