JLL Says Data Center Scarcity Is Raising the Value of Existing Capacity

JLL says data center scarcity has reached a decisive point, despite years of construction driven by cloud computing and artificial intelligence. Limited power, delayed equipment, and preleased capacity are making operational facilities more valuable than undeveloped sites.

The reversal matters because the industry has focused heavily on new hyperscale campuses. Existing facilities were often treated as aging infrastructure that would struggle with high-density AI systems. Now, a working grid connection and a credible upgrade path can outweigh the advantages of a new building.

A Facilities Dive report surfaced through Google News captures this shift. JLL’s research describes a market where available capacity is scarce, construction delays remain common, and enterprises must reserve space well before deployment. Morningstar adds an important qualification: many older facilities still need expensive technical changes before they can run dense AI workloads.

The result is not a simple boom for every data center owner. Powered buildings are gaining strategic value, but their cooling systems, electrical distribution, and rack designs determine whether that value can support modern computing. The market is rewarding usable capacity, not buildings alone.

Power and Equipment Delays Have Changed the Market

The scarce resource is no longer an empty building or a promising parcel. It is dependable electrical capacity that can reach servers on schedule.

JLL’s global outlook estimates that worldwide data center capacity will grow from 103 gigawatts in 2025 to 200 gigawatts in 2030. That represents a 14% compound annual growth rate.

Yet the same report shows why developers cannot translate demand directly into completed facilities. Grid connection lead times for new 50-megawatt projects stretch across several years in many major markets. Dallas faces a pause extending into 2028, according to JLL’s market comparison.

A grid connection is the utility approval and physical infrastructure needed to deliver power to a site. Without it, a finished shell cannot operate its computing equipment. Land ownership and construction progress provide little protection when the necessary electricity remains unavailable.

Equipment has become another constraint. JLL reports a global average lead time of 33 weeks for critical data center equipment. That remains 50% above pre-2020 levels. The United States averaged 42 weeks, which was 83% above its 2019 level.

Transformers, generators, and switchgear sit between utility supply and computing hardware. They convert, route, and protect electricity before it reaches server racks. A delay involving any of these components can hold back an otherwise completed project.

JLL found that 57% of projects experienced delays lasting at least three months during 2025. Large operators have responded by keeping six to 12 months of strategic inventory for essential components. Smaller developers and enterprise buyers often lack the purchasing scale to follow that strategy.

These bottlenecks arrive while customers are committing to space earlier. A 2025 capacity analysis reported that North American colocation vacancy had fallen to 2.3%. Vacancy dropped below 1% in several of the largest markets.

Colocation lets multiple customers rent computing space, power, cooling, and network access inside a professionally operated facility. It gives enterprises an alternative to building and maintaining their own sites.

Nearly three-quarters of the construction pipeline covered by the analysis was already preleased. Enterprises were reserving capacity 18 to 24 months before they expected to use it. In effect, buyers had to commit before seeing the finished product.

This changes the value of an existing facility. A functioning site can remove years of uncertainty involving permits, equipment procurement, construction, and grid access. Even an older building can offer something a proposed campus cannot guarantee: a starting date.

The supply squeeze has also pushed rents higher. JLL found that North American data center rents had risen 50% over five years, even as inventory expanded in major regions. New supply did not catch demand because cloud providers and AI operators absorbed capacity quickly.

Hyperscalers contributed heavily to that competition. These companies operate very large cloud platforms and data center fleets. Their scale allows them to lease entire buildings or reserve large blocks of future capacity.

Enterprise buyers rarely need the same volume. However, they compete for the same utility connections, construction labor, and electrical equipment. A hyperscaler’s reservation can remove enough capacity to change the options available to dozens of smaller customers.

The market therefore assigns a premium to operational certainty. A data center with power, connectivity, and an experienced staff can serve customers while a newer project remains in an interconnection queue.

That does not make every old data center valuable. It makes time, power, and upgrade potential the leading parts of the valuation.

Existing Data Centers Gain Value, but Only If They Can Be Upgraded

Existing data centers have become strategic assets because replacement takes years, but their physical limits still determine what workloads they can accept.

JLL argues that data centers rarely become completely obsolete. Operators can replace servers, networking equipment, cooling systems, and electrical components in stages. The underlying shell often remains useful through several technology cycles.

Computing equipment also follows a different replacement schedule from the building. JLL says tenants typically own their graphics processors, central processing units, and networking hardware. Those components receive upgrades on an accelerated cycle of roughly five years.

