Free tools Windows power users keep installed
One-click scans. No signup required.
BloombergNEF’s 106GW projection was a real December 2025 forecast, but it is no longer the firm’s latest view. On July 21, 2026, BloombergNEF projected 194GW of U.S. data-center capacity online by 2035, with 118GW by 2030. That is an 83% increase over its earlier estimate and illustrates how quickly artificial-intelligence construction plans are changing the electricity outlook.
What the original 106GW forecast said
BloombergNEF’s report, AI and the Power Grid: Where the Rubber Meets the Road, published in early December 2025, put U.S. data-center power demand or capacity at approximately 106GW by 2035. The estimate was about 36% higher than its previous projection of roughly 78GW.
The increase reflected a rapidly expanding project tracker. BloombergNEF said nearly 150 significant U.S. projects had been added during the prior year, and nearly one-quarter were larger than 500MW—more than twice the previous year’s share. AI campuses require unusually dense, continuous power, so a few very large projects can materially change a regional forecast. Utility Dive’s account of the December forecast provides the contemporaneous figures.
The latest BloombergNEF number is 194GW
BloombergNEF’s July 21, 2026 outlook projects 194GW of U.S. data-center capacity online by 2035 and 118GW by 2030. The new 2035 figure is 83% above the December 2025 forecast, after BloombergNEF identified another 101GW of announced U.S. pipeline capacity.
Do these 3 things before closing this tab:
1Clear out junk files and repair common Windows errors2Scan for outdated or missing drivers - takes under a minute3Repair Windows errors before they cause bigger problems#1 Best Overall
- Save valuable floor space: 6U wall mount server cabinet Dimensions: 13.78" H x21.65" W x17.72" D.Maximum mounting depth is 14.2"
- Keep critical network equipment secure: glass door and side panels are lockable to prevent unauthorized access. Front door can be installed on either side of the front of the cabinet to satisfy your door swing orientation preference
- Easy equipment configuration: Fully adjustable mounting rails and numbered U positions, with square holes for easy equipment mounting with top and bottom punch-out panels for easy cable access
- Durability: Made of high quality cold rolled steel holds up to 110lb (50kg) (Easy Assembly Required)
- PCI & HIPPA and EIA/ECA-310-E compliant
BloombergNEF estimates data centers could represent about 5.9% of U.S. electricity consumption today, 12% in 2030 and 20% in 2035. Those percentages are forecasts, not guaranteed outcomes. Announced facilities can be delayed, downsized, canceled or denied power service. The latest assumptions are summarized by BloombergNEF.
What does “194GW” actually measure?
Gigawatts measure power, not energy consumed over a year. BloombergNEF’s summary describes data-center capacity online; it should not be read as 194GW of continuous demand or as a specific number of annual terawatt-hours.
- Nameplate or nominal capacity: the power capability associated with planned or operating facilities.
- Peak demand: the highest instantaneous draw from the grid or on-site equipment.
- Average demand: the typical draw over time, which depends on utilization and workload scheduling.
- Annual consumption: electricity used over a year, measured in kilowatt-hours or terawatt-hours.
- IT load versus total facility load: servers are only part of a facility’s requirement; cooling, power conversion and other auxiliary systems add to it.
EPRI cautions that announced nominal megawatts are a pipeline indicator rather than a near-term peak-load forecast. Ramp schedules, non-IT loads, on-site generation and flexible operations can all change the load that a utility actually serves. Its 2026 scenarios place U.S. data-center nominal capacity in 2030 between 56GW and 132GW, demonstrating the uncertainty range. See the EPRI executive summary.
Why the forecasts are rising so fast
AI workloads are power-dense
Training and inference clusters use far more concentrated power than many conventional enterprise workloads. Higher server density and more accelerators increase the electrical requirement at a single site.
Quick wins for a faster PC:
Scan for outdated or missing drivers - takes under a minuteDriver Scan →Clear out junk files and repair common Windows errorsFree Scan →Fix the driver behind crashes, sound loss and screen glitchesFind Drivers →Campuses are getting larger
Developers are proposing gigascale campuses rather than isolated buildings. A single project can require hundreds of megawatts and can shift a utility’s expected load growth.
