AI Drives Doubling of Data Center Energy Consumption, Energy and Hardware Leaders Benefit

Overview

This report, based on the latest data from the International Energy Agency (IEA), deeply explores the multi-dimensional impact of Artificial Intelligence (AI) on energy demand, climate impact, and innovation potential. The core conclusion is: AI-driven computing power demand will explode in the coming years, leading to a doubling of data center electricity consumption before 2030, which will reshape the global energy landscape and benefit power infrastructure and semiconductor equipment suppliers. Meanwhile, AI has enormous potential in improving energy efficiency in traditional industries, which may offset its own carbon emissions, but faces multiple challenges including physical bottlenecks, cost allocation disputes, and social resistance during implementation.

Core views

Data center energy demand will witness historic growth. According to IEA's baseline scenario, global data center electricity consumption will double from 485 TWh in 2025 to 950 TWh by 2030, accounting for nearly 10% of global electricity demand growth during the same period. The United States will account for 45% of new demand, other developed economies for 19%, and China for 6-7%. This growth is mainly driven by AI hardware, with AI-related servers expected to account for more than half of data center growth by 2030. However, this growth is strictly limited by physical bottlenecks, including tight supply of high-end IT hardware (especially memory), grid connection waiting times of 4-5 years, gas turbine delivery delays, and shortages of skilled electricians. Therefore, data center cumulative capital expenditure from 2026 to 2030 is expected to reach $3.9 trillion, with approximately 20% ($780 billion) to be used for energy-related expenditures, including grids, power generation, backup generators, and UPS systems. The energy structure will change significantly. To support AI loads, renewable energy will become the largest source of new electricity, expected to contribute 360 TWh by 2030, accounting for more than one-third of total power generation; natural gas follows closely, expected to reach 340 TWh, especially in the US market. Additionally, on-site battery storage and UPS capacity will grow from the current approximately 5 GW to 20-25 GW. Although AI technology itself shows astonishing improvements in energy efficiency (energy consumption per query decreases by an order of magnitude), due to the 'Jevons Paradox' effect, where efficiency improvements stimulate broader use and more complex workloads, total energy consumption increases instead of decreases. For example, simple text queries require only 0.3 Wh, while intelligent agent tasks with reasoning capabilities require up to 50 Wh. The main drivers of future demand will come from high-energy consumption scenarios such as video generation, multimodal processing, and embedded enterprise inference, which account for 75-85% of future inference demand. AI's impact on climate shows a 'double-edged sword' characteristic. On one hand, AI has enormous potential in optimizing energy systems, with AI applications in industry, transportation, and construction expected to save energy equivalent to 3% of global terminal energy consumption by 2035, even exceeding data centers' own emissions. On the other hand, currently the vast majority of claimed climate benefits come from traditional AI, not high-energy consumption generative AI. Additionally, because rebound effects are not incorporated into models and implementation obstacles exist, uncertainty remains about whether AI can achieve net-positive emission contributions. For example, the Google-American Airlines flight path reduction trial showed that while technically 62% of contrails could be reduced, the actual execution rate was only 11.6% due to operational obstacles. Risks and opportunities investors should pay attention to. Although AI construction is irreversible, opposition from local communities (NIMBY movements) has become a substantial risk, with $156 billion in projects already delayed or canceled in 2025 due to local opposition or bans. Meanwhile, pressure from rising electricity prices is mainly concentrated in local congested areas, and policymakers need to avoid passing costs on to ordinary consumers through reasonable cost-sharing mechanisms. In terms of investment opportunities, JP Morgan is optimistic about the capital goods (CapGoods) and IT hardware sectors, considering them direct beneficiaries of the AI energy transition, and has recommended multiple 'Overweight' companies including Legrand, Schneider Electric, Siemens, and ASML.

Analysis framework

The report adopts a combined top-down and bottom-up macro and micro analysis framework. First, it uses IEA's baseline scenario model as an anchor, modifying the growth path of data centers through physical constraints (such as hardware supply, grid connection cycles) to distinguish between supply-demand balance states under different scenarios. Second, it introduces the economic concept of 'Jevons Paradox' to explain why technological progress带来的 energy efficiency improvements failed to reduce total energy consumption, but instead increased total demand due to the complexity of application scenarios (from text to agents). Finally, it evaluates the actual implementation capability of AI in emission reduction by combining specific cases (such as Google's flight path optimization, comparisons of carbon commitments among major tech companies) and field research data, and distinguishes the essential differences in environmental impact between traditional AI and generative AI.

Methodology notes

• Industry/Industrial Analysis FrameworkSupply-demand framework

  AI data center energy demand is limited by physical bottlenecks (hardware, power grids, labor), not just technical demand

  The research report points out that although AI demand is strong, the growth ceiling before 2030 is determined by supply-side factors such as memory supply, grid access time, and labor shortages, which is a typical supply-demand framework analysis, emphasizing short-term supply rigidity.

• Macroeconomic framework

  Jevons Paradox

  The report uses Jevons Paradox to explain the phenomenon: when technological progress improves resource utilization efficiency (such as decreased energy consumption per AI query), lower costs反而 stimulate broader demand and more complex applications (such as agent tasks), leading to increased total resource consumption rather than reduction.

