🔌 DOE + LBNL + S&P Global 451 Research · 2024–2026
Virginia Data Centers Already Consume More Power Than The State Makes
176 TWh consumed by US data centers in 2024. Virginia: 1 in 5 kWh goes to data centers. Texas hits its capacity wall during the next bad summer. You're subsidizing ChatGPT on your electricity bill.
176 TWhUS data center power 2024 (4.4% of all US electricity)
1-in-5Virginia kWh going to data centers (Dominion)
2027–28When top states hit their capacity wall
$29BRate increases utilities seeking in 2025 (you pay this)
Choose your depth. The data doesn't change — just the explanation.
AI data centers are eating electricity at an extraordinary rate. In Virginia, 1 in every 5 kilowatt-hours of electricity now goes to data centers. By 2028, data center demand in top states may exceed their entire available surplus power.
US data centers consumed 176 TWh in 2024 — 4.4% of all US electricity. The DOE projects this will double or triple by 2028. Virginia has already hit 20% of grid load from data centers. The GPU power curve: each generation uses 40-50% more watts, though inference efficiency per token is improving.
Forward projections have wide uncertainty ranges: LBNL conservative (efficiency gains realized) to BCG worst-case (1,050 TWh). The BCG figure is an extreme scenario, not a consensus forecast — it's presented alongside LBNL's range on this page but should be weighted accordingly. Liquid cooling (immersion and direct-to-chip) is commercially deployed at scale by Google, Meta, Microsoft — delivering 40-50% PUE reductions. GPU TDP increases don't map linearly to grid demand. The 'capacity wall' is defined as when projected data center demand exceeds available excess generation capacity in PJM interconnection (Virginia's grid).
LBNL data center energy data: eta.lbl.gov/publications/united-states-data-center-energy. EIA state electricity data: eia.gov/electricity/state. Dominion Energy Virginia load data: dominionenergy.com/about-us/electric-grid. GPU power data: techpowerup.com/gpu-specs. OpenAI/Anthropic compute consumption: not publicly disclosed — use inference pricing x volume as proxy.
State-by-State: Who Hits the Wall First
This isn't a national average story. The crisis is hyperlocal. Virginia is already past the tipping point. Texas's grid reserve margin disappears during the next heat wave. These are physical constraints that no amount of investment can fix quickly.
🔴 Virginia
12.1 GW
Data center demand 2025. State generation: ~73 TWh. Virginia imports power — demand already exceeds in-state generation.
⚠️ Already past the wall
🔴 Texas (ERCOT)
9.7 GW
ERCOT summer peak reserve margin was 7.4% in 2023. Data centers = ~13% of peak capacity. At current growth: no reserve margin during summer 2026-2027.
⚠️ Summer 2026–2027
🟡 Ohio
3.5 GW
Fastest growth corridor. 40% YoY growth rate. Fastest path to capacity constraint in the Midwest.
⚠️ 2028–2029
🟡 Georgia
2.8 GW
Major hyperscaler investment. Google, Microsoft, AWS all building. Grid under pressure.
⚠️ 2028–2030
🔵 California
5.2 GW
Large but spread across large grid. Renewable capacity buffer. Regulatory environment slowing growth paradoxically helps.
Watching: 2030+
🔵 Arizona
2.1 GW
Hot climate = high cooling costs. Water scarcity adding constraint alongside power.
Watching: 2029+
The Projection Curves — How Bad Does It Get?
LBNL conservative estimate: 325 TWh by 2030. BCG worst case: 1,050 TWh — that's equivalent to 71% of Texas's total annual generation. All of it, for data centers.
US Data Center Power Demand: Three Scenarios (2020–2030)
Low: LBNL conservative. Mid: DOE consensus. High: BCG worst case. The range reflects uncertainty in AI training demand, efficiency gains, and economic conditions.
State Data Center Power Demand 2025 (GW)
Top 8 states. Virginia and Texas already in crisis territory. Ohio is the fastest-growing.
GPU Power Progression — Why Demand Is Accelerating
Each GPU generation requires more power. B200 = 1,000W vs H100's 700W. Training large models: 25+ MW continuous. Requirements doubling annually.
Comparing to Things You Know
US Data Centers 2024
176 TWh
NYC total electricity
50 TWh
BCG worst case 2030
1,050 TWh
Texas total generation
~480 TWh
Training one LLM
25+ MW
NVIDIA B200 per GPU
1,000W
EIA: DC % of commercial electricity by 2050
20%
DOE: demand doubles or triples
by 2028
🔌 You're Paying for This
Utilities are seeking $29 billion in rate increases for 2025. Why? Data center infrastructure buildout. Residential customers in Virginia, Texas, and Ohio are subsidizing hyperscaler data center expansion through higher electricity bills. The same companies charging you $20/month for AI subscriptions are also causing your electricity bill to rise. This is not disclosed as a cost of AI on your utility statement.
⚡ The Hidden Cost: Fossil Fuel Lock-In
Some states are expanding natural gas generation and delaying coal plant retirements specifically to power data centers — locking in emissions for 20-30 years. The irony: AI companies tout sustainability goals while their compute demand is the single largest driver of new fossil fuel generation in the United States. The climate math doesn't add up, and it's buried in state utility filings that nobody reads.