America’s artificial intelligence buildout is moving into a more expensive phase, where the limiting factor is no longer simply access to chips, land or capital but the ability to produce enough electricity at the right locations. Moody’s Ratings estimates that the US will need about $110 billion in new power-generation investment through 2030 to support roughly 45 gigawatts of additional capacity associated with the expanding data center economy. The estimate puts a financial scale on a problem that has increasingly surfaced in utility planning, corporate power procurement and the economics of large AI campuses. Data centers are emerging as a major source of new US electricity demand after years in which power consumption grew slowly enough that utilities had little incentive to accelerate generation construction.
The Moody’s projection draws on the International Energy Agency’s expectation that US data centers could consume about 426 terawatt-hours of electricity in 2030. That would roughly double the sector’s share of national electricity consumption to about 10% from 2025 levels, turning hyperscale computing into a much more consequential component of the US power market. One gigawatt represents roughly the output of a traditional nuclear reactor, making the projected 45 GW requirement a substantial addition to the country’s generation fleet. The number also illustrates why individual AI projects increasingly resemble utility-scale infrastructure investments rather than conventional commercial real estate developments. A data center can be built comparatively quickly once a site, equipment and financing are secured, but generation, transmission and interconnection projects often operate on timelines measured in years. That difference creates a structural bottleneck in which the speed of AI investment can outpace the physical infrastructure required to supply it.
Natural Gas Will Carry Most of the New Generation
Natural gas is expected to provide more than 30 GW of the additional generation identified by Moody’s, making it the dominant source of new supply through the end of the decade. The associated demand could require roughly 4 billion cubic feet a day of incremental natural gas supply, creating another infrastructure requirement upstream of the power plants themselves. Gas generation remains attractive to data center developers because operators can dispatch it when demand rises rather than depending on the availability of sunlight or wind. That controllability matters for AI workloads, which can create large and sustained electricity requirements across training, inference and cloud-computing operations. The choice also reflects a broader reality in power markets: adding generation capacity is not enough if the resource cannot reliably produce electricity during the hours when large computing loads require it.
Solar generation and energy storage will account for much of the remaining capacity in the Moody’s projection, while nuclear restarts will contribute less than 5% of the total. That mix underscores the limits of relying on a single technology to solve the electricity requirements created by AI infrastructure. Solar can add large quantities of generation, while batteries can shift some of that output toward periods of higher demand, but neither automatically provides the same operating characteristics as dispatchable generation. Nuclear assets, meanwhile, can provide substantial quantities of continuous power but face constraints involving plant economics, licensing, maintenance and restart requirements. The resulting portfolio will therefore depend heavily on how utilities and regulators value reliability alongside the nominal cost of generation. AI developers have increasingly treated power availability as a strategic input, and the generation mix emerging around their campuses could shape both operating costs and project timelines for years.
Data Centers Could Absorb Billions of the Cost
The $110 billion generation requirement does not translate directly into a single national surcharge on electricity customers. Moody’s estimates that the new power plants alone could add between $25 billion and $30 billion a year to US electricity costs, but the eventual allocation will depend on local rate-making processes and rules governing infrastructure costs. Ryan Wobbrock, senior vice president of ratings at Moody’s Ratings’ global infrastructure finance group, said in an email that how those costs reach consumers will depend on the rate-making process and other rules for allocating costs in each area. Some jurisdictions may require large customers to absorb a greater share of the infrastructure they trigger, while others could distribute portions of those investments across broader customer classes. The resulting policy decisions could determine whether AI expansion becomes primarily a corporate infrastructure expense or a broader electricity-system obligation.
