Crashes, No Sound, or Screen Glitches?
Random freezes, missing sound and display glitches usually trace back to one bad driver. Find and replace yours safely.Free scan · under a minutePC Slower Than It Used to Be?
A free scan shows the junk files, broken settings and background clutter dragging Windows down - then fixes them in one click.Free scan · Windows 10 & 11AI’s electricity needs could slow the construction and expansion of U.S. data centers, but that does not mean the country is about to run out of energy. The more immediate risk is that a particular region cannot deliver enough reliable power to a proposed site quickly enough—because generation, transmission, substations, equipment, or grid approval is not ready.
Former Google CEO Eric Schmidt made that warning at a House Energy and Commerce Committee hearing on April 9, 2025, arguing that AI facilities could reach 1 to 10 gigawatts and that energy infrastructure is becoming part of the competition to build advanced AI. The scale is striking, but a proposed campus is not the same as an operating facility drawing its full planned load.
What Eric Schmidt said about AI and energy
In written testimony to the House Energy and Commerce Committee, Schmidt, then chair of the Special Competitive Studies Project, argued that AI development is moving faster than the energy system and public processes can adapt. He said planned AI data centers could require 1 to 10 gigawatts and called for an “all of the above” approach to energy supply. His testimony also linked reliable, abundant electricity to U.S. competitiveness. Read Schmidt’s written testimony and the House hearing page.
A gigawatt is 1,000 megawatts of power. A continuous 1-GW load would consume 8,760 gigawatt-hours in a year if it operated at full power every hour; actual annual consumption depends on utilization. A 5-GW campus would represent a very large, sustained load, but the comparison with power plants depends on their output and operating conditions. The 1-to-10-GW figure describes a possible facility scale, not a current average or proof that every announced campus will be built and fully energized. Projects may be phased, reduced, delayed, or fail to receive grid connections.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
#1 Best Overall
The committee’s summary of the hearing also highlighted Schmidt’s concerns about the scale of planned facilities, the need for substations, and delays in natural-gas turbines. Those are pieces of a broader infrastructure problem, not evidence that one type of power plant alone can resolve it. House committee summary.
What the energy-demand numbers do—and do not—show
The International Energy Agency (IEA) estimates that data centers worldwide used about 415 terawatt-hours (TWh) of electricity in 2024. Its base case projects roughly 945 TWh in 2030, more than double the 2024 estimate. U.S. data centers accounted for about 180 TWh in 2024, close to 45% of the global total, and the United States is expected to have the largest absolute increase. These are estimates and projections, not guaranteed outcomes. IEA, Energy and AI.
Annual electricity use and peak demand answer different questions. TWh measures energy consumed over time; gigawatts and megawatts measure the rate of power delivery at a moment. The IEA projects that data centers could rise from roughly 6% of U.S. peak electricity demand today to 13% by 2030. A growing annual total raises the amount of electricity the system must supply, while a high local peak can strain equipment and networks even if the national annual supply appears adequate.
That is why “the U.S. is running out of electricity” is too broad. The central concern is whether power can be generated, moved, and delivered reliably to specific sites on a schedule that matches construction and computing plans. The Atlantic Council identifies transmission shortages, aging infrastructure, uncertain load forecasts, and local acceptance as important constraints. Atlantic Council analysis.
The Tool Desk
Outbyte PC Repair FREERepair Windows errors before they cause bigger problemsFix Now →Outbyte Driver Updater FREEScan for outdated or missing drivers - takes under a minuteDriver Scan →Why AI data centers can put unusual pressure on the grid
AI training runs large groups of accelerators at once, while inference—the process of serving model responses—can create continuing demand as users and applications make queries. High-performance facilities need more than a large electricity supply. They need dense power delivery, cooling, redundant feeds, substations, backup systems, and enough network capacity. When many large facilities cluster in one region, their combined demand can grow faster than local infrastructure.
Rank #2
Not every workload has the same flexibility. Interrupting a long training run may waste computing time or delay a job. Some inference can be batched, delayed, or shifted to another region, depending on how quickly a service must respond. The Atlantic Council estimates that energy represents roughly 2% to 6% of AI training costs in its analysis. That cost share does not make power physically optional: a facility without adequate electricity cannot run its equipment, regardless of electricity’s share of a model’s overall cost.
