Beyond Commissioning: The Hidden Subscriptions Inside The Data Centre Boom

Last week, I wrote about where data centre money actually flows once the press releases fade. What surprised me was not that the investors agreed, but why they did. The response was not about growth projections or megawatt counts. It was about tangibility. About finally being able to see who actually gets paid when a data centre gets built.

But, that story only goes halfway. There is another layer to the global data centre boom that investors are only beginning to appreciate. It has very little to do with announcements and a lot about what happens after the facility goes live.

Most investors still carry an old mental model of data centres. You build them. You fill them. You sweat the asset. If utilisation ramps up fast enough, margins take care of themselves. That model made sense when data centres were low-density, relatively stable, and slow to change. That is not the world we are in anymore.

A modern data centre, once commissioned, behaves far less like a factory and far more like a bundle of subscriptions you can never cancel. From the moment the first hall switches on, a set of costs locks in and start repeating. They scale with footprint. They tighten with regulation. With AI workloads, they intensify far faster than most models assume.

The important part is this: Many of these costs exist whether the utilisation is 40% or 95%. That distinction is what separates one-time beneficiaries from businesses that quietly compound.

One way we at Greyhound Research have been making sense of this is by grouping data centre participants based on when and how they get paid, rather than what they build. Looked at through that lens, four categories emerge, each with very different risk and return characteristics.

The first bucket is one-time commissioning spend. Civil construction. Basic fit-out. Parts of engineering and integration work. These players are essential to getting a facility live, but their revenue lives and dies by project starts. Delays in power approvals, permitting, or capital timing hit this layer first and hardest. This is also where optimism risk is highest.

When projects slip, this bucket feels it immediately. The second bucket is phase-driven expansion spend. Every incremental megawatt pulls in electrical balance-of-plant, cabling, transformers, and more cooling capacity. This spend repeats, but only when capacity expands.

It feels recurring, but it remains conditional. If utilisation lags or rollout cadence slows, this layer pauses quickly. The third bucket is AI-driven retrofit and density upgrades. This is where older assumptions break. AI workloads do not politely fit into yesterday’s designs.

Higher rack densities force thermal redesigns, power upgrades, and redundancy additions even in relatively new facilities. This spend is not optional. It arrives reactively, often earlier than planned, and usually at a higher cost per unit than greenfield builds.

The fourth bucket, and the one investors consistently underestimate, is existence-linked annuity spend. These are businesses that get paid not because a data centre grows, but because it exists and must stay alive, compliant, and insurable. This is where the subscription analogy really holds.

Take power. Electricity is not a variable input for data centres. It is a fixed obligation. Capacity is reserved in advance. Redundancy is engineered upfront. As AI pushes power draw per rack higher, operators do not negotiate their way out of higher electrical spend. They absorb it.

What often goes unnoticed is that this spending is not merely persistent. It is frequently anti-cyclical. When expansion slows, operators do not relax reliability standards. They tighten them. Deferred growth budgets are redirected toward resilience, redundancy, and efficiency upgrades.

In slowdowns, existence-linked spend behaves less like growth capex and more like defensive infrastructure maintenance. Cooling follows a similar path, but with greater intensity. What used to be treated as a facilities cost has become a performance constraint.

AI breaks linear cooling assumptions. Thermal optimisation stops being episodic and becomes continuous. Systems are tuned, retrofitted, partially replaced, and re-engineered to manage heat, water usage, and energy efficiency.

As density rises, cooling stops being a cost to optimise and becomes a risk to contain. That shift quietly accelerates reinvestment, especially in mature facilities where thermal headroom disappears faster than utilisation grows.

Backup power pushes the subscription logic even further. Generators, batteries, uninterruptible power systems, and fuel infrastructure exist as insurance. They are tested, serviced, audited, and replaced on schedule. Downtime tolerance has collapsed, while penalties for failure have risen sharply through contracts, regulation, and reputational exposure.

Here again, cyclicality behaves counter-intuitively. When demand softens, uptime expectations do not. Reliability spend is one of the last things operators cut, and often one of the first areas they reinforce.

Then there is security, monitoring, and compliance. This layer rarely features in investor conversations because it looks small. But it never goes away. Physical security, access control, fire systems, building management, audits, and reporting scale with footprint and regulation, not traffic.

Over time, these systems become harder to simplify, not easier. Each new requirement adds permanence. At this point, though, the subscription story would still be incomplete if we stopped at physical infrastructure.

Modern data centres also carry a dense layer of software subscriptions that renew annually and expand with complexity. Virtualisation, orchestration, networking software, storage control planes, observability, security tooling, and AI infrastructure software all behave like recurring revenue layered on top of physical assets.

These subscriptions scale with cores, nodes, throughput, features enabled, and regulatory surface area. As environments become denser and more heterogeneous, the software footprint expands. But here the behaviour diverges. When budgets tighten, software subscriptions can be rationalised.

Licenses are renegotiated. Vendors are consolidated. Features are optimised. Physical subscriptions do not behave this way. Power cannot be renegotiated away. Cooling capacity cannot be deferred without consequence. Compliance does not pause. Infrastructure is endured, not optimised.

This asymmetry matters. Infrastructure subscriptions exist because the data centre must stay alive. Software subscriptions exist because the data centre must stay intelligible. One keeps the lights on. The other keeps the system governable. They scale differently, but they scale together.

This is also where valuation mistakes creep in. Many investors expect data centre economics to behave like software. They do not. Revenue may scale linearly with utilisation, but reinvestment does not. It arrives in steps, triggered by density thresholds, regulatory shifts, or resilience gaps that are often discovered late.

AI, in particular, introduces non-linear reinvestment pressure. A single density jump can force simultaneous upgrades across power, cooling, redundancy, and monitoring. These moments are hard to forecast and rarely smooth. What looks like operating leverage on paper often becomes reinvestment gravity on the ground.

This framing also explains why capital is concentrating. As power availability, water access, permitting timelines, and execution bandwidth tighten, capital does not disappear. It migrates. Projects move to resource-secure corridors. Orders gravitate toward proven suppliers. Late entrants get crowded out.

In that sense, data centres are starting to look less like pure technology plays and more like infrastructure utilities with software layered on top. They reward scale, execution discipline, and balance-sheet strength. They punish leverage, shortcuts, and optimistic assumptions about static cost structures.

If you approach this cycle looking only for headline beneficiaries, you risk missing the quieter compounding beneath the surface. If last week was about following the money, this week is about following the payments that never stop. The data centre boom is not just about who builds capacity. It is about who gets paid, month after month, to keep that capacity within tolerance.

Disclaimer: The views expressed in this article are solely those of the author and do not necessarily reflect the opinion of NDTV Profit or its affiliates. Readers are advised to conduct their own research or consult a qualified professional before making any investment or business decisions. NDTV Profit does not guarantee the accuracy, completeness, or reliability of the information presented in this article.