For Windows users, IT leaders and organizations planning cloud or AI-heavy workloads, this is no longer only an energy-industry issue. Power availability can affect where capacity is developed, how confidently providers can plan expansions, and whether a proposed facility represents usable computing capacity or merely a project waiting for a connection. For households and policymakers, the harder question is who pays when very large new loads require new generation, transmission and delivery equipment.
Demand is rising faster than simple headlines suggest
The International Energy Agency’s Base Case puts US data-center electricity consumption at 183 terawatt-hours in 2024 and 426 terawatt-hours in 2030. That is a substantial projected increase over six years. It is a scenario-based forecast, however, rather than a measurement of future operating data centers or a guarantee that every announced project will be energized on schedule.
That distinction matters. Electricity demand is not the same thing as a completed data center, a utility interconnection request, a signed power contract, or construction already underway. Each has a different likelihood of turning into a running facility. A developer may have land, permits, financing or customer interest while still lacking the transmission access and generation needed to operate at full scale.
The available evidence therefore supports a more precise conclusion than the most dramatic framing: data-center expansion is placing exceptional pressure on the systems that connect large loads, and that pressure is already changing investment plans and regulation. It does not supply one national measurement that compares all US data-center construction against all grid buildout on a single timetable.
The connection queue can be the real schedule
The physical grid takes time to expand. The IEA estimates that grid constraints could delay around 20% of globally planned data-center capacity scheduled for construction by 2030. It also reports that new transmission lines can take four to eight years to build in advanced economies.
Those figures help explain why a large facility’s development timeline and its power timeline can diverge. A planned site may require upgrades well beyond its own property: additional local delivery infrastructure, transmission capacity, or enough dependable generation to serve the new load. The result is that a company can be ready to develop computing space before the relevant power system is ready to serve it.
Queue timing is also highly regional. The IEA lists typical US connection queues of one to three years, while identifying Northern Virginia as a market with waits of up to seven years. The longer figure should not be treated as a national norm. It is instead an important illustration of how a highly sought-after location can face a distinctly tougher interconnection problem than the countrywide range suggests.
This is a critical point for enterprise technology planning. An announced expansion in a major data-center market is not, by itself, evidence that capacity will be available at the desired date. Customers evaluating long-term compute needs should distinguish between operational capacity, capacity under construction, and capacity whose electricity connection remains subject to grid processes.
More power plants are forecast, with gas carrying much of the modeled load
One forecast from Moody’s Ratings, reported by Bloomberg, estimates that the US data-center buildout would require $110 billion to construct 45 gigawatts of new generation through 2030. Bloomberg’s account says more than 30 GW of that modeled new supply would be gas-fired. Most of the remaining supply would come from solar and storage, with less than 5% tied to nuclear restarts.
This is an analytical projection, not an announcement that $110 billion has already been committed or spent. The underlying Moody’s report was not publicly available in the material reviewed, so its detailed assumptions, regional allocations and sensitivity cases cannot be independently assessed here. Still, the reported mix is revealing: the modeled response is not simply more renewables, nor simply a return to nuclear. It relies heavily on gas-fired generation while adding solar and storage.
That has practical and policy consequences. The generation mix determines more than the construction bill. It shapes fuel exposure, permitting requirements, local environmental debates and the type of grid support available when demand is high. None of those outcomes can be assumed from a national capacity number alone, because the operational need is local and timing-dependent.
Developers are also pursuing on-site natural-gas generation when grid connections are slow. The IEA says a large number of such projects are advancing, largely in the United States. But it also notes that many remain at an early stage. On-site generation is therefore a genuine response to the bottleneck, not evidence that data centers can universally bypass the grid.
Even facilities with their own generation still face questions about construction timing, fuel arrangements, reliability and how their operation interacts with the surrounding power system. The broader implication is that the power constraint may move rather than disappear: from a transmission queue to a generation-development, permitting or fuel-supply challenge.
A big project pipeline is not the same as delivered capacity
A June 2026 report on Jefferies research found that only 12 GW of the 24 GW of US data-center capacity scheduled for that year was under construction. The reported obstacles included permitting, interconnection, energy supply, labor and contracting issues.
That finding cuts against the assumption that data-center plans automatically become physical infrastructure. It also offers a useful caution for market analysis. A schedule can describe developer intent, but only construction and eventual energization turn that intent into live capacity.
