The rapid proliferation of artificial intelligence (AI) applications is increasing electricity demand through both large-scale model training and continuously running inference services. There is intense debate in the US about the scale and urgency of this increase: some argue that capacity is sufficient and the problem can be solved with efficiency improvements and investments; others claim that demand has exploded and the existing infrastructure cannot keep up. The truth lies in the combination of these two perspectives.

1. Training and Inference: Energy Profiles
Training: Training large language models involves operations that take days or weeks and require high-intensity GPU/TPU usage. These processes create short but intense peak energy demands.

Inference: Serving millions of users with trained models creates a continuous but lower-power load. As the number of users increases, total energy consumption grows rapidly.

Result: Training creates one-off but intense peaks; inference, on the other hand, creates continuous and, as it scales, a large overall energy demand.

Companies that could benefit: NVIDIA ($NVDA), AMD ($AMD), Alphabet/Google ($GOOG), Microsoft ($MSFT), Amazon ($AMZN)

2. Data Center Efficiency and Technological Improvements
Hardware optimization: TPU, ASIC, and next-generation GPUs increase energy efficiency. Software techniques such as model compression, quantization, and distillation reduce energy cost per extraction.

Design and cooling: Modern data centers are becoming more efficient with PUE improvements, liquid cooling, and co-location strategies.

Renewable energy: Major providers are turning to renewables; however, intermittent production requires storage and demand management.

Companies that could benefit: Alphabet/Google ($GOOGL), Meta ($META), Amazon ($AMZN), Microsoft ($MSFT), Broadcom ($AVGO), Supermicro ($SMCI)

3. US Grid Impacts: National vs. Regional
National level: Some analyses suggest that the additional demand from AI is manageable within total US electricity consumption.

Regional risks: The clustering of data centers in regions such as Virginia, Texas, and Oregon is putting pressure on local transmission and distribution infrastructure. Grid connection queues have extended to 4–5 years. Prices are soaring in the Virginia and PJM markets, increasing reliability risks.

Company response: Giants like Microsoft, Google, Amazon, Meta, xAI, and OpenAI are building their own power plants, making nuclear reactor deals, and turning to off-grid gas power plants.

Companies that could benefit: Dominion Energy ($D), NextEra Energy ($NEE), Duke Energy ($DUK), Exelon ($EXC), Vistra ($VST)

4. Demand Explosion in Numbers
In 2023, US data centers consumed ~176 TWh of electricity (4.4%).
In 2025, US data centers consumed ~224 TWh of electricity (5.2%).
In 2027, US data centers will consume ~371 TWh of electricity (8.0%).
DOE/LBNL projection: Could reach 649 TWh in 2030 (11.8%).
Goldman Sachs: Data center power demand will increase from 31 GW in 2025 to 66 GW in 2027.
IEA: 130% increase by 2030 (~240 TWh additional demand).

Companies that could benefit: NVIDIA $NVDA, AMD $AMD, Supermicro $SMCI, Micron Technology $MU, Arista Networks $ANET
A large portion of this increase comes from GPU-based "accelerated" servers. In some years, data centers account for half of the total increase in electricity demand in the US.

5. Scenarios
Scenario A — Rapid, unplanned growth: Large model trainings, poor location selection, and delayed renewable integration create immediate capacity pressure on the grid.

Scenario B — Planned, efficiency-focused growth: Demand can be managed through model optimization, training scheduling, renewable integration, and storage investments.

Realistic picture: In the short-to-medium term (2026–2028), energy access has become one of the most significant factors limiting AI growth. In the long term (post-2030), investments and efficiency can provide a balance.

Companies that could benefit: Tesla Energy ($TSLA), NextEra Energy ($NEE), Brookfield Renewable ($BEP), Fluence Energy ($FLNC), Bloom Energy ($BE)

6. Policy and Business Recommendations
Policymakers: Consistently implement regional grid planning with data center projects. Incentivize renewable energy and storage investments.

Companies: Reduce peak demand by scheduling training programs; prioritize efficiency investments. Select data center locations based on renewable resources and cold climates.

Local governments: Demand transparent impact assessments and community participation in new projects; plan infrastructure investments in advance.

Companies that could benefit: General Electric ($GE), Siemens Energy ($SIEGY), Schneider Electric ($SBGSF), Eaton ($ETN)

7. Environmental and Economic Dimensions
The increase in energy demand is critical in terms of carbon emissions, energy costs, and local environmental impacts. The use of renewable energy and carbon pricing mechanisms can reduce this cost. Economically, data centers bring jobs and investment, but energy price pressures have different effects on communities.

Companies that could benefit: NextEra Energy ($NEE), Ørsted ($ORSTED), Enphase Energy ($ENPH), First Solar ($FSLR)

8. Grid Obsolescence and Infrastructure Issues
The US power grid is struggling to meet the increased demand driven by artificial intelligence, not only due to capacity limitations but also due to obsolescence.

Transmission lines: Lines in many regions are 40–50 years old. Modernization necessary for high-capacity data centers is progressing slowly.

Transformer and distribution infrastructure: Local transformer capacity is limited; in densely populated areas (Virginia, Texas), old equipment frequently creates bottlenecks.

Gas turbines and generation fleet: A significant portion of the US's current thermal capacity consists of aging power plants. Renewal and new turbine investments are delayed due to supply chain congestion.

Planning gap: Grid connection queues extending to 4–5 years are a direct result of obsolescence and lack of modernization.

Companies that could benefit: Quanta Services ($PWR), AEP ($AEP), National Grid ($NGG)

Conclusion
AI growth highlights not only the need for new capacity but also the rapid renewal of aging infrastructure. Without modernization, energy access will remain one of the most critical factors limiting AI growth in the short term. In the US, AI growth is also an energy race: national capacity is still large and the risk of blackouts is low; however, regional bottlenecks, speed, and reliability issues are urgent. In the short term, energy access may limit AI growth; in the long term, technology, investments, and efficiency can provide a balance. The most appropriate approach is a coordinated policy and industry strategy that includes scenario-based planning, efficiency investments, and renewable energy + storage integration.