The separation matters for owners. A landlord does not necessarily need to rebuild the entire property when a tenant adopts newer processors. The owner needs enough power, cooling, floor strength, and distribution capacity to support the replacement equipment.

North American and European operators have generally favored systematic retrofits over demolishing older sites, according to JLL. Asia-Pacific markets have used selective retirement more often for facilities built before 2015.

Retrofitting means modifying an operating building to support new technical requirements. It can include stronger electrical feeds, liquid cooling, new backup systems, or redesigned rack layouts.

An existing site begins with several advantages. It may already have a utility relationship, fiber connections, security controls, operating permits, and a trained facilities team. Those features can take years to reproduce elsewhere.

Location also matters. Many established data centers sit near dense fiber routes and major customer populations. Those connections support workloads that cannot tolerate long network delays, including financial transactions, healthcare systems, and interactive business applications.

Existing facilities can also let enterprises deploy incrementally. A company might reserve several racks for a pilot, then expand after measuring demand. That approach reduces the execution risk of committing to an entire new building.

However, the physical upgrade can be difficult. Morningstar’s AI infrastructure study found that most traditional colocation facilities were not designed for densities above 20 kilowatts per rack.

Rack density measures the electrical power consumed by equipment inside one server cabinet. Higher density packs more computing into a smaller space, but it produces more heat and requires stronger electrical distribution.

AI systems can demand far more power than traditional enterprise servers. Training large models uses clusters of graphics processors running together for extended periods. Production inference also creates sustained demand when applications serve many users.

Morningstar expects AI data center capacity to increase sixfold and represent 64% of United States capacity. That forecast strengthens the case for upgrades while exposing how much existing infrastructure must change.

Air cooling becomes less effective as rack density rises. Fans and chilled air can handle ordinary enterprise systems, but tightly packed accelerators produce concentrated heat. Operators increasingly use liquid cooling, which moves heat through coolant near the processors.

Liquid cooling is not a simple equipment swap. It can require new piping, coolant distribution units, leak detection, maintenance procedures, and changes to the facility’s heat-rejection system. Operators must complete those changes without disrupting existing tenants.

Electrical systems present similar difficulties. An older facility might have enough total utility power but lack the internal distribution needed for dense racks. Upgrading switchgear or busways can require planned shutdowns and careful sequencing.

This creates a divided market among existing facilities. A power-ready building with adaptable cooling can support new demand relatively quickly. A rigid facility may remain suitable only for conventional workloads.

Those conventional workloads still matter. Databases, storage, networking, backup systems, and business applications do not all need AI-level rack density. Enterprises often operate mixed environments rather than converting every system to graphics processors.

That diversity supports older data centers. Owners can place dense AI clusters in upgraded zones while serving conventional applications elsewhere. They can also reserve the most modern halls for customers willing to pay for specialized capacity.

The key measure is not a building’s age. It is how much useful power reaches each rack, how efficiently heat leaves the room, and how safely the operator can change those systems.

Enterprise Deployments Are Being Squeezed Between Hyperscalers and AI Specialists

The supply shortage puts enterprise technology teams in direct competition with buyers that can reserve more capacity, accept longer contracts, and move faster.

JLL reported global data center occupancy of 97% at the end of 2025. It also found that 77% of capacity under construction had already been committed to tenants.

Those figures challenge the idea that construction activity automatically creates an oversupplied market. Much of the future inventory has a customer before it opens. Enterprises searching after completion can find few meaningful options.

Hyperscalers occupy more than half of global data center space, according to JLL. Their long leases and strong balance sheets make them attractive tenants. Owners can finance projects more easily when a large cloud company has committed to the capacity.

The same preference creates pressure for ordinary businesses. A regional bank, manufacturer, or healthcare provider may need a few megawatts rather than an entire campus. Its requirement can be economically important but less useful for financing a large development.

AI infrastructure companies add another source of demand. Neocloud providers, which rent graphics-processor capacity for AI workloads, often seek dense deployments with rapid delivery. Their business depends on putting expensive processors into service quickly.

Sean Farney, JLL’s vice president of data center strategy, told industry reporting that less than 10% of United States inventory could support a true AI-dense critical load. He described intense activity around power-dense environments.

That estimate should not be treated as a universal engineering threshold. Facility capabilities vary by room, rack, and cooling design. Still, it illustrates how little inventory combines operating status with AI-ready density.

Enterprises face three broad choices. They can lease colocation space, use public cloud services, or keep infrastructure on their own premises. Many organizations combine all three in a hybrid architecture.

Hybrid IT places applications across on-premises systems, colocation facilities, and public clouds. The model lets a company match each workload with the environment that best fits its security, performance, and operational needs.