The original Jefferies research note was not independently retrieved, and announced capacity can include requests that do not become actual load. The 12 GW figure should therefore be understood as reported analyst evidence of execution risk, not a definitive nationwide inventory. Its central lesson remains sound: the supply chain for data centers is more than servers and buildings. It includes permits, contracts, workers, generation and a power-system connection.
For organizations that run Windows workloads in hosted environments, the immediate effect may not be a visible change to a desktop or server. The nearer-term impact is on capacity planning. Businesses should not treat broad regional announcements as a substitute for provider-specific commitments on deployment timing and available power. Where a workload has fixed compliance, latency or business-continuity requirements, decision-makers should ask how location and power availability affect the proposed timeline instead of assuming that every major market can expand identically.
Wholesale power pressure is real in PJM, but retail bills are less certain
The public debate often jumps from data-center growth to a simple prediction of higher household electricity bills. The evidence is more complicated.
PJM’s independent market monitor estimated that data-center load added $11.26 per megawatt-hour to total wholesale power prices in the first five months of 2026, a 24.4% increase. That is significant evidence of a recent impact in one major wholesale market. At the same time, the monitor stated that data centers were not responsible for all of the higher capacity costs.
The qualification is essential. A wholesale market-price effect is not automatically the same as a nationwide retail-rate increase. Retail bills depend on local utility regulation, rate design, contracts, the allocation of new infrastructure costs and other factors. The question is not merely whether new infrastructure costs money; it is which party is assigned those costs.
There is also countervailing historical research. A 2026 preprint estimated that data centers modestly reduced average US retail electricity rates from 2015 through 2024, while cautioning that future supply constraints could reverse that result. As a preprint, it should not be treated as the final word. But it is an important corrective to claims that data centers have already raised retail rates everywhere.
Taken together, these findings establish neither universal consumer harm nor a guarantee of consumer benefit. They show why national averages can obscure regional reality. PJM provides evidence of a material recent wholesale effect; the historical national estimate points in the other direction for average retail rates over an earlier period. The future outcome will depend heavily on whether power systems can add capacity in time and whether regulators prevent cost shifting.
Regulators are moving from forecasts to rules
Federal and state authorities are beginning to formalize the problem as one of large-load integration and cost allocation.
In June 2026, the Federal Energy Regulatory Commission ordered all six of its jurisdictional grid operators to justify or reform tariff provisions for large loads. The reform areas explicitly include preventing cost shifting and improving transparency into transmission costs. That action does not itself decide who will pay in every region, but it confirms that regulators see existing rules as needing scrutiny in the face of large new electricity customers.
Texas has adopted more immediate constraints. ERCOT and the Public Utility Commission of Texas approved a batch process intended to connect large users only in quantities and locations that the grid can reliably support. Separately, the Texas governor directed that a data-center audit be completed before queued projects could move forward.
Texas is thus not simply an example of rapid expansion. It is also evidence that a state can welcome major development while requiring more disciplined sequencing when grid reliability is at stake. The policy model is moving toward a basic test: a project may be economically attractive, but can it connect at the requested size and location without creating unacceptable system risk or shifting unsupported costs?
The White House-backed Ratepayer Protection Pledge addresses the same concern by asking participating operators to build, bring or buy power supply and to fund delivery infrastructure required by their facilities. Its intent is clear: large new loads should contribute to the resources they require. But the pledge is voluntary and has no federal enforcement mechanism. It should not be described as a guaranteed shield against higher consumer bills.
What to watch next
The key question through 2030 is not whether demand forecasts are large; they are. It is whether the electrical system can translate forecasts into dependable, locationally appropriate supply fast enough, under rules that allocate costs transparently.
For technology buyers, several indicators will be more useful than headline construction announcements:
- Whether planned capacity is actually under construction rather than merely proposed.
- Whether a site has a viable electricity connection on a stated schedule.
- Whether the relevant region faces unusually long queues, as Northern Virginia does in the IEA’s example.
- Whether new supply and delivery infrastructure are being funded by the large load, the utility, other customers, or a combination of them.
- Whether new large-load tariff rules alter project timing or economics.
The data-center boom may ultimately bring new generation and grid investment that supports broader demand. That is a possible outcome, not an established one. It depends on siting, generation choices, infrastructure delivery, utilization and rate design.
For now, the most defensible assessment is that US data-center demand is exposing a grid-planning problem that had been easier to defer when large new loads were less concentrated and less urgent. The constraint is not solely the number of power plants, nor solely the number of data centers. It is the ability to align computing growth, generation, transmission, connection rules and consumer protection on timelines that currently do not always match.