Spending Billions on Buildings to Sell Something Called the Cloud
Cloud computing is marketed as software and financed like heavy industry. The capital required to deliver it changes how the business should be measured.
Software Economics, Industrial Requirements
Traditional software has a distinctive financial shape: high development cost incurred once, then near zero cost to serve each additional customer. Gross margins are very high and capital requirements are modest.
Cloud infrastructure breaks this. Delivering computing, storage and networking to customers requires physical capacity, and serving more customers requires more of it. The business carries genuine marginal cost in the form of hardware, electricity and cooling.
A cloud provider does not sell software. It sells access to buildings full of machines, on a subscription, and it must build the buildings first.
What the Capital Buys
The spending falls into several categories with different lives and different economics.
| Asset | Characteristic |
|---|---|
| Land and shell construction | Long lived, slow to build |
| Power and cooling infrastructure | Long lived, constrains capacity |
| Servers and networking equipment | Short lived, replaced regularly |
| Fibre and interconnection | Long lived |
The third row drives the economics. Server equipment has a useful life measured in a handful of years, so the depreciation charge is large and continuous. A provider must keep spending simply to maintain existing capacity, before adding any.
This makes the assumed useful life of server equipment a meaningful accounting judgement. Extending it reduces annual depreciation and raises reported operating profit, and several large operators have revised these estimates upward over time. The change is disclosed and its earnings effect can be substantial, which is why it deserves attention when comparing periods.
Capacity Must Precede Demand
Data centres take years to plan, permit, build and connect to power. Capacity therefore has to be committed well before the revenue that will use it exists.
This creates an unavoidable timing problem. Build too slowly and customers cannot obtain the capacity they need, which pushes them to competitors in a market where switching at the margin is feasible. Build too quickly and expensive assets sit underused, depreciating against no revenue.
Utilisation is consequently the central operating variable. The same physical asset can be highly profitable or loss making depending on how much of it is sold, and the fixed cost base means the profit swing between the two states is large.
Power Is the Binding Constraint
The limiting factor for expansion has shifted from capital availability to electricity. Large facilities require substantial and continuous power, and connecting them depends on grid capacity, interconnection queues and local generation.
This has made site selection a question of energy availability rather than proximity to customers, and it explains why operators sign long term power agreements and invest in generation directly. It also introduces a cost input that is volatile and, in some markets, politically contested.
Why the Revenue Is Sticky
Against these capital demands sits an unusually durable revenue base. Once an organisation has moved workloads onto a platform, moving them again is expensive and disruptive. Data must be transferred, applications rebuilt against different services, and staff retrained.
Providers reinforce this through proprietary managed services that have no direct equivalent elsewhere, and through charges for moving data out. The result is high retention and revenue that expands as customers grow, which supports committing capital years in advance.
Long term contracted commitments give providers visibility into future demand, and disclosed backlog figures are a useful indicator of whether current construction is supported by contracts already signed.
How to Read the Numbers
Gross margin alone is misleading, because it may exclude the depreciation that represents the real cost of the capacity. The more informative measures are capital expenditure relative to revenue, the trend in that ratio, disclosed backlog or remaining performance obligations, and any change in depreciation assumptions.
A provider whose capital spending is accelerating faster than revenue is either investing ahead of contracted demand, which is a bet, or fulfilling commitments already made, which is not. The backlog disclosure is what separates the two.
The Bottom Line
Cloud infrastructure pairs subscription revenue with the capital intensity of heavy industry, and the assets involved depreciate quickly enough that maintenance spending alone is substantial. Capacity must be committed years before the demand that fills it, which makes utilisation the variable that determines profitability. Assessing such a business requires looking at capital intensity, contracted backlog and depreciation assumptions rather than at gross margin, which describes a software business the operator does not have.