Developers are chasing “speed to power”
Sites with available land, fiber, cooling resources and a plausible electricity path are attractive even when the permanent grid connection will take years. Grid delays encourage developers to consider on-site generation or direct power arrangements.
Rank #2
- Save valuable floor space: 12U wall mount server cabinet Dimensions: 24.25" H x21.65" W x17.72" D. MAXIMUM MOUNTING DEPTH is 14.2".
- Keep critical network equipment secure: glass door and side panels are lockable to prevent unauthorized access; Front door can be installed on either side of the front of the cabinet to satisfy your door swing orientation preference
- Easy equipment configuration: Fully adjustable mounting rails and numbered U positions, with square holes for easy equipment mounting with top and bottom punchout panels for easy cable access
- Durability: Made of high quality cold rolled steel holds up to 110lb (50kg) (Easy Assembly Required)
- PCI & HIPPA and EIA/ECA-310-E compliant
The assumptions are unusually sensitive
Outcomes depend on AI-model efficiency, chip utilization, data-center utilization, project financing, permitting, transmission construction, equipment availability and whether a campus uses grid power, on-site power or both. Better algorithms could reduce electricity per unit of computation; a stronger AI investment cycle could increase total demand.
Where the pressure is concentrated
The United States does not face one uniform data-center shortage. Load is concentrated in particular transmission and wholesale-market regions, including PJM, MISO and ERCOT, with additional growth in states such as Georgia, Ohio, Indiana, Pennsylvania, Louisiana and Mississippi.
In PJM, BloombergNEF projected data-center capacity could reach 31GW by 2030, nearly matching the 28.7GW of new generation the Energy Information Administration expected in that region over the same period. BloombergNEF also warned that ERCOT reserve margins could enter risky territory after 2028 if demand grows faster than supply. These are regional projections, not proof that every American utility will face the same condition. The American Public Power Association summarizes the PJM comparison.
Can the grid supply the buildout?
The answer depends on location, timing and the type of service a project requires. A region may have enough national generation on paper while a proposed campus waits years for a substation, transformer, transmission upgrade, fuel connection or interconnection study.
- Generation: new plants must be built or existing plants retained to cover firm load and reserves.
- Transmission and substations: high-voltage lines, transformers and local distribution equipment can be the binding constraint.
- Interconnection queues: studies, permits and construction often take longer than a developer’s building schedule.
- Fuel deliverability: gas-fired assets need pipeline capacity during stressed periods.
- Reserve margins: fast demand growth can reduce the cushion available during heat waves, cold snaps or generator outages.
The EIA has warned that, in a high-demand case, natural gas could supply much of the incremental generation if low-carbon generation and transmission do not arrive quickly enough. That is a scenario, not a prediction that the entire country will run short of electricity. EIA’s analysis explains the assumption.
How developers and utilities are trying to add power
| Option | Potential advantage | Main trade-off |
|---|---|---|
| Utility-scale generation and grid upgrades | Can serve multiple customers and provide durable capacity | Permitting, construction and interconnection can take years |
| Behind-the-meter gas engines or turbines | Fast, firm power while grid access is delayed | Fuel-price exposure, emissions, air permits and possible stranded equipment |
| Fuel cells | On-site firm output with potentially lower local combustion emissions | High project cost and fuel-supply dependence |
| Batteries and microgrids | Backup, peak management and power-quality support | Batteries alone generally cannot provide multi-day firm supply without major oversizing or another generator |
| Renewables paired with storage | Lower operating emissions and potential contracted supply | Intermittency, land, transmission congestion and the need for firming |
| Nuclear contracts, uprates or restarts | Firm, low-carbon generation | Existing assets may be scarce; new reactors usually cannot solve an immediate 2026–2030 need |
| Flexible workloads and demand response | Can reduce peaks by shifting training or other interruptible computing | Inference and latency-sensitive services are less flexible |
| Reuse of retired power or crypto-mining sites | May offer existing electrical infrastructure and suitable land | Infrastructure condition, local permits and available capacity still require verification |
BloombergNEF reports that developers are turning to on-site gas because grid connections can take years. It presents that strategy as a response to current constraints, not a guarantee that every temporary unit will be retired when grid service arrives. See BloombergNEF’s analysis of gas generation.
