Enterprises looking for colocation space to support a digital transformation initiative, to launch an AI project or just to keep pace with demand for compute power to run day-to-day operations are in for a rude awakening. The vacancy rate for colocation space is as low as it has ever been, and there’s no end in sight to the colocation space crunch. As a result of this supply/demand imbalance, prices in some locations are spiking by nearly 20% year-over-year.
Enterprises that currently have a colocation lease agreement and are looking to re-up should be prepared for bigtime sticker shock. And enterprises that require new colo space are likely to face an even bigger shock: little or no available space at all, and a waiting list that could stretch out for 18 months or more, according to experts.
For enterprise IT execs who already have a lot on their plates, the lack of available colocation space represents yet another headache to deal with, and one with major implications. Nobody wants to have to explain to the CIO or the board of directors that the company can’t proceed with digitization efforts or AI projects because there’s no space to put the servers.
IT execs need to start the planning process now to get ahead of the problem.
In a world of mergers, acquisitions, reorganizations, tariffs, etc., coming up with a long-range estimate of data center requirements isn’t easy. And many companies are chomping at the bit to get in on the AI boom without a clear idea of what specific projects they have in mind, or any clear methodology to predict future AI workload demands. But don’t despair. There are options and tactics that organizations can take to make sure their need for compute capacity is met.
How bad is it?
Andrew Batson, senior director of Americas data center research and strategy at JLL, a real estate services company, says that the colocation vacancy rate at the end of 2024 hit an all-time low of 2.6% “amid insatiable demand and despite several years of record construction levels.”
Batson paints a pretty dire picture. “Vacant and immediately available data center space is incredibly limited. Across North America there are very few blocks available larger than 5 MW. Any second-generation space that becomes available is re-leased within weeks. Nearly 6.5 GW is under construction, of which 72% is preleased. Tenants looking to lease any sizable amount of data center capacity must wait 24 months on average.”
According to a commercial real estate services firm CBRE, “Demand continues to outpace new supply across both core and emerging hubs.” Inventory across the four largest U.S. data center markets—Northern Virginia, Chicago, Atlanta and Phoenix—increased 43% year-over-year in Q1 2025.
But that increase in inventory was overwhelmed by skyrocketing demand. Northern Virginia remained the tightest market, with a vacancy rate at 0.76%. Phoenix was at 1.7%, Chicago at 3.1% and Atlanta’s vacancy rate was 3.6%.
What’s driving the colo crunch?
Demand has outstripped supply due to multiple factors, according to Pat Lynch, executive managing director at CBRE Data Center Solutions. “AI is definitely part of the demand scenario that we see in the market, but we also see growing demand from enterprise clients for raw compute power that companies are using in all aspects of their business.”
Batson agrees. “AI is driving demand, but it’s not the sole driver. We estimate AI workloads are about 20% of all data center workloads.”
The big wild card contributing to the colo space shortage is that the hyperscalers are snapping up colo space as fast as it comes on the market, as they try to stay ahead of the surge in demand for AI processing from their big customers.
The hyperscalers often make grand announcements about investing billions to build new data centers, but the reality is that they can’t build fast enough. For example, Oracle recently conceded that orders were outstripping available data center space. And Alphabet reported that Google Cloud was unable to meet surging customer demand because of constrained data center capacity.
Lynch says the hyperscalers are pre-leasing colo space before those facilities are even completed. And they don’t just rent a few racks, they want the whole building. That essentially means the hyperscalers are exacerbating the colo space crunch and, by leasing entire facilities, making it harder for enterprises to find small blocks of space.
The latest data from Synergy Research Group shows that the hyperscalers now account for 44% of the worldwide capacity, with more than half of that inside their own data centers and the rest in leased facilities. Non-hyperscale colocation capacity accounts for only 22% of total capacity.
Since the hypervisors have significantly more inventory to begin with, they can still meet the vast majority of enterprise needs. But hyperscalers aren’t an option for enterprises that can’t put sensitive data in the public cloud for regulatory or privacy reasons. Colo space is generally cheaper than public cloud, making it an attractive option. And many enterprises simply want full control of their IT assets. For those enterprises, the colo space crunch is a real problem.
Power constraints
It’s not GPU chip shortages that are slowing down new construction of data centers; it’s power. When a hyperscaler, colo operator or enterprise starts looking for a location to build a data center, the first thing they need is a commitment from the utility company for the required megawattage.