Schmidt’s 1-to-10-GW range should therefore be read as a warning about possible frontier-scale plans, not a description of typical data centers. The Institute for Progress has argued that the largest AI clusters could approach 5 GW by 2030 if current compute-growth trends continue; that is an analytical projection, not an established forecast. Institute for Progress analysis.
Where a power bottleneck actually forms
Power systems have several linked stages. Having fuel or a power plant somewhere in the country does not guarantee that a data center can draw electricity at a specific site. The relevant questions include how much power the facility needs, where and when it needs it, how firm the supply must be, and what upgrades are required to connect it.
Free tools Windows power users keep installed
One-click scans. No signup required.
Generation
New gas, nuclear, hydro, wind, solar, geothermal, and other generation can contribute to supply, but a resource is not the same as a completed power plant. Projects require equipment, permits, financing, construction, fuel or transmission arrangements, and an approved grid connection. Schmidt specifically warned about natural-gas turbine delays; the House committee summary described his concern that turbine backlogs and higher costs could impede near-term development. Gas can provide firm generation, but it also brings emissions, fuel-delivery, and permitting considerations.
Transmission and substations
High-voltage transmission moves electricity across regions; substations and local distribution systems bring it to a facility. A data center can be near power generation and still lack the lines, transformer capacity, or substation equipment needed to connect. Transmission projects often involve complex planning and approvals, while major electrical equipment can be difficult to procure on a data-center construction schedule.
Interconnection and permitting
A grid interconnection is a technical and regulatory process, not a simple place in line. Studies assess reliability and identify network upgrades; parties must determine costs and approvals before a project can connect. A data center may be funded and under construction while its power connection remains uncertain. Permits for generation, transmission, pipelines, and local facilities can add more time.
Reliability and local acceptance
Operators need power through peak demand and equipment failures, not only on an average day. Heat waves, cold snaps, generator outages, fuel shortages, or transmission failures can reduce available supply. Projects may also face local concerns over land, noise, water use, air emissions, and who pays for infrastructure built to serve a very large new customer.
Why the constraint is regional, not simply national
Electricity markets and grids are regional. A national supply estimate can obscure a shortage of transmission or substation capacity in the area where companies want to build. Northern Virginia is described by the Atlantic Council as the world’s largest data-center market by operational capacity; Texas, Georgia, Ohio, and other states are also expanding as data-center hubs.
The House committee said signed agreements in Central Ohio could bring data-center demand there to 5,000 MW by 2030. That is a local projection based on those agreements, not proof that all 5,000 MW will be connected or consumed. Still, concentrated plans can require a utility to build generation and network infrastructure for a relatively small number of very large customers. House committee summary.
How flexibility could relieve some grid pressure
AI workloads do not all have to run at maximum power at every moment. The IEA estimates that U.S. data centers could potentially integrate up to 70 GW of additional capacity into the existing system if operators reduced grid demand for approximately 1% of the time. This is a model-based estimate, not a guarantee of spare capacity in every region. The IEA says stress events generally last a few hours and points to workload shifting, storage, and backup generation as possible forms of flexibility.
- Move non-urgent training or batch jobs to lower-demand hours.
- Shift eligible inference between locations or delay selected requests.
- Use batteries to reduce demand from the grid during short peak periods.
- Coordinate backup generation or on-site supply where regulations and operating conditions allow.
- Contract for demand response, with clear terms for when and how much load can be curtailed.
Flexibility has limits. A frontier training run may not tolerate repeated interruption without cost, and some inference services require rapid responses. A 1% reduction in demand time does not mean a data center can operate without a dependable power supply the other 99% of the time. It means that carefully timed reductions could help the grid handle peak stress while infrastructure catches up.
Recommended Free Tools
Efficiency may slow demand growth, but it may not reverse it
More efficient chips, cooling, algorithms, model compression, and accelerator utilization can reduce the energy needed for a given computation. Smaller specialized models and better scheduling may also avoid using the largest systems for every task. If AI adoption grows more slowly than expected, total demand could be lower than current projections.