Rank #3
- Sturdy:4u server rack is construct from cold rolled steel, with a weight capacity of 110lbs(50kg); Electrostatic powder coat prevents rust and corrosion,quality finish
- Direct use:Open and use, not having to assemble it.Network rack can be placed flat or mounted on the wall,also can be installed vertically under the table
- Design Features:maximum mounting depth of 14 in,cables can be fixed on the side panel;Open frame server rack achieves effortless inspection, replacement and assemble
- Installation:wall mount network rack is easy to install,with instructions or videos for reference;Equipped with multiple accessories, suitable for different needs
- Application:EIA/ECA-310-E Compliant;wall mounted 4u rack fits all 19" racks and cabinets to hold various IT, network, and AV equipment;wall mount rack available in 4U, 6U, and 8U to choose
Who pays for the infrastructure?
The financial effect is determined by contracts and rate design, not by the headline gigawatt figure alone. Utilities may spend on plants, transmission, substations and distribution equipment before a campus reaches full operation.
Costs that may be assigned directly to the data center
- Interconnection facilities and dedicated substations.
- Minimum-payment or take-or-pay commitments.
- Demand charges based on the customer’s highest draw.
- Special large-load tariffs and contributions toward network upgrades.
Costs that may be shared
Broader transmission, generation and distribution investments can enter a utility’s rate base or a regional wholesale market. If a speculative project does not materialize, unused equipment can create stranded-asset risk. State tax abatements and public infrastructure spending can also shift part of the cost to taxpayers.
That does not mean household bills will rise nationwide solely because of data centers. The result depends on the utility’s rules, regional market, financing, customer commitments and whether regulators protect residential and small-business customers from speculative load.
Environmental and community effects
- Emissions: gas generation used for speed or reliability can increase carbon dioxide and other pollutants.
- Air quality: engines, turbines and backup generators require permits and can affect nearby communities.
- Water: cooling systems and upstream electricity production can consume significant water, depending on technology and climate.
- Land and transmission: campuses, substations and new corridors have physical and visual impacts.
- Noise: cooling equipment and generators can create persistent local noise.
- Efficiency: newer chips, cooling designs and power-management systems may reduce electricity intensity per unit of compute.
An independent facility-level academic analysis estimated that hyperscale data centers represented about 1.8% of total U.S. electricity consumption in its central scenario and attributed roughly 54% of their generation supply to fossil sources. Its sample and methodology differ from BloombergNEF’s, so it is context rather than a direct check of the 194GW projection. Read the study.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
What could make 194GW too high—or too low?
- Announced projects may be canceled, delayed, downsized or unable to secure power.
- AI models may become more efficient, reducing electricity per task.
- Chip utilization and cloud demand may be lower than developers assume.
- Training workloads may move geographically or pause during grid emergencies, while inference remains relatively inflexible.
- Transmission, gas turbines, transformers or permits may arrive later than planned.
- Conversely, a stronger AI investment cycle or additional gigascale campuses could push demand above current expectations.
- Different forecasts may count overlapping announcements or rely on the same scarce generation resources.
What the outlook means for households and investors
For personal finances, the practical questions are local: which utility serves you, whether it is in a constrained market, how it allocates upgrade costs and whether regulators approve special data-center tariffs. Watch state commission filings for large-load contracts, minimum bills, transmission plans and proposed rate-base additions rather than assuming a national bill increase.
Investors and businesses should distinguish a signed, financed and permitted campus from an announcement. A project with a binding power contract and a credible interconnection schedule has a different risk profile from a headline pipeline entry. Communities should request public estimates of jobs, tax revenue, water use, emissions, infrastructure cost and protections against abandoned projects.
The Bottom Line
Bottom line: 106GW was BloombergNEF’s genuine December 2025 forecast, not a fabricated number. But the latest BloombergNEF projection is 194GW of U.S. data-center capacity online by 2035, with 118GW by 2030. Neither figure is certain: the outcome depends on which AI projects are built, how efficiently they run, where they connect and who pays for the generation and grid upgrades.
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.
