Data-center and warehouse investments are global, but many of their effects are tied to one substation, watershed, freeway or neighborhood. A useful local account asks what infrastructure a project needs, when it needs it, who pays, and what has actually been built.

Large forecasts do not equal a local electricity bill

The International Energy Agency’s October 2026 analysis says the world’s largest data-center operators are set to spend more than $700 billion in capital expenditure in 2026, mostly on new data centers. It also says the effect on electricity bills is not clear-cut: new grid or generation costs can raise bills if they are spread across all users, while a large user can help spread existing fixed costs when it pays for its share. The allocation depends on local tariffs, utility investments and the project’s actual load.

That distinction matters because a project announcement is not a metered load. Lawrence Berkeley National Laboratory’s 2025 update, published in June 2026, models U.S. data centers at 649 terawatt-hours in its 2030 reference case, or 11.8% of projected U.S. electricity use. Its sensitivity scenarios span 521–843 TWh and 9.5–15.3%. These are modeled estimates, not observed 2030 consumption.

United States · 2030 reference case

649 TWh

LBNL model; 11.8% of projected national electricity use. Scenario range: 521–843 TWh.

United States · 2023 indirect water

~800B L

LBNL estimate for water consumed in electricity generation serving data centers—not on-site cooling water.

Grid congestion is another real cost signal, but it must not be assigned to one sector without evidence. The IEA reports $12 billion in U.S. congestion costs and €4.3 billion in the European Union in 2024. Those are system-wide estimates, not data-center bills or costs caused only by data centers. The newer IEA analysis says large-load tariffs can reduce cost-shifting risks when they account for reserved capacity, long-term commitments, phased connections and the possibility that a project never connects.

Water is a basin question as well as a facility question

The IEA estimates that nearly 40% of identified AI data-center capacity is in areas with high or extremely high baseline water stress. This is a share of identified capacity, not a count of operating sites. Water-use choices also interact: reducing on-site cooling water can require more electricity, and water used upstream to generate electricity is part of the footprint.

LBNL’s U.S. 2024 report estimates that data centers used about 176 TWh of electricity in 2023 and had nearly 800 billion liters of indirect water consumption associated with that electricity use. This is a national, model-based estimate based on grid mix. It is not a measurement of direct water withdrawals at individual facilities. Local reporting should separate cooling-water withdrawals and consumption from the indirect water tied to power, and show the source, season and watershed for each.

Warehouses bring freight activity into specific neighborhoods

Warehouse impacts can include truck trips, vehicle miles, yard equipment and traffic around nearby homes and roads. A facility’s floor area alone does not show how much freight it moves or where its emissions occur. South Coast AQMD’s WAIRE program illustrates one regional approach: it applies to warehouse buildings of at least 100,000 square feet in that air district and requires covered operators to earn points through actions, approved plans or a mitigation fee. It is a local program, not a national warehouse rule.

For other regions, the same ground-level questions remain useful: how many truck visits and miles are expected, what vehicles and yard equipment are used, how are freight emissions addressed, and which road, air-quality or emergency-access reviews apply?

Keep projects, policies and company programs distinct

The current Directory spans different regions and stages. At the U.S. Savannah River Site, DOE/NNSA selected Amentum to enter lease negotiations for a proposed 1-GW data center with about 2 GW of on-site generation. The agency says this is not a final lease; safety, security, permitting and other approvals remain pending. California’s September 2026 laws address water and electricity disclosure, grid-upgrade costs and land-use review. India’s NITI Aayog framework offers state-level evaluation guidance; it is not a binding rule or facility approval.

Supply-chain examples also need accurate labels. Tesco reports construction underway at its London Gateway distribution centre, with opening expected in 2029; its cited results describe a multi-million-pound investment but no exact project cost. Walmart’s Prepaid Consolidation program routes supplier shipments across 42 U.S. regional distribution centres and uses a price-per-case rate; Walmart describes it as a logistics program, not a new warehouse build. General Mills’ FY2026 filing reports about $101 million in expected restructuring charges, including about $33 million in cash, for a multi-year competitiveness initiative, and a separate warehouse/distribution footprint of about 16 million square feet, nearly all leased. Those restructuring figures are not a warehouse construction budget or a site operating-cost estimate.

A practical local infrastructure ledger

California’s new data-center laws and South Coast AQMD’s warehouse program show that local and state rules can define some of these obligations. Elsewhere, the same information can help residents, utilities and planners compare proposals without assuming that every project has the same footprint or who bears its costs.

Evidence limits: This is an editorial synthesis of public sources, not a peer-reviewed paper, field study, project audit or cost estimate. LBNL figures are national models; IEA figures are forecasts or system-wide analysis; company project stages and corporate costs are operator-reported. Where no project-level cost is published, none is inferred.

Sources and evidence stage

Explore Open DirectoryRead the local infrastructure cost-accounting noteSuggest a correction