AI's Energy Demand: The Flow Numbers That Matter

The demand for electricity from AI data centers is not a trend; it is a structural shift in global energy consumption. The scale of this increase is massive and accelerating, with projections showing a near-doubling of power needs within a decade. This surge is the direct result of AI's insatiable appetite for computing power, turning data centers into the new industrial giants of energy use.

Global data center electricity consumption is projected to more than double by 2035, reaching 8.6% of US demand. This represents a steep climb from today's 3.5% share. The fuel for this growth is clear: training sophisticated models like GPT-4 requires immense power, with one estimate showing GPT-4 training alone required around 30 megawatts. This demand is now outpacing other emerging energy sectors like electric vehicles.

The near-term forecast is even more aggressive. Goldman Sachs Research forecasts global power demand from data centers will increase 50% by 2027 and by as much as 165% by the end of the decade compared to 2023 levels. This creates a clear supply-demand tension, with occupancy rates for data center infrastructure projected to peak above 95% in late 2026 before moderating. The bottom line is a fundamental reconfiguration of power grids, driven by a handful of tech giants building massive, power-hungry facilities.

The Supply Bottleneck and Market Risk

The immediate constraint on meeting AI power demand is not a lack of capital, but a grid that cannot expand fast enough. Utilities face severe delays in permitting and building new transmission lines to carry power from generation sources to high-density data centers. This creates a physical bottleneck that will force a premium on power in key growth regions, regardless of how much money is deployed.

This supply-side friction sets up a clear financial risk: a potential oversupply of data center capacity by 2027. Occupancy rates for this infrastructure are projected to peak above 95% in late 2026 before moderating. As more facilities come online and AI-driven demand growth slows, the market could see a surplus of high-capacity space, pressuring rental rates and returns on new builds.

The concentration of this demand in specific regions exacerbates the strain. The Northern Virginia market, for example, is the top data center hub in the U.S., housing a massive cluster of facilities. This regional concentration means local power grids are being pushed to their limits, making them particularly vulnerable to the transmission delays that hinder broader supply growth.

The Catalysts and What to Watch

The resolution of the AI power supply-demand imbalance hinges on three key catalysts. First, the pace of new energy projects is critical. The industry's seven-year development cycle for data centers means supply must be built in advance. Yet, as noted, the bulk of the electricity to power them must come from fossil fuel-based power plants if new clean capacity doesn't materialize quickly enough. The need to monitor new nuclear and renewable energy projects is therefore paramount to see if supply can keep up with the forecasted 50% global increase by 2027.

Second, regulatory actions and utility rate changes pose a direct risk to data center economics. As facilities strain local grids, utilities may impose higher fees or require costly infrastructure upgrades. This could increase operating costs for operators, potentially slowing expansion or altering the profitability of new builds. The financial impact of these regulatory shifts will be a key variable in the market's trajectory.

Finally, efficiency improvements in AI hardware and software offer a potential mitigating factor. Innovations like DeepseekV3's "Mixture of Experts" architecture promise better training efficiency, which could lower capex and energy use per model. If these gains accelerate, they could soften the demand curve and help avoid the oversupply risk projected for 2027. The bottom line is that the market's path will be determined by the interplay between the speed of new power generation, the cost of grid access, and the rate of AI efficiency gains.