Ecology & Planetary Boundaries
Abundant energy and rapid data-centre expansion, offset by volatile climate policy and grid constraints.
Where does the USA stand, and where does it break?
US energy-related CO₂ emissions stood at approximately 4.8 billion tonnes in 2024 — well below their peak in the 2000s, though declining only slowly (U.S. EIA). The power sector has shifted markedly away from coal towards natural gas and renewables: natural gas accounted for the largest share of electricity generation in 2024 at around 43%, followed by renewables (approximately 23%, including hydropower), nuclear (approximately 19%) and coal (approximately 15%) (U.S. EIA). The decisive new factor is the power appetite of AI data centres: according to EIA and industry estimates, their share of US electricity consumption could rise to around one tenth by the end of the 2020s — driving a meaningful increase in power demand for the first time after years of stagnation. The United States holds a locational advantage: abundant natural gas, falling costs for solar and wind power, and industrial electricity prices below European levels. At the same time, climate policy is fragmented and volatile. State-level programmes such as California's cap-and-trade scheme sit alongside states with no carbon pricing at all, and federal policy shifts with each change in majority.
The economy ahead is compute-intensive: every large AI model, every inference request consumes electricity and cooling water. The expansion of data centres is pushing US electricity demand higher again after two decades of stagnation. This confronts climate targets with a central question: whether the additional demand will be met by renewables and nuclear power, or backstopped by new gas-fired plants. Grid bottlenecks, slow permitting for transmission lines, and the pace of build-out risk keeping fossil capacity online longer than planned. For the United States, this presents a dual dilemma: the energy and compute advantage draws value creation into the country, yet without the expansion of clean generation, grids, and storage, it could push emissions upward again. Whether AI becomes a driver or a brake on decarbonisation depends on how quickly clean capacity and grids can keep pace with the data-centre boom.
Paths for the USA
The United States is accelerating the permitting and construction of solar, wind, nuclear, storage and transmission infrastructure so that the new demand for computing power can be met on a climate-neutral basis. This is the critical switch point: the data-centre boom is driving electricity demand upward for the first time in years. The Inflation Reduction Act has already triggered a surge in investment through tax credits. The risk lies in sluggish permitting processes and in grid expansion that can barely keep pace with the speed of data-centre deployment.
Trade-off: build-out speed versus planning and grid reality
Feasibility: The key lies in reforming permitting procedures
The United States is betting on nuclear energy — existing reactors, life extensions, and new Small Modular Reactors — as a baseload-capable, low-emission power source feeding data centres directly. Major technology companies are already signing offtake agreements with nuclear operators. The appeal is clear: AI data centres require stable power around the clock, which wind and solar alone struggle to deliver. The downside: new reactors are expensive, slow to build, and subject to demanding regulatory requirements; SMRs have barely been proven at commercial scale.
Trade-off: security of supply versus cost and construction pace
Feasibility: Realistic at scale only by the end of the decade
The US is reducing specific consumption through more efficient chips, improved cooling and flexible load management that shifts computing workloads to periods of high renewable generation. This matters because a portion of compute growth can be absorbed through efficiency gains without the need for new generation capacity. Data centres can even stabilise the grid by acting as flexible loads. The limiting factor is the rebound effect: more efficient AI becomes cheaper and is used more intensively, meaning that absolute demand continues to rise despite efficiency improvements.
Trade-off: efficiency versus absolute demand growth
Feasibility: A complementary measure only, not a standalone solution
The United States is using natural gas capacity in the near term to backstop compute demand, flanked by state-level carbon pricing and methane rules to contain emissions. This is pragmatic, given that gas is readily available and flexible. The cost: without a federal framework, carbon pricing remains a patchwork, and gas plants running longer than intended jeopardise climate targets. Political volatility makes reliable price signals difficult to establish.
Trade-off: security of supply versus climate targets
Feasibility: Without a federal carbon framework, the incentive effect remains weak
Trigger-based phases
Trigger: the data-centre build-out causes US electricity demand to rise meaningfully for the first time after two decades of stagnation.
Utilities and regulators face connection and grid bottlenecks. The first conflicts over electricity prices, water consumption, and local permitting emerge. The question of how the new demand will be met moves to the centre of energy policy.
Trigger: grid operators and corporations determine whether the additional demand will be met by renewables, nuclear power, or new gas-fired plants.
The outcome will define the emissions trajectory of the coming decade. States and federal policy drift apart. Permitting and grid expansion become the bottleneck that determines both the pace and the cleanliness of the build-out.
Trigger: The emissions curve shows whether growth and compute demand have been decoupled from emissions.
If clean capacity expands at the pace of data centres, AI becomes a driver of decarbonisation. If it falls behind, emissions will rise again. The choices made in these years will shape the US climate record through to 2040.
Real-world pilot programmes
Status: Partially in force (the largest clean-energy investment package in US history); many tax credits were reduced or terminated in 2025
The IRA supports solar, wind, storage, nuclear and hydrogen through long-term tax credits and has triggered a surge in investment in clean generation. It is the test case for whether market-based incentives can drive build-out quickly enough to meet compute demand on a climate-neutral basis.
Status: Under implementation (power purchase agreement for data centres; restart planned)
Microsoft has secured the entire output of a reactivated nuclear reactor for its data centres through a long-term power purchase agreement. This real-world project illustrates how technology companies are procuring low-emission baseload power directly for AI compute, and how nuclear energy is returning as a source of compute power.
Status: In force (state-level carbon pricing as a model)
California's emissions trading scheme prices carbon where federal policy provides no framework. As a regulatory laboratory, it demonstrates how a single state can manage emissions through market-based mechanisms. At the same time, it reveals how fragmented US climate policy remains in the absence of a federal framework.
What we are watching
This measures the extent to which the compute boom is placing strain on the electricity system. The threshold matters because demand growing faster than clean generation keeps fossil-fuel capacity on the grid for longer. If the balance tips, emissions will rise despite climate targets. For decarbonisation, the pace is what counts.
Emissions show whether growth and CO₂ remain decoupled. The threshold matters because a renewed rise signals that compute demand is being met by fossil fuels. A reversal upwards would mean the AI economy is adding to the climate burden rather than alleviating it. For 2040 targets, the direction of the trend is what counts.
This share measures how cleanly growing electricity demand is being met. The threshold matters because stagnation amid rising demand indicates that new capacity is being added predominantly through fossil fuels. If the share flatlines, the United States will miss the decoupling target. For investors, a rising share signals the success of the energy transition in the age of compute.