Data Centres: A Grid Connection Is Not Just an Electricity Price
Beyond electricity price, the hidden constraint on data-centre growth is firm power that can be delivered at the needed location and time.
One-Sentence Insight
For a data centre, one megawatt theoretically available is not one megawatt that can be connected when needed. Location, grid capacity and connection terms have become part of the computing market.
Why This Market Is Changing
Developers can acquire servers, land and construction contracts without securing deliverable power at the required location and date. Capital then sits idle. Real access depends on upgrades, application queues and technical commitments. These local conditions can matter as much as the quoted electricity tariff.
Simple Example
You want to open a bakery with a large electric oven. Your utility quotes an attractive tariff, but the local transformer is full. The price is real; your ability to operate is not yet secured. Replace the oven with a data centre and the distinction becomes a major investment risk.
The Actual Access Right
The route to operation has several gates: application, load and network studies, possible securities, upgrade work, schedule and final connection terms. A commercial indication is not a deliverable megawatt. Certainty varies by stage and jurisdiction. Separate requested, studied, contracted and energized capacity.
Flows and Actors
A developer submits intended power, load profile and timing. The network operator evaluates constraints and upgrades. An energy retailer may sell electricity without controlling physical interconnection. Financiers price delay risk. Authorities protect reliability and other consumers. Confusing these roles leads to poor location decisions.
Evidence
In its April 2025 Energy and AI study, the IEA estimates data centres used about 415 TWh in 2024 and could use 945 TWh by 2030 in its base case. It estimates that around 20% of planned data-centre projects could face grid-related delays if bottlenecks persist. This concerns the share of projects, not a measured share of their total power capacity. These are global estimates and scenarios, not forecasts for any specific site.
Connection Economics
Cheap land can be expensive if a two-year grid delay postpones revenue or requires major upgrades. A dearer site with firm access may reduce financing and delivery risk. Decision-grade economics combine land, electricity, upgrades, likely energization date, operating flexibility and cost of delay. No global average substitutes for a local binding offer.
Constraints
Scheduling software cannot create transformers, cables or permits. Speculative applications may crowd queues, but queue rules differ across countries. On-site generation can help without eliminating fuel, emissions or resilience requirements. Load flexibility also depends on workloads and customer commitments; not every computing task can be paused.
What People Miss
People compare per-kilowatt-hour tariffs without asking when the kilowatts are physically deliverable. Uncertain future access is not economically equivalent to firm capacity today. Even a renewable power purchase agreement does not, on its own, prove that the physical grid can deliver the requested load at a particular address.
Critical View
The IEA does not say exactly one-fifth of all projects will certainly fail. Its figure refers to planned projects potentially facing delays, not a measured share of power capacity delayed. Connection timelines change with policies and investment. Phasing, siting and load flexibility may help particular sites, not all. International comparisons require consistent definitions of connection stages.
Sidy's Synthesis
My synthesis: AI infrastructure now has a geography of genuinely usable power. A data centre’s megawatts are more than a slide-deck number: they must be deliverable, reliable and economically viable by a stated date.
I would compare projects by firm megawatts at a specified date, graded by evidence: indicative letter, completed study, accepted agreement, completed upgrade or verified energization. This prevents an application queue position being priced like a functioning asset. A slightly smaller site with earlier firm power may be worth more than a larger promise with no credible date.
AI and Future Lens
AI could analyse workloads to find tasks that can shift in time without harming service quality. It might help process applications and identify scheduling conflicts in grid projects. It cannot grant a permit or conjure physical capacity from a forecast.
As computing supply becomes more specialized, customers may trade off site, timing, reliability, cost and environmental impact. Comparison systems must expose uncertainty and protect sensitive grid data. A single 'best country for data centres' score would obscure the actual decision.
Build From This
Build: a site comparison record for data centres or other major industrial loads. For each site, log requested power, documents supporting access, milestone dates, known upgrades, flexibility options and a range for delay costs. Timestamp every item and label it verified, declared or unknown.
Pilot with five real sites using public evidence and authorized network-operator responses. Never invent contracted capacity. Measure selection time saved and undocumented assumptions identified before money is committed.
Do not sell queue positions or promise megawatts you do not control. Proceed only if decision-makers can improve site choices using inspectable evidence.
First Steps
For a live project, request four items: firm power, date, conditions and required upgrades. Compare two sites using an explicit hypothetical delay-cost scenario.
Remember This
An electricity price is not a connection right. Promised power is not delivered power. Timing, reliability and conditions determine the economic value of megawatts.
Primary sources
Facts, figures and quotations should be traceable to the sources below. Sidy's synthesis is labeled as synthesis and does not replace sourced facts.