• Valuation MethodDCF Discounted Cash Flow

  Reverse DCF valuation method used to determine target price

  When valuing individual stocks such as Legrand and Prysmian, the report uses a reverse DCF model, which first sets a target EV/EBITA multiple, then back-calculates implied future cash flow expectations, to verify whether current stock prices are reasonable.

Asset mapping & comparison

Structured mapping from thesis to named assets (strengths, weaknesses, peers, risks).

• ASML (ASML.US)

  As the sole supplier of EUV lithography machines in the EU, directly benefiting from the explosion of AI chip manufacturing demand, especially with rising DRAM prices and the adoption of high-NA EUV equipment.

  Strengths

  Monopoly position, extremely high technical barriers, benefiting from increased lithography intensity at advanced process nodes (such as 3nm/2nm).

  Weaknesses

  Facing export control risks (especially targeting China), economic downturn may affect capital expenditure.

  Comparison

  Compared to other semiconductor equipment manufacturers, its irreplaceability in the EUV field is stronger.

  Risks

  Export restrictions due to geopolitics, valuation suppression from rising interest rates.

• Schneider Electric (SU FP)

  Data center business accounts for 26% of its sales, providing critical power management and automation solutions, directly benefiting from the data center construction boom.

  Strengths

  Attractive long-term growth portfolio, excellent organic growth and profitability performance, high ESG rating.

  Weaknesses

  Recent financial execution affected by tariffs and exchange rate pressures, management changes causing investor concerns.

  Comparison

  Leading position in data center electrical infrastructure, second only to Siemens Energy.

  Risks

  Management uncertainty, extended transition period for new CFO may affect confidence rebuilding.

• Legrand (LR FP)

  Data center business accounts for as high as 30%, one of the companies with the highest data center exposure in the report, benefiting from growing low-voltage distribution demand.

  Strengths

  Strong execution record, strong pricing power, valuation attractive relative to historical average, growth outlook better than past five years.

  Weaknesses

  Slow recovery in European residential market, slowing M&A pipeline may drag overall growth.

  Comparison

  Data center revenue percentage higher than Schneider Electric, regarded as the best stock in this segment.

  Risks

  Data center market growth below expectations, continued weakness in European real estate.

• Siemens Energy (ENR.DE)

  Benefiting from AI-driven power demand growth and grid investment with leading gas turbine, grid equipment, and offshore wind portfolio.

  Strengths

  Service contracts lock in long-term supply-demand imbalance, strong free cash flow, wind energy business nearing profitability.

  Weaknesses

  Project execution risks, high dependence on power demand growth, aggressive competitor pricing.

  Comparison

  Significant advantages in gas services and grid technology, regarded as a strong cash flow trading stock.

  Risks

  Project execution delays, valuation downgrade due to slowing power demand growth.

Key data

• Data center electricity consumption in 2030950 TWhDouble from 485 TWh in 2025, accounting for nearly 10% of global electricity growth
• Data center cumulative capital expenditure from 2026-2030$3.9 trillionOf which approximately 20% ($780 billion) is used for energy-related expenditures
• Percentage of new data center demand in the United States45%Far exceeding other developed economies (19%) and China (6-7%)
• Renewable energy generation in 2030360 TWhAccounting for more than one-third of total data center power generation
• Energy consumption of agent tasks~50 Wh/taskApproximately 160 times that of simple text queries (~0.3 Wh)
• AI potential energy savings3% of global terminal energy consumptionExpected comprehensive energy saving effects of AI in industry, transportation, and construction by 2035
• Amount of obstructed projects in 2025$156 billionLosses of data center projects due to local opposition, bans, and litigation

Impact & implications

For the energy industry, AI will become the largest electricity demand growth point after electric vehicles, forcing utility companies and power generators to accelerate deployment of renewable energy and flexible regulation resources (such as energy storage, gas turbines). For investors, this means that power infrastructure, cables, transformers, and semiconductor manufacturing equipment manufacturers will face long-term structural growth opportunities. Meanwhile, policymakers need to establish new electricity pricing mechanisms and approval processes to balance data center development with community interests, preventing drastic electricity price fluctuations. For technology companies, how to meet computing power demand while fulfilling carbon neutrality commitments, especially solving the high energy consumption problem of generative AI, will be key to future competition.

Risks

• Physical bottlenecks (memory, turbines, grid connections) may cause data center construction progress to fall short of expectations.
• Local community opposition and regulatory bans (NIMBY) may significantly increase project costs and timelines, or even lead to project cancellations.
• Electricity price increase pressure is concentrated in local congested areas, potentially triggering policy intervention or cost-passing disputes.
• The high energy consumption characteristics of generative AI may prevent it from achieving expected net-positive climate benefits, facing greenwashing accusations.
• Geopolitical tensions may lead to restricted exports of semiconductor equipment, affecting supply chain security.
• Macroeconomic recession may suppress capital expenditure plans of tech giants.

What to watch

• Actual implementation of data center capital expenditure, especially the proportion of energy-related spending.
• Changes in grid connection queue times and delivery cycles of key hardware (such as transformers, gas turbines).
• Policy trends regarding data center electricity price allocation and cost recovery in various countries.
• Execution of 24/7 carbon-free energy commitments by large tech companies (Big Tech).
• Evolution of AI workload structure, especially adoption speed of agentic AI and video generation.
• Changes in the number of local legislation and bans targeting data center construction.