Scarcity changes those decisions. A company might prefer colocation for predictable performance and control, but no suitable capacity may be available nearby. The organization then faces a longer reservation period, another market, or a larger cloud commitment.

Moving to a secondary market can reduce competition for space. It can also increase network latency, complicate staffing, and separate systems from users or business partners.

Using public cloud capacity avoids a construction project, but it does not remove the underlying infrastructure constraint. Cloud providers must still secure data center space, power, cooling, and equipment. Their scale changes how they manage the shortage.

On-premises expansion can work when an enterprise already controls a suitable building and utility connection. Yet many corporate facilities were designed for lower-density servers. Their electrical and cooling systems may need the same upgrades as an older colocation site.

The pressure extends beyond AI workloads. Hyperscalers can absorb capacity that enterprises need for ordinary databases, storage, or disaster recovery. A shortage created by AI spending can therefore raise costs and delay unrelated technology projects.

Facilities and technology teams must plan together earlier. A computing strategy that ignores electrical capacity can fail before hardware procurement begins. Conversely, a facilities plan based on historical rack density may not support the next application portfolio.

The competition also changes contracting. Enterprises must make commitments before software demand becomes certain. They may reserve more capacity than current usage requires because waiting carries a risk of losing access.

That behavior can reinforce scarcity. Early reservations remove future capacity from the market, encouraging other buyers to commit sooner. It resembles a queue where every participant advances its order because everyone expects a delay.

Google News coverage often frames the data center boom through announcements involving new campuses. For enterprise buyers, the more consequential story is the disappearance of flexible, near-term capacity inside existing markets.

The winner is not automatically the organization with the largest computing budget. It is the buyer that can connect application planning, capacity forecasting, procurement, and facility engineering before the preferred space disappears.

The Upgrade Thesis Still Faces Technical and Financial Limits

Scarcity raises the option value of an existing facility, but it cannot erase structural limitations or guarantee that every retrofit will produce acceptable returns.

The strongest skeptical argument concerns the mismatch between traditional data centers and modern AI equipment. An operational grid connection has value, but the facility may not deliver enough power to the right place.

Morningstar noted that retrofitting existing facilities has proved difficult and costly. It cited Meta’s 2022 decision to cancel expansions at several developing sites and pursue new construction instead.

That precedent matters because Meta has substantial engineering resources. If a large platform chooses a new design, smaller operators should not assume that every legacy property can be economically converted.

Cooling is one obstacle. Dense accelerator racks can overwhelm an air-cooled room even when the property has sufficient utility supply. Adding liquid cooling may require changes across the mechanical system, not just inside the rack.

Space creates another constraint. New pipes, heat exchangers, pumps, and electrical distribution equipment need room. An older building may have little unused floor area or insufficient ceiling height.

Structural loading can also limit a conversion. Dense computing racks and cooling equipment add weight. Floors designed for older servers may require reinforcement before accepting the new configuration.

Operational continuity makes every change harder. Colocation customers expect constant availability. Operators must install new systems around active equipment, schedule maintenance carefully, and preserve redundant power and cooling throughout the project.

The shortage can encourage optimistic underwriting. Buyers may price a facility according to future AI capacity before completing engineering studies. If the utility, cooling plant, or structure cannot support the planned density, the expected value can evaporate.

Time-to-revenue is central to that risk. GPUs lose economic value while sitting idle because faster hardware continues to enter the market. A delayed retrofit can undermine the business case even if the facility eventually opens.

Farney warned that idle hardware can break revenue assumptions and complicate refinancing. This connects physical construction risk directly with capital availability.

Financing conditions add pressure. A stabilized data center with reliable tenants can attract long-term capital. A conversion project with uncertain engineering requirements looks more like a development risk.

JLL expects core fund formation for stabilized data center assets to exceed 50 billion in 2026. It also forecasts additional liquidity from asset-backed securities and commercial mortgage-backed securities.

Those capital flows favor proven income and credible operators. They do not mean every powered building will receive financing. Investors will still examine tenant quality, lease duration, upgrade costs, and utility commitments.

Local resistance presents another uncertainty. JLL reports a 58-point gap between national support for data centers and acceptance near proposed facilities. National support reached 93%, while local support stood at 35%.

Communities have raised concerns about electricity prices, water consumption, noise, land use, and limited local employment. An existing facility can avoid some permitting risk, but a major expansion may reopen those disputes.

Energy costs can also weaken the upgrade thesis. A JLL energy assessment found that industrial electricity prices across six major markets rose about 18% between 2019 and 2024.