According to a McKinsey study, data centers are consuming more power due to the proliferation of the power-hungry GPUs required for AI. Ten years ago, a 30 MW data center was considered large. Today, a 200 MW facility is considered normal.
McKinsey points out that average power densities per rack have more than doubled in just two years, from 8 kW to 17 kW. Training models like ChatGPT can consume more than 80 kW per rack, while Nvidia’s latest chip may require rack densities of up to 120 kW.
McKinsey predicts that global demand for data center capacity could rise between 19%-22% a year from 2023 to 2030, “raising the potential for a significant supply deficit.”
The cost equation for enterprise IT
There are two cost scenarios that enterprises face: renewals and new contracts.
Lynch says organizations that negotiated leases five years ago are in for a shock when they look to re-up. Prices for colo space have been increasing by double digits every year for the past 4-5 years. And that’s in dramatic contrast to the situation in the 2015-2020 period, when there was an oversupply and prices were actually dropping.
He recommends that IT execs should “set expectations” among procurement and finance teams that costs will increase dramatically for the same amount of colo space.
In terms of new space, global data center pricing rose a moderate 3.3% year-over-year in Q1 2025, according to CBRE. But this is real estate, so location, location, location. Prices in Northern Virginia rose 17.6%, Chicago was up 17.2%, Atlanta pricing rose by 13.0%.
What should enterprises do?
The first step for enterprises is to accelerate and prioritize the planning process. “Get ahead of it,” says Lynch. He recommends that organizations plan at least three years out.
And enterprises need to get their ducks in a row internally. It’s important to bring in all stakeholders, including business units, software development teams, c-suite executives, procurement departments and data center managers to create a comprehensive roadmap of future compute and storage requirements.
In addition, organizations need to be ready to move fast. Lynch says organizations need to have processes in place that enable them to sign agreements quickly, without a lot of red tape or bureaucracy.
Batson adds that if an organization renews or reserves new colo space for 12 or 18 months down the road, they need to understand that they’re writing a check upfront. “It’s not like reserving a table at a restaurant for dinner. You need to put down a financial commitment,” he says.
Here are some options for enterprises to consider:
Stranded power
This option certainly doesn’t apply to all enterprises, but there are some who own their own data center and have vacant space within the facility because they migrated applications to the cloud over time and de-commissioned legacy servers or mainframes. Batson says those facilities still have a dedicated power supply from the utility company. Organizations could tap into that power, and purchase new, more powerful, more efficient data center servers. Or they could partner with a service provider who would operate the on-premises equipment for them; the managed private cloud option.
Piecing it together
Lynch says organizations need to be flexible and willing to accept alternatives to their first choice. In other words, if an enterprise wants 5MW of colocation space in Northern Virginia within six months, they might run smack into a brick wall. But they might be able to grab 1-2MW in Baltimore, and another 1-2MW in Richmond.
Batson agrees: Organizations looking for 5MW or more may find that their options are limited, but “smaller spaces can be found” in other locations that are becoming data center hotspots, such as Phoenix or Des Moines.
There are downsides to this approach, however, which include having to negotiate and managing multiple contracts, plus the added complexity of having IT infrastructure scattered at multiple locations. But it’s better than nothing.
Public cloud
Assuming the organization is unable to reserve all of the colo space it needs for high-priority projects, there could be ways to reallocate the current application inventory. Identify current colo workloads that might be safely moved to the public cloud.
Batson points out that the public cloud might be more expensive than colo, but it’s an option to consider. This can be tricky because many organizations have concerns about putting certain application data, including AI-related data, in the public cloud.
Powered shell
Another alternative for organizations that want to maintain control over their IT infrastructure is to build their own small, modular, data center facilities. A powered shell, essentially a physical structure with power, but no equipment, can be built relatively quickly. Organizations can then add servers as the need arises.
If the organization has to wait 24 months to get new colo space anyway, it might make sense to take the DIY approach. “I do believe as a result of supply constraints, we’re going to see more powered shell developments,” says Lynch.
Along the same lines, organizations can order a turnkey, customized, modular data center that comes with the latest backup power, cooling, server and storage virtualization, high-speed networking and other advanced features. In these cases, the advantages are speed, plus the ability to maintain full control of IT assets.
These are not all-or-nothing options; organizations that plan ahead can cobble together solutions that incorporate multiple approaches to make sure they have enough compute power for current and future requirements.
Source:: Network World