AI’s Influence on the Data Center Colocation Market

AI has changed what buyers need from a colocation provider.

A standard server rack draws about 7.6 kW of power. One rack of current AI hardware can draw 120 kW, roughly sixteen times more, and most existing data centers were never wired or cooled to handle that level of density.

Capital is arriving faster than new capacity can be built.

Global data center spending is expected to reach roughly $2.9 trillion through 2028, while much of the available capacity in major markets is already committed.

Only 1.4% of data center capacity in major North American markets sits vacant today, a record low.

This article walks you through how AI is reshaping colocation power requirements, cooling infrastructure, capacity availability, market pricing, and what buyers should look for when choosing an AI-ready data center.

How Much Are Companies Spending on AI Data Center Infrastructure?

Enterprises and service providers are projected to commit $1.6 trillion to AI infrastructure worldwide by 2029. 

This investment is driving rapid growth in accelerated servers, power systems, cooling infrastructure, and large AI data centers.

Major technology companies are projected to spend more than $2.8 trillion on AI infrastructure through 2029, with hyperscaler AI capital expenditures reaching about $490 billion by the end of 2026. 

Global AI computing demand could require an additional 55 GW of power capacity by 2030, with each incremental gigawatt costing roughly $50 billion.

AI Data Center Statistics

The most useful AI data center statistics describe different parts of the market rather than one single market size.

The figures below separate infrastructure commitments, broad technology-company investment, and AI-linked data center demand.

Table 1. AI Infrastructure Forecasts

Source	Metric	Figure	Target Year
IDC	Worldwide AI infrastructure commitments by enterprises and service providers	$1.6 trillion	2029
Citi	AI-related infrastructure spending by major technology companies	More than $2.8 trillion	Through 2029
Cushman & Wakefield	Annual AI-linked data center demand in its 2024 market forecast	$75.3 billion	2028

Note: These forecasts measure different parts of the AI infrastructure market and should not be treated as directly comparable market-size estimates.

For additional industry metrics, see Brightlio’s data center statistics.

How Tight Is Colocation Capacity in Primary Markets?

Colocation capacity is extremely tight across North America’s primary data center markets.

CBRE reported a vacancy rate of only 1.4% in the first half of 2026, despite total supply reaching a record 10,903 MW.

Another 7,481.1 MW was under construction, but 80.4% had already been preleased. This left less than 1,500 MW of future capacity available, equal to about six months of demand at the current rate.

Northern Virginia remained the largest market and recorded a vacancy rate of just 0.2%. Buyers may need to reserve power and capacity long before their planned deployment dates.

A complete comparison of the largest U.S. data center markets should cover more than total inventory. Available power, delivery schedules, preleased capacity, and support for high-density deployments now determine how much usable capacity a market can offer.

Table 2. Selected Primary-Market Capacity and Pricing Snapshot

Market	Vacancy Rate	250 to 500 kW Asking Rent	Q1 2026 Pricing Context
Chicago	2.2%	$200 to $230 per kW/month	Up 14.7% year over year
Northern Virginia	0.3%	$190 to $235 per kW/month	Among the tightest major markets
Frankfurt	5.0%	$235 to $265 per kW/month	Highest rate range among major European markets
Singapore	2.0%	Average $403 per kW/month	Highest average among major Asia-Pacific markets

Source: CBRE Global Data Center Trends 2026. Q1 2026 values are used here because they provide a comparable cross-market pricing snapshot; CBRE’s H1 2026 North America update reports more recent regional capacity conditions.

How Much Power Do AI Racks Require?

AI racks can require 30 to 150 kW per rack, with current GPU deployments reaching about 100 kW per rack. Traditional CPU racks typically use only 3 to 10 kW per rack.

Colocation providers now support much higher rack densities. Digital Realty offers high-density colocation from 30 kW to 150 kW per cabinet. 

Available cooling options include air-assisted liquid cooling and direct liquid cooling. 

These figures give a more accurate picture of current AI-ready facilities than older examples of 50 to 70 kW per cabinet.