The wait for a large new grid connection was approaching five years in major markets. Operators have considered onsite generation, batteries, power-purchase agreements, and private wires to move projects forward.

These approaches improve scheduling flexibility, but they introduce fuel, maintenance, permitting, and emissions questions. Onsite power is not equivalent to an unlimited utility connection.

Data center owners therefore need disciplined technical diligence. They must confirm the capacity of utility feeds, substations, switchgear, cooling plants, and structural systems before assigning an AI premium.

Enterprise customers need similar caution. A provider’s claim that a site is AI-ready should include measurable rack density, cooling specifications, redundancy, commissioning results, and an achievable delivery schedule.

The market’s reversal is real, but conditional. Existing capacity is more valuable because new supply is slow. It remains valuable only when engineers can convert that advantage into dependable computing.

Three Signals Will Show Whether Scarcity Keeps Favoring Existing Capacity

Grid delivery, retrofit performance, and enterprise leasing will determine whether existing data centers retain their advantage over new construction.

The first signal is the delivery time for grid connections. JLL’s data shows that electricity access has become the controlling schedule item in many markets. Shorter queues would weaken the scarcity premium attached to operational facilities.

Utilities could improve delivery by expanding substations, transmission lines, generation, and interconnection staffing. Regulators could also change how large loads enter the queue or pay for upgrades.

Progress will vary by region. A national improvement would not help an enterprise that needs capacity near one metropolitan area. Buyers should examine utility-level timelines rather than relying on broad market averages.

If connection times remain near several years, existing powered sites will retain leverage. Developers will keep pursuing behind-the-meter generation, battery storage, and private energy agreements.

The second signal is verified retrofit performance. Owners need to show that existing facilities can support denser racks without sacrificing uptime, efficiency, or maintenance access.

Useful evidence includes completed cooling conversions, measured rack density, commissioning results, and the time required to place customer hardware into service. Announced plans offer less value than operating deployments.

Modular designs could improve the retrofit case. JLL expects the modular data center market to expand from 11 billion in 2025 to 48 billion in 2030.

A modular data center uses factory-built power, cooling, or computing units that technicians assemble at the operating site. Standardized production can reduce onsite labor and shorten construction schedules.

JLL says modular implementation could reduce logistics costs by 40% and enable eight-week delivery cycles. Those estimates remain forecasts, but successful deployments would strengthen the advantage of existing powered campuses.

Modular systems cannot solve every problem. They still need land, electrical supply, cooling interfaces, permits, and network connections. Their value comes from simplifying parts of a project, not eliminating infrastructure requirements.

The third signal is the share of enterprise leasing compared with hyperscaler absorption. Cloud and technology providers accounted for nearly two-thirds of North American leasing activity during the first half of 2025.

If enterprises regain access to smaller blocks of near-term capacity, the market may be developing a healthier range of offerings. That would support hybrid infrastructure without forcing every company toward a large cloud commitment.

If hyperscalers continue absorbing most new supply, enterprise buyers will need earlier planning and more flexible locations. Existing corporate data centers could also receive additional investment because replacing them remains difficult.

Morningstar’s utility forecast expects United States data center electricity demand to rise from 5% of total consumption to 34% by 2035. The estimate is aggressive and should be monitored against actual utility load.

Morningstar also identifies affordability and regulation as important risks. Rising infrastructure costs can trigger political limits on electricity rates or utility profits. Those responses can slow projects even when technical demand remains strong.

The next several quarters should reveal whether the industry can turn announced power agreements into operating capacity. Investors should separate utility requests, construction starts, completed connections, and energized server halls.

Enterprise buyers should track a more practical set of measures. They need confirmed delivery dates, usable rack density, cooling compatibility, and expansion rights. Headline capacity means little if it cannot support the intended hardware.

This is also where Google News readers should look beyond construction announcements. A campus plan describes ambition. Energized capacity describes what customers can actually use.

Existing data centers now sit at the center of that distinction. Their grid access offers a head start, while their technical limitations impose a test.

For facilities leaders, the immediate action is clear. Audit current power, cooling, network, and structural capacity before the next computing project reaches procurement.

Technology leaders should connect workload forecasts with facility requirements at the same time. Waiting until servers arrive turns infrastructure planning into crisis management.

Investors should value operating certainty without assuming that every legacy building can become an AI facility. A credible upgrade plan needs engineering evidence, customer demand, and a realistic commissioning schedule.

The larger question is whether utilities and developers can deliver new capacity faster than enterprises consume what already exists. Until that changes, powered and adaptable data centers will remain unusually valuable.