But lower energy per task does not automatically mean lower total electricity use. Cheaper queries can encourage more use, while longer reasoning or agentic tasks may require more computation per request. AI is also being considered for more applications, from coding and office software to science, robotics, and industrial systems. The Atlantic Council notes uncertainty about inference growth and model “thinking”; both efficiency and expanding use shape future demand.
A serious forecast must therefore track not just the efficiency of a model or chip, but also how often systems are used, how much computation each task consumes, and whether workloads can move across time and place.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.Who pays, and what are the environmental trade-offs?
Meeting data-center demand can require new generation and grid upgrades, and those costs do not disappear because the customer is a technology company. Utilities and developers must decide how much infrastructure a project pays for and whether other customers bear any costs. If upgrades are spread across a utility’s customer base, households and businesses may face higher bills; if the data center pays more directly, its operating costs rise. The details depend on local regulation and negotiated arrangements.
The generation options involve different trade-offs:
- Natural gas: Can provide firm power, but adds greenhouse-gas emissions and may require pipeline capacity; turbine availability and permitting can constrain timelines.
- Nuclear: Can provide firm, low-carbon electricity, but new projects involve substantial financing, licensing, construction, and fuel-cycle considerations. It is not automatically a near-term fix.
- Wind and solar: Can supply substantial electricity, but output varies with weather and time. Transmission, storage, flexible demand, or other firm capacity may be needed to match a constant load.
- Batteries: Can help shift electricity and cover short-duration peaks, but their usefulness depends on storage duration, cycling, grid connection, and local market rules; they do not by themselves replace all firm generation.
Communities also weigh land use, noise, water consumption, local air pollution, and the distribution of economic benefits. Data centers can bring construction activity, tax revenue, and utility investment, but those gains do not settle how costs or environmental impacts should be allocated. A House hearing held in 2026 focused explicitly on meeting growing demand while protecting ratepayers, underscoring that cost allocation is now a central policy question. House hearing notice; committee hearing page.
What the energy issue could mean for U.S. competitiveness
Schmidt framed energy as part of the U.S.-China AI competition. The argument is straightforward: if companies cannot obtain reliable power and connect facilities on time, their ability to expand compute can be constrained. That is a strategic judgment, not proof that electricity alone will determine which country leads in AI.
The United States has natural-gas resources, major technology companies, capital markets, semiconductor and cloud ecosystems, and a mix of existing generation. It also faces fragmented regulation, aging infrastructure, long interconnection processes, equipment backlogs, and local opposition. The testimony and energy analysis cited here do not establish that China has solved its own energy constraints or that one country is certain to win. The more defensible point is that infrastructure execution can affect how quickly each country can turn investment into usable computing capacity.
Quick wins for a faster PC:
Clear out junk files and repair common Windows errorsFree Scan →Fix the driver behind crashes, sound loss and screen glitchesFind Drivers →How to judge whether a specific project is power-constrained
For a data-center proposal, the headline capacity alone is not enough. A useful assessment asks:
- How much power? Distinguish peak MW from average load and annual TWh; clarify expected utilization and whether the figure covers one building or a phased campus.
- Where? Identify the utility, grid region, transmission zone, and available substation capacity rather than relying on a national supply figure.
- When? Separate the first phase and its expected connection date from a campus’s ultimate announced capacity.
- How firm? Determine what workloads can shift or pause, what backup is available, and whether the operator requires around-the-clock low-carbon supply.
- Who pays? Check whether the data-center operator, utility, or other ratepayers fund network upgrades and new generation.
- What happens under stress? Consider heat, cold, outages, fuel constraints, and delayed upgrades, along with the site’s backup and demand-response plans.
The bottleneck thesis would weaken if projected AI loads do not materialize, facilities are phased or relocated to areas with spare capacity, efficiency outpaces adoption, or flexible workloads and new infrastructure arrive in time. It would strengthen if announced loads become firm connection requests while transmission, equipment, and generation projects continue to lag.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