Table 3. Rack Power Density by Deployment Type

Deployment Type	Typical Cited Density	Cooling Implication
Conventional CPU server rack	3 to 10 kW per rack	Conventional air cooling is typical
High-density colocation service range	30 to 150 kW per cabinet	Liquid-cooling options become increasingly important
CBRE 100-kW GPU example	About 100 kW at rack-level density	Closed-loop liquid cooling

Sources: CBRE H1 2026 and Digital Realty high-density colocation. Exact cooling requirements depend on server design, rack layout, facility engineering, and environmental conditions.

How Much Electricity and Water Do AI Data Centers Consume?

AI-focused hyperscale data centers can consume enormous amounts of electricity. Pew Research Center reports that a typical facility can use as much electricity each year as 100,000 households. 

Some of the largest new facilities could consume about 20 times more. In 2023, data centers used 26% of Virginia’s electricity, 15% in North Dakota, 12% in Nebraska, and 11% in both Iowa and Oregon.

Servers use about 60% of a data center’s electricity on average. Cooling systems use between 7% and more than 30%, depending on facility efficiency. 

U.S. data centers directly consumed about 17 billion gallons of water in 2023, and hyperscale and colocation facilities represented 84% of that total. Hyperscale facilities could consume 16 to 33 billion gallons per year by 2028.

What Are Data Center Sustainability Requirements, Why Do They Matter, and What Is Driving Demand?

Data center sustainability requirements are standards for reducing energy use, water consumption, carbon emissions, and environmental impact.

They matter because electricity and cooling capacity can affect operating costs, expansion plans, and corporate environmental targets.

AI is driving stricter requirements because high-density computing consumes more electricity and requires intensive cooling.

Some cooling systems use significant amounts of water, creating resource concerns in water-stressed markets.

In 2024, natural gas supplied more than 40% of U.S. data center electricity. Renewables supplied 24%, nuclear supplied 20%, and coal supplied 15%.

Colocation buyers should assess energy sources, cooling efficiency, water use, renewable energy options, and future power availability.

Why Is AI Accelerating Edge Computing Demand?

AI is increasing demand for edge computing because many applications need fast processing close to users, devices, and data sources. 

Autonomous systems, IoT devices, real-time analytics, personalization tools, and interactive applications can experience delays when every request travels to a distant data center.

Limited capacity in major hubs is pushing development into new markets.

JLL reported in August 2026 that frontier markets represented 77% of North American capacity under construction. 

Many tenants reserving capacity now are receiving delivery dates in 2028. Cushman & Wakefield ranked Austin-San Antonio and West Texas among the leading secondary and tertiary markets in its 2026 global comparison.

Buyers that can operate outside the largest data center hubs may consider markets such as Salt Lake City, Austin, San Antonio, and Reno. Brightlio’s data center market trends coverage explains how power limits and development delays are changing location decisions.

How Is AI Affecting Colocation Pricing?

AI is pushing colocation prices higher because demand is strong and suitable capacity is limited. AI deployments may require upgraded electrical systems, liquid cooling, more mechanical equipment, and expensive facility changes. 

Buyers are competing for a limited supply of power available in the near term.

CBRE reported higher asking rents across every deployment size in the first half of 2026. 

Rates increased 4.3% for 250 to 500 kW deployments, 7.9% for 500 kW to 3 MW, 8.3% for 3 to 10 MW, and 6.7% for deployments above 10 MW. 

These increases show that pricing pressure affects deployments of all sizes.

The advertised price per kW represents only part of the total cost. 

Buyers should review contracted power, rack-density limits, cooling fees, installation costs, cross-connects, redundancy, expansion capacity, and future power-delivery dates.

What Does AI Mean for Colocation Buyers?

AI changes what buyers need from colocation facilities. Dense GPU systems place more electrical load and heat into each rack. Buyers should evaluate available power, rack-density limits, cooling systems, expansion capacity, connectivity, and contract terms.

