In 2019, Digital Realty Trust applied for permission to build a data center in Silicon Valley. It still sits empty. Not because the company ran out of money. Not because demand disappeared. Because the local utility cannot provide enough power. Six years of waiting. No resolution in sight.
That building, vacant in the heart of the world's technology capital, is the most honest summary of where the artificial intelligence race actually stands today.
The Scale of What Is Happening
Global data center electricity consumption reached approximately 415 terawatt hours in 2024, according to Brookings, which works out to about 1.5 percent of all the electricity used on earth. That share has been growing at a compound annual rate of 12 percent since 2017, a pace more than four times faster than total global electricity consumption growth. The number is large. The trajectory is what matters. At 12 percent annual growth, electricity demand from data centers doubles roughly every six years. The infrastructure being built to supply it takes longer than that.
Morgan Stanley's research team projects that U.S. data center demand alone could reach 74 gigawatts by 2028, against a projected shortfall of 49 gigawatts in available power. That is a gap of 49 gigawatts — roughly the combined generating capacity of 49 large nuclear plants — that does not exist and cannot be conjured quickly. The firm forecasts annual global electricity demand will rise by more than one trillion kilowatt-hours per year through 2030, with data centers accounting for nearly a fifth of that growth. Annual data center power consumption is expected to grow by 126 gigawatts through 2028. Canada's entire power demand is 133 gigawatts. The math is not comfortable.
Large technology companies are expected to commit more than one trillion dollars in spending in the 2025 to 2026 period alone, much of it now directed not at chips or models but at securing electricity. The CEO of an on-site electricity generation company, speaking at a Morgan Stanley conference in December 2025, described the situation this way: "If AI can create a $10 trillion GDP improvement in the economy through productivity gains, no one has really contemplated how much additional power all those users are going to need. It's mind-boggling."
Where the Grid Is Breaking
The United States electrical grid is, in most respects, a mid-twentieth century infrastructure project being asked to perform twenty-first century work. Ben Hertz-Shargel of Wood Mackenzie told CNN that the country has "basically run out of headroom, largely speaking." That framing is accurate and worth sitting with: the world's largest economy, home to the world's largest technology companies, is running out of the electricity needed to power the products those companies are building.
The constraint is not hypothetical and it is not evenly distributed. In 2023, data centers in Virginia consumed 26 percent of all electricity in the state. Ireland, which hosts major facilities for Google, Microsoft, and Amazon, has implemented a de facto cap on new data center grid connections in the Dublin region. The Netherlands imposed a nine-month moratorium on new hyperscale data center permits to assess their impact on the national grid. Frankfurt cannot expand an existing data center because the local utility cannot provide sufficient power until 2033. Google cancelled plans for a data center near Berlin for the same reason. A data center in Santa Clara cannot start operating because the local utility lacks capacity.
These are not cautionary projections. They are things that have already happened, in major cities, in wealthy countries, affecting some of the most capitalized companies in history.
The Three Bets Companies Are Making
Faced with a grid that cannot keep up, the hyperscalers have organized their responses around three strategies, each carrying real tradeoffs.
The first is nuclear. Microsoft, Google, and Amazon have all signed offtake agreements for nuclear power, betting on a generation source that is dense, reliable, and carbon-free. The difficulty is time. Ropes and Gray's analysis of data center investment in 2026 notes that new nuclear build timelines and cost overrun risks "remain material." The plants that would power 2026's data centers at scale will not come online before the early 2030s, at the earliest. The Energy Department is exploring nuclear-powered AI data centers on federal land through new public-private partnerships, but the same constraint applies: the power is needed now, and nuclear is a decade away.
The second bet is natural gas. Morgan Stanley forecasts that natural gas will meet roughly one-fifth of the world's new power needs outside of China, and gas investments have been hitting record highs since 2024. The tradeoff is that natural gas air permits are, as Ropes and Gray describes them, "a major gating issue," and the emissions profile sits awkwardly alongside the stated climate commitments of the companies making the bet. Community opposition is real and growing.
The third bet is geographic. Companies are building where the power exists, and the map of that search is reshaping global investment flows in ways that have little to do with traditional considerations of talent, infrastructure, or tax policy. Microsoft committed $15.2 billion to develop data centers in the UAE, a decision explicitly tied to that region's electricity availability. Amazon committed $6 billion to Malaysia and $5 billion to Thailand. Microsoft has shifted investment away from Ireland and the United Kingdom toward the Nordic countries. NTT Global Data Centers announced plans in March 2026 to double its global capacity to 4 gigawatts. The common denominator across every one of these decisions is not geography. It is the local grid's ability to provide power at scale.
What Is Actually Being Decided
Stephen Byrd, Global Head of Thematic Research for Morgan Stanley, described the situation at the firm's Powering AI conference: "The upcoming infrastructure CapEx cycle will create islands of wealth, and literal power." That is an unusual formulation, and a precise one. The countries and regions that can provide electricity at scale are receiving investment, jobs, tax revenue, and the long-run economic growth that follows from hosting this infrastructure. The ones that cannot are not.
Maeva Cousin, chief trade and climate economist at Bloomberg Economics, put it plainly: "Countries that fail to meet rising electricity demand risk missing out on defining investments that will shape economic prospects for decades." This is not a warning about some future scenario. It describes what is happening now, in real time, in decisions being made by the executives of the largest companies in the world.
The AI race is usually narrated as a competition between models, between chips, between labs, between the industrial policies of rival superpowers. That story is not false. But underneath every model and every chip and every lab is a building full of servers, and underneath every server is a wire connected to a grid. In Frankfurt, in Dublin, in Santa Clara, and in Berlin, that wire is currently the binding constraint. The Digital Realty building has been waiting six years for electricity.
The AI race is waiting too.
Works Cited
Brookings Institution. "Global Energy Demands Within the AI Regulatory Landscape." April 10, 2026. brookings.edu
CNN Business. "There Are Fixes for AI's Toll on the Power Grid. Here's Why They're Not Happening." April 23, 2026. cnn.com
Morgan Stanley Research. "Energy Markets Race to Solve the AI Power Bottleneck." February 27, 2026. morganstanley.com
Ropes and Gray LLP. "Data Center Investment in 2026: AI Demand, Power Constraints, and Private Equity Trends." May 2026. ropesgray.com
Tech-Insider. "AI Data Centers: 1,000 TWh by 2026." June 2026. tech-insider.org
Yahoo Finance. "AI Growth Strains Electrical Grid as Tech Companies Expand Operations." 2026. finance.yahoo.com