Article
How power is becoming the new geography of growth
As AI data centers expand, access to reliable power may increasingly shape where infrastructure is built, how quickly companies can grow and which regions capture the next wave of investment.
This article is based on research published by UBS Global Research – ESG.
You can’t power a data center without electricity. In the US, roughly USD 511 billion may need to be spent on new power generation capacity by 2030 to meet a 3% compound annual growth rate in electricity demand.1 That figure excludes the transmission and distribution networks needed to deliver the power to where it is required. For a digital economy often described in terms of code, chips and data, the scale of that number is a reminder that the next constraint may be strikingly physical.
For decades, technological progress appeared to depend principally on computing capacity. Faster processors, larger server farms and more advanced software set the pace. Electricity was essential, but it was often treated as a utility input, available enough to remain in the background of strategic decision-making. That assumption is weakening. The growth of AI and high-performance computing is turning power availability into a determinant of where digital infrastructure can be built and how quickly it can come online.
The change is driven by a different kind of demand. AI data centres require continuous, reliable electricity. Some new AI campuses require power at a scale comparable to the electricity consumption of a mid-sized city. This is a different load profile from the more familiar rises and falls of residential or commercial consumption. It creates sustained pressure on power systems designed for an economy with far more modest growth in electricity demand.
That matters because digital infrastructure and energy infrastructure run on different clocks. A hyperscale (i.e. very large, and typically owned and operated by a tech giant like Google or Microsoft) or colocation (also very large, but multi-tenant) data centre can often be approved and built within 12 to 24 months.2 Expanding power capacity, including generation, transmission, and substations, can take five to ten years because of regulatory processes, interconnection delays, supply chain constraints, and community resistance. The result is a mismatch between the speed at which digital demand is being committed and the speed at which physical infrastructure can respond.
This mismatch is already visible in the plumbing of the grid. Interconnection queues have reached historically high levels, with requests of around 2.3TW, roughly twice installed US power capacity as of 2024.3 Median timelines from new service request to commercial operation stand at around six years, up from about four years in 2021, according to the Lawrence Berkeley National Laboratory. Lead times for critical equipment have stretched as well, with large power transformers now taking around 2-4 years in some cases.4 These are not obstacles that capital alone can solve immediately. They reflect physical manufacturing limits, specialized materials, transport constraints and skilled labor shortages.
The central issue is that power scarcity is often local. There is no single, uniform US power grid that can be assessed through national averages alone. Electricity systems are fragmented across regions, utilities, regulatory structures and transmission boundaries. Surplus generation in one region cannot automatically serve a data centre in another. For a hyperscaler seeking to build in a constrained market, excess power elsewhere may be of limited practical use if transmission connections and local substations cannot deliver it.
The challenge is not simply to build more of one kind of power asset. The system requires a portfolio, with each option carrying trade-offs. Natural gas remains central to reliable dispatchable power, but gas turbines face stretched order books and delivery schedules extending into 2029 and beyond. Solar can be deployed relatively quickly and continues to narrow the economic gap with gas in some cases, but it is variable and cannot serve round-the-clock loads without storage or other firming resources (firming refers to the stabilization of variable power supply). Batteries can help shift supply across the day and relieve pressure on constrained systems, yet they do not generate electricity and are not a standalone replacement for baseload power. Fuel cells appear more promising than expected in certain cases, especially where waste heat can be redirected to chiller applications, but the long-term outlook remains challenging if demand growth slows materially.
This helps explain growing interest in behind-the-meter generation (i.e. power produced on the customer’s side of the energy meter, such as on-site solar), demand response and battery storage. These measures can reduce dependency on congested grid connections, help manage peaks and provide interim relief while larger infrastructure works through long development cycles. Growing deployment of battery energy storage systems, particularly alongside data centres and other behind-the-meter applications, reflects increasing demand for flexible, on-site energy solutions that can support resilience and power availability.5 Yet flexibility has limits. It can redistribute demand and improve utilisation of existing capacity, but it cannot remove the need for substantial new electricity supply.
The next phase will be defined by a practical question: can energy systems adapt quickly enough to support the ambitions of the digital economy? The answer is unlikely to rest on a single technology. Gas, solar, storage, fuel cells, transmission buildout, demand management and, over a longer horizon, nuclear all feature in the discussion. But the more important shift may be strategic rather than technical. Power is no longer merely an operating cost to be managed. It carries growth implications and, in turn, location implications.
The USD 511 billion figure is eye-catching because of its size. Its deeper significance lies in what it reveals about the digital economy’s foundations. The next era of computing may be shaped as much by substations, transformers and interconnection queues as by chips and algorithms. The digital economy is subject to a physical constraint, and the map of future growth may be drawn by the places where power can actually be delivered.