Colocation buyers should confirm these six factors before selecting a facility:

Factor	What to Confirm	Warning Sign
Power	Confirm the power available at move-in and the power reserved for expansion.	Future power has no firm delivery date.
Rack Density	Confirm the maximum kW supported per cabinet, row, and suite.	The provider gives only total facility capacity and omits cabinet-level limits.
Cooling	Confirm the cooling system and maximum supported heat load.	The required cooling system needs major facility upgrades.
Expansion	Confirm that cabinets, power, cooling, and activation dates are reserved.	Expansion depends on utility capacity that has not been committed.
Location and Connectivity	Confirm carrier options, latency, cloud access, and available power.	Limited network or power capacity restricts the planned workload.
Total Cost	Calculate power, cooling, cross-connect, installation, support, and expansion fees.	The quoted price leaves out major infrastructure or future capacity costs.

How Much Power Can the Facility Actually Deliver?

The power committed to a customer’s suite determines how much equipment the deployment can run.

Campus capacity shows the total power available across the facility. The electrical system serving the suite shows how much power the customer can actually use.

The capacity plan should include three figures: power available at move-in, power reserved for expansion, and the activation date for each future power block.

Projects that need new high-voltage transmission lines or generation capacity may face grid connection timelines of 24, 36, or more than 48 months.

A project planning to add several megawatts within two years should reserve that power early.

Every power figure needs a clear definition. Critical IT load, installed electrical capacity, redundant upstream capacity, and utility service capacity measure different parts of the electrical system.

The facility’s redundancy design and data center tier provide more information about power delivery and reliability.

How Much Rack Density Can the Facility Support?

Maximum rack density shows whether the facility can support dense GPU systems. Cabinet, row, and suite limits reveal how much power can reach the deployment.

Buyers should confirm the maximum kW per cabinet, maximum row density, busway and breaker limits, PDU capacity, and total density available within the contracted suite.

A campus-level megawatt figure does not show these limits.

Modern GPU racks place much more electrical load into a small area. This makes high power density an important facility design requirement.

Hardware upgrade plans should be part of the calculation because newer accelerators can increase power and cooling needs within the same number of cabinets.

What Cooling Does the AI Deployment Require?

Cooling requirements depend on the heat produced inside each rack. An NVIDIA GB200 NVL72 rack has approximately 120 kW of designed rack power and uses liquid cooling. Its cooling needs are far greater than those of conventional low-density server racks.

Facilities supporting this type of hardware need coolant distribution, heat-removal capacity, leak detection, monitoring, and operating procedures for dense computing systems.

Direct-to-chip cooling transfers heat from CPUs and GPUs through cold plates. Immersion cooling places equipment in dielectric fluid that removes heat directly.

The facility’s liquid cooling systems must match the hardware and rack density.

The contract should state who is responsible for coolant distribution units, piping, customer equipment, maintenance, monitoring, testing, and upgrade costs.

Some lower-density GPU deployments can use air cooling or rear-door heat exchangers. Rack-scale systems designed for liquid cooling need liquid infrastructure from the start.

How Much Expansion Capacity Is Reserved?

Reserved expansion capacity determines whether the facility can support future deployment phases.

CBRE reported that 80.4% of capacity under construction across North America’s primary markets was already preleased in the first half of 2026. The figure was 74.3% one year earlier.

Less than 1,500 MW of capacity under construction remained available across those markets. This represented about six months of demand at the current rate.

A strong expansion agreement reserves cabinets, power, cooling capacity, and activation dates together. Space, power, and cooling must all be available for the next deployment phase.

The expansion plan should cover the initial deployment, the next hardware addition, and the expected final footprint. Every phase needs a defined power allocation and delivery date.

Which Location Provides the Right Power and Connectivity?

The right location must provide enough power and suitable network connectivity. Vacancy across North America’s primary data center markets fell to 1.4% in the first half of 2026, compared with 1.6% in the first half of 2025.

Demand filled new capacity quickly and kept available inventory near record lows.