Data center operators are already moving toward arrangements that place some generation directly behind the meter, allowing large computing campuses to secure dedicated power without relying entirely on the traditional grid. Moody’s estimates that these projects could represent about 30% of the total generation buildout through 2030 and that data centers could directly pay as much as $15 billion of the associated costs. Behind-the-meter generation can give developers greater control over capacity and reduce exposure to grid constraints, but it also changes the relationship between computing companies and the electricity system. A campus that finances its own generation can effectively behave more like an industrial power customer with dedicated infrastructure than a conventional commercial load. That model could become increasingly attractive as grid queues lengthen and utilities struggle to synchronize transmission upgrades with the arrival of new AI facilities.
Regulators Are Becoming Part of the AI Infrastructure Timeline
Affordability has added another layer of uncertainty because regulators now have to evaluate how rapidly utilities should expand infrastructure for customers with unusually large and concentrated loads. Moody’s said that “federal and state politicians and regulators are requiring more time to review and structure rates, creating a new source of potential delays for new data centre developments”. Those reviews can affect the economics of projects before developers commit billions of dollars to buildings, servers and power systems. The central question is no longer only whether a utility can serve an AI campus but whether the proposed rate structure appropriately assigns the cost and risk of doing so. If regulators determine that other customers could bear too much of the burden, they may impose additional conditions, deposits, tariffs or contractual requirements on data center operators.
The political stakes are rising because the benefits of AI infrastructure and the costs of expanding the power system do not necessarily fall on the same groups. Whether households and smaller businesses face higher electricity costs will depend on how utilities and regulators allocate infrastructure investments across customer classes. The broader buildout has also drawn opposition over concerns involving electricity costs, pollution, water use and impacts on land and property. The White House has framed the global AI race as a national imperative, strengthening the political case for faster infrastructure development and greater domestic computing capacity. But national urgency does not eliminate the local economics of electricity systems, where regulators must still decide who pays for new generation, transmission and reliability upgrades. That tension is likely to become one of the defining infrastructure questions of the US AI expansion.
The AI Power Bill Will Reshape Data Center Economics
The $110 billion figure matters because it shifts the discussion around AI infrastructure from capital expenditure on servers and buildings toward the full cost of the electricity ecosystem required to operate them. A data center project can look financially compelling when evaluated primarily through land, construction and computing capacity, yet its economics can change substantially once generation, transmission upgrades, interconnection costs and electricity-rate exposure enter the model. The Moody’s analysis suggests that power procurement will increasingly become a core component of the competitive strategy for hyperscalers, cloud providers and specialized AI infrastructure companies. Projects with access to firm generation could command a strategic advantage over facilities that depend on uncertain grid expansion schedules. Developers may therefore increasingly evaluate power sites based not only on available megawatts but also on generation technology, fuel access, transmission topology, regulatory structure and the ability to control long-term electricity costs.
The emerging power constraint could also change where America builds its next generation of AI capacity. Regions with abundant generation but limited transmission may compete with areas closer to major load centers, while states with faster permitting and clearer large-load tariffs could attract projects that might otherwise face years of uncertainty. Direct power arrangements, long-term generation contracts and dedicated generation could become more common as companies seek to reduce dependence on grid expansion alone. At the same time, utilities may need to redesign planning assumptions around customers whose demand can arrive faster and at a much larger scale than traditional industrial loads. The resulting investment cycle will extend well beyond the data center itself, reaching natural gas pipelines, renewable projects, batteries, transmission networks and generation equipment manufacturers.
America’s AI Race Now Depends on Building Power Faster
The central challenge for the US AI boom is becoming increasingly physical: the country must build enough reliable electricity infrastructure before computing demand reaches the limits of the existing grid. Moody’s $110 billion estimate provides a financial measure of that challenge, while the projected 45 GW of new generation illustrates the scale of capacity required through 2030. Natural gas will likely provide the largest share, supported by solar, storage and a smaller contribution from nuclear restarts. Data center companies could directly finance a meaningful portion of the investment through behind-the-meter projects, but the broader electricity system will still determine how much additional cost reaches other customers. Regulators will consequently play a larger role in AI infrastructure economics as they weigh speed, reliability, affordability and cost allocation.
