With lofty valuations in AI-related stocks, it’s understandable that investors are on edge about anything that might go wrong. Recently, concerns have centered on the ability to bring new AI data centers online at the volume AI providers require. While funding for construction is flowing, a lack of electricity and permitting — two very low-tech problems — might be the Achilles’ heel of this otherwise high-tech industry.Demand for AI is growing unabated. On Tuesday, AMD (NASDAQ:AMD) CEO Lisa Su told investors that the company’s AI data center revenue should grow by about 80% per year over the next three to five years. As a result, AMD’s total revenue is forecast to increase by roughly 35% annually over the same period, which exceeded analysts’ estimates. The company’s stock price has been soaring (chart).But AI data center provider CoreWeave (NASDAQ:CRWV) threw cold water on any excitement after warning that its Q4 results would miss expectations because a developer failed to deliver data centers on time. Infrastructure that was expected to come online in Q4 will instead come online in Q1-2026 and Q2-2026.The delay didn’t cost the company any customers, and demand for space in its data centers is still insatiable, CEO Mike Intrator told CNBC. But the company did reduce its 2025 revenue forecast range to $5.05 billion to $5.15 billion, below analysts’ forecast of $5.29 billion.Microsoft (NASDAQ:MSFT) CEO Satya Nadella also recently raised concerns about construction delays. “The biggest issue we are now having is … the ability to get the [data center] builds done fast enough, close to power. So if you can’t do that, you may actually have a bunch of chips sitting in inventory that I can’t plug in. In fact, that’s my problem today. It’s not a supply issue of chips. It’s actually the fact that I don’t have warm shells to plug into.” A “warm shell” is a new data center ready for occupancy.We wonder if this is the beginning of many problems the industry will face trying to build and deliver data centers on time. Here’s a look at some of the numbers surrounding AI data center demand, supply, and electricity:1. A building spreeEven before AI hit the scene, the number of data centers in the US had been on the rise as more people and businesses increasingly relied on cloud computing to process, analyze, and store their ever-increasing data. The advent of AI pushed that trend into hyperdrive.Monthly spending on data center construction starts rose to a record $4.2 billion in August, based on a 12-month moving average. That’s up 100% from the August 2024 level and 400% from August 2023, ConstructConnect reports. Spending ytd through August at $40 billion had already surpassed the full-year 2024 record (chart). Building has also gotten more expensive, with the average cost per square foot rising to $977 in August, up from $665 a year prior.The US leads the world in data centers, with 4,189 to date — far exceeding the number in any other country, according to the Data Center Map. The UK has 511 data centers, followed by Germany (487), China (381), France (321), Canada (294), India (276), Australia (275), Japan (247), Italy (209), and others had even fewer—those tracked range from hyperscale data centers to edge data centers.2. A power problemUS data centers consumed 183 terawatt-hours (TWh) of electricity in 2024, according to IEA estimates cited in a Pew Research Center October 24 report. That works out to more than 4% of the country’s total electricity consumption last year. By 2030, consumption is projected to grow by 133% to 426 TWh.A third of data centers are in Virginia, Texas, and California, Pew reports. Because data centers are often geographically concentrated, they can tax electric grids. For example, about 26% of the total electric supply in Virginia is consumed by data centers.There are growing concerns that data center demands are driving up electricity prices today and will continue to do so in the future. In recent years, electricity rates have increased as utilities have replaced aging equipment to protect against extreme weather events and cyberattacks.The typical US household was billed $142 a month for electricity last year, up 25% from $114 a month in 2014 (chart). Consumers worried about their electric bills are unlikely to support the construction of new data centers in their communities.Access to electricity is also causing headaches for those building data centers. Utility connection delays of up to five years are the most significant obstacle for data center growth, reports Bain & Company.3. Looking to the starsCurrently, data centers receive their energy from traditional electricity sources. Natural gas-powered utilities supplied more than 40% of the electricity used by data centers in 2024. Renewables, like wind and solar power, supplied 24% of the electricity, nuclear about 20%, and coal around 15%, Pew reported.Going forward, data center companies are looking at traditional sources as well as new ones to augment power supplies. Google Research is exploring space via its Project Suncatcher. The moonshot project is described as placing a constellation of solar-powered satellites carrying Google TPUs into low-earth orbit and connecting them via free-space optical links to create space-based AI infrastructure (chart).To turn this idea into a reality, Google (NASDAQ:GOOGL) will need to overcome several challenges, including establishing high-bandwidth communication between the satellites, managing orbital dynamics, and protecting equipment from radiation damage. Placed in the right orbit, a solar panel can be eight times more productive in space than it is on Earth, producing power nearly continuously and reducing the need for batteries.Google believes that by the mid-2030s, the cost of launching satellites should fall enough to make the cost of launching and operating a space-based data center roughly comparable to the energy costs of a data center on Earth. It plans to launch two prototype satellites by early 2027 to test how its models and TPU hardware operate in space and validate the use of optical intersatellite links for distributed machine learning tasks.Original Post