This limited availability makes power a major location requirement. Primary markets offer large carrier networks, cloud connections, skilled workers, and mature infrastructure.

Selected emerging markets may provide larger development sites and more opportunities for new power capacity.

Latency-sensitive AI inference needs carrier diversity, cloud access, suitable network routes, and proximity to users.

Large AI training clusters need large power blocks, firm delivery dates, and room for long-term expansion.

A suitable site must provide the electricity and connectivity needed to transfer training data, connect systems, and deliver inference results within the required response time.

What Is the Real Cost of AI Colocation?

The real cost of AI colocation includes rent, power, cooling, connectivity, installation, support, and reserved expansion capacity.

In the first half of 2026, rental rates across primary North American markets increased 8.3% for deployments requiring 3 to 10 MW and 4.3% for deployments requiring 250 to 500 kW.

A complete colocation pricing model should include high-density power distribution, cooling equipment, cross-connects, remote hands, network capacity, installation, testing, and reserved expansion capacity.

Contract terms can significantly affect the cost of a multi-year AI deployment.

Buyers should review committed power, rent increases, installation duties, cooling fees, expansion rights, cross-connect charges, remote-hands rates, and future activation dates.

Every shortlisted facility should be compared using the same rack density, power requirement, cooling system, connectivity needs, and growth schedule.

Final Thoughts

AI is changing the criteria organizations use to evaluate colocation. 

Power density, cooling capability, available megawatts, connectivity, expansion options, delivery timelines, and cost now need to be assessed together. 

Organizations planning AI deployments should define those infrastructure requirements before comparing facilities so they can distinguish genuinely AI-ready capacity from space that cannot support the intended workload.

Brightlio Delivers AI Colocation Solutions

Brightlio sources colocation capacity for dense AI footprints and traditional deployments through a global network of data center partners. We help organizations compare facilities based on available power, supported rack density, cooling, connectivity, location, expansion options, and budget.

If you need help identifying facilities that match a specific AI infrastructure profile, contact Brightlio to discuss your deployment requirements.

Frequently Asked Questions About AI and Data Center Colocation

What Is the Influence of AI on Data Centers?

AI is increasing rack power density, electricity demand, cooling requirements, and competition for available capacity.

Data centers that support GPU-heavy workloads may need upgraded electrical distribution and liquid-cooling systems, while operators must secure more power and plan capacity farther ahead than many conventional enterprise deployments require.

How Much Power Do AI Data Center Racks Require?

Power density varies by hardware and design. Conventional CPU racks historically operated around 3 to 10 kW per rack, while current high-density AI deployments can reach roughly 100 kW or more.

Some colocation providers now support cabinet densities from 30 kW up to 150 kW.

Why Do AI Data Centers Need Liquid Cooling?

High-density GPU infrastructure produces much more heat in a smaller physical footprint. As rack density rises, conventional air cooling becomes increasingly difficult to use efficiently.

Direct liquid cooling, air-assisted liquid cooling, rear-door systems, or immersion can move heat away from high-density equipment more effectively, depending on the deployment.

How Are AI Workloads Different From Traditional Data Center Workloads?

AI workloads often concentrate far more compute and power into each rack. Training and inference systems can use large GPU clusters, high-speed networking, and specialized storage, which increases electrical and cooling requirements.

Traditional workloads may use lower-density CPU infrastructure and can therefore fit more easily within conventional data center designs.

Is AI Increasing Colocation Costs?

Yes. AI contributes to higher colocation costs through stronger demand, scarce power, high-density electrical infrastructure, advanced cooling, and facility retrofit requirements.

CBRE reported rent increases across all major deployment sizes in H1 2026, although the exact price still varies significantly by market, contract size, and facility design.

What Should Companies Look for in an AI-Ready Colocation Facility?

Confirm available and future power, supported rack density, cooling architecture, redundancy, carrier connectivity, expansion capacity, and power-delivery timelines.

Companies should also understand whether liquid cooling is available for their hardware and evaluate the total deployment cost rather than comparing facilities solely on advertised space or per-kW pricing.