This article contains affiliate links. If you buy through them, MyrinNew.com may earn a commission at no extra cost to you.
🕒 2 min read
Updated October 23, 2026. Investment in AI infrastructure is a real economic commitment, but a large construction budget does not establish a profitable business. The useful question is how capacity, energy availability, customer demand, and operating cost fit together.
What the energy evidence says
The International Energy Agency’s April 2026 Key Questions on Energy and AI report projects global data-centre electricity consumption rising from 485 TWh in 2025 to about 950 TWh in 2030. It identifies bottlenecks in grids, energy equipment, chips, and approvals. This is a scenario-based outlook, not a guarantee that every announced project will be built or used. Source: IEA 2026 executive summary (CC BY 4.0).
Separate capacity from utilization
A facility’s rated capacity describes what it can support under defined conditions. Utilization describes what customers actually use. Revenue depends on the commercial arrangement, while profit also depends on operating costs and the capital committed. These measures can move in different directions.
Affiliate pick: tech accessories
Belkin US
Belkin BoostCharge Pro 2-in-1 Wireless Charging Pad with Official MagSafe Charging 15W, Black
"BOOSTCHARGE PRO 2-IN-1 PAD WITH MAGSAFE BLACK"," | Belkin BoostCharge Pro 2-in-1 Wireless Charging Pad with Official MagSafe Charging 15W, Black
$44.99
$79.99
Price as of November 5, 2026
When evaluating an infrastructure announcement, ask whether it describes a proposed site, a power agreement, an operating facility, or capacity already serving customers. Treating these as interchangeable can turn a project pipeline into a misleading picture of completed supply.
A practical set of questions
- Demand: Is there contracted customer demand, or only an expectation that it will arrive?
- Power: Is supply connected and deliverable, or dependent on a future grid upgrade?
- Economics: Which costs are included in the claimed margin?
- Technology: What happens if the equipment’s useful economic life is shorter than expected?
- Execution: Which permits, suppliers, and construction steps remain outstanding?
What smaller businesses can learn
The same reasoning applies to buying an AI service. Start with a workload and a service level. Measure cost per useful completed task, including review and correction. Do not infer business value from how large a model is, how much hardware a supplier has announced, or how impressive a demonstration looks.
A sensible pilot defines a baseline, a small group of representative tasks, an error threshold, and a stop condition. If the result is useful, expand deliberately. If it is not, the pilot has still produced information before the business commits to a larger dependency.
The evidence supports taking AI’s physical infrastructure seriously. It does not remove the need to examine each proposed business model on its own terms.
