The AI buildout is generating demand that has to land somewhere, and a significant portion of it could be landing in colocation. This blog examines the decision facing many operators right now, what it actually takes to evaluate it well, and why the operators who come out of this cycle strongest will have won on operations, not just on timing.
The headlines tell two stories at once. Hyperscaler CapEx is approaching $700 billion, national vacancy rates have hit a historic 1%, and the pipeline of enterprises needing colocation space is growing faster than new capacity can come online. At the same time, construction moratoriums are spreading, power interconnection queues are stretching into years, and the first contraction in US data center construction in five years is underway.
The assumption that AI and HPC growth somehow competes with traditional colocation demand gets the dynamic backwards. Every AI workflow ultimately produces data, applications, inference results, or customer-facing services that still need to live somewhere operationally. AI infrastructure may change the density profile of the environment, but it doesn’t eliminate the need for colocation capacity. In many cases, it expands it.
For colocation operators, that tension is an opportunity, but not a simple one.
The demand is real, but so are the potential costs of capturing it. Read on for ways to evaluate the opportunity clearly, what tenants are actually scrutinizing, and what the operators best positioned to capture it are doing differently
The Question on Every Operator’s Desk
Colocation operators sitting on existing capacity today are asking themselves a difficult question: whether to retrofit that capacity for AI and HPC workloads for potentially higher margins, but make significantly higher capital investment.
Retrofitting existing colocation space for AI/HPC often requires significant investment in liquid cooling infrastructure, CDU deployment, upgraded power distribution, and structural modifications that can materially increase cost per deployed megawatt. Depending on the environment, operators may be evaluating retrofit costs that can cost several million dollars per MW. But the revenue case, while compelling on paper, is still a little premature.
Smart operators are thinking about this the right way. Some facilities, particularly those in markets where traditional colocation demand is softening, are well-positioned to retrofit for AI/HPC deployment. Others are better served maximizing the profitability of their existing capacity. The decision isn’t the same for everyone, and the operators who get it right will be the ones who approach it with data rather than intuition.
That starts with understanding profitability per megawatt, on both sides of the ledger.
What Profitability Per Megawatt Actually Requires
While it’s a straightforward concept, this metric is surprisingly hard to calculate in practice. On the revenue side: what are you charging tenants per megawatt, and how much of that revenue are you actually retaining after SLA penalties, service credits, and contract remediation? On the cost side: what is your fully loaded overhead per megawatt — capital, OpEx, maintenance, labor — and how does that vary across your portfolio?
The gap between those two numbers is your profitability per MW:
Realized Profitability per MW (monthly) = (Contracted Revenue per MW − SLA Credits & Remediation) − Operating Cost per MW (Cooling + Maintenance + Labor) − Capital Recovery per MW (amortized retrofit/build)
For colocation operators evaluating AI/HPC retrofits, even small swings in any one of those variables can materially change whether the business case holds over time, but producing that number for a single site is an exercise that can take weeks for those without easily accessible data. Producing it across a portfolio of 10, 20 or 50 sites is something many organizations haven’t done at all.
For the vast majority of organizations, producing profitability per MW reporting still requires pulling maintenance costs from one system, power data from another, incident records from email threads, and SLA remediation details from spreadsheets managed separately by operations and finance teams.
Operators who can measure profitability per MW consistently across their portfolio are in a position to evaluate retrofit decisions more confidently, negotiate from a position of strength, and identify where operational costs are eroding margin. Operators who can’t are making significant capital decisions based largely on estimates.
What Today’s Tenants Are Actually Evaluating
The tenants queuing up for colocation capacity right now aren’t lowering their standards because their options are limited. They’re raising their scrutiny precisely because the stakes are higher.
An enterprise running AI inference workloads will evaluate your uptime history, of course, but they’ll also be evaluating whether your operations are mature enough to give them the data they need to manage their own cost and performance picture. When tenants are deploying infrastructure that may represent millions of dollars per megawatt in combined GPU and facility investment, they need a colocation partner who can help them understand performance at the megawatt level:
- Real-time power consumption trends
- Cooling efficiency by deployment zone
- Incident history tied to specific infrastructure
- Maintenance costs at the asset level
- Operational performance across multiple facilities
Who can win customers in this market and keep them when capacity opens up again, will be the ones who treat operational data as a tenant service. Uptime is just the baseline, visibility is the true differentiator.
The Operational Foundation That Makes the Decision Defensible
Whether the decision in front of you is to retrofit for AI/HPC, to maximize profitability on your existing footprint, or both, the underlying requirement is the same: you need clean, accessible portfolio visibility.
That means maintenance costs that are tracked at the asset level, not estimated at the site level. SLA performance that is documented and reportable, not reconstructed after a tenant dispute. Incident history that is structured and searchable, not buried in email threads. And power consumption data with equipment-level reporting (not just at the facility level), so that when a tenant or a regulator asks for it, the answer is already there.
This market won’t stay constrained forever. The operators who come out of this cycle retaining tenants will be the ones who treated this moment as a forcing function: to build the kind of operational and financial clarity that makes every future decision easier. The capacity crisis may have created the opportunity, but operational discipline will determine who actually captures it.
*Capital Recovery= the per-MW cost to build or retrofit the capacity, amortized over its useful life to recover both the investment and the required return on it (a capital charge, not just depreciation). Excludes ongoing maintenance, which sits in operating cost.
The operators who can proactively identify their gaps have a meaningful advantage over the ones who are about to find out they existed. Talk to one of our experts about an Operational Audit.
