Securing power sources vital to fueling data center, AI infrastructure investing boom
The boom in generative artificial intelligence is fueling investments around data centers and infrastructure but also raising questions about how much power will be needed.
The investment amount needed to sustain the growth in AI will be enormous. Blackstone Chairman and CEO Stephen A. Schwarzman recently estimated that in the U.S. alone, over $1 trillion in capital expenditures will be needed to build and facilitate new data centers over the next five years.
Through the end of July, $33.7 billion was invested globally in data centers, according to an analysis from law firm Linklaters. Large deals are helping drive AI growth. In May, CoreWeave, a specialized provider of critical cloud infrastructure, or AI hyperscaler, secured $7.5 billion in debt financing led by Blackstone and Magnetar Capital along with Coatue Management, Carlyle Group, C$434 billion ($318 billion) pension fund manager Caisse de Dépôt et Placement du Québec, DigitalBridge Credit and others.
In June, Apollo Global Management and affiliates agreed to acquire a 49% equity interest for $11 billion in a joint venture related to U.S. chipmaker Intel’s Fab 34 high-volume manufacturing facility in Ireland.
The power challenge
The amount of power needed to fuel the AI boom is immense. The International Energy Agency estimated in a 2024 report that electricity consumption from data centers, AI and cryptocurrency could double by 2026. Data centers globally consumed approximately 460 terawatt-hours in 2022 and that could hit 1,000 TWh in 2026, with the report noting “demand is roughly equivalent to the electricity consumption of Japan.”
DigitalBridge, the $84 billion global digital infrastructure firm, has been investing in data centers since 2016, said Kevin Smithen, chief commercial and strategy officer. A few years ago, a hyperscale technology campus would be around 50 megawatts with an initial customer deploying 20 to 30 megawatts. Today, standard AI data center campuses are using at least 250 megawatts, he said.
“You’re going to need a number of these campuses across North America for each large AI customer. So, the power density that is required for these new facilities is quite extensive,” he said, adding that DigitalBridge has prioritized securing power to sustain sector growth, often through renewables.
He added, “Those that are coming into the space that have not secured that power today are going to have to either build it (the energy infrastructure) or wait.”
DigitalBridge sees a natural adjacency to providing power to data centers and is considering as a strategy partnerships to provide power to data centers, with an eye to renewables, Smithen said.
The No. 1 challenge with data centers is accessing power, said Stephen Dowd, CIO for private infrastructure strategies at CBRE Investment Management. He’s seen a similar power growth in recent years with data centers going from the range of 30 to 50 megawatts to now being built for 100 to 200 megawatts.
“Then a subsidiary challenge to that is, what kind of power are you accessing?” he said, adding, “there’s still a lot of pressure” around investors asking, “can you make that power clean power?”
Dowd said that is “a tough nut to crack,” giving the example that big solar farms may not be located near data centers, which tend to be closer to population centers.
A recent CBRE report on data centers found they currently account for 2.5% of U.S. electricity consumption, and in northern Virginia, where almost half of the country’s data center inventory is found, they account for almost 20% of electricity consumption. The low cost of energy on a relative basis and state tax incentives for data centers make the area attractive.
One of those northern Virginia facilities is Corscale’s Gainesville Crossing, a 130-acre hyperscale data center campus. Development began in 2022 and construction has already begun on the second building on the campus. When fully developed, the campus will be over 2.3 million square feet with 326 megawatts of available power.
Corscale is a data center development company. Nic Bustamante, the chief technology officer for Affinius Capital affiliate Corscale, said the facility was built so that it won’t become obsolete anytime soon.
“We make it modular,” he said, describing how the building’s shell is meant to be there for the long term but the power and cooling systems can be unbolted and removed as new technology is developed.
Carrington Brown, senior managing director for development at Affinius who oversees data center investments, thinks the projected forecasts for the AI marketplace are wrong. Affinius manages $8 billion in a data center portfolio.
“I think the growth we will see in the next four to five years will exceed, frankly, every report that I’ve read,” he said. “The only thing that will stop that is the supply of energy.”
Brown said there will be need for new forms of power generation, including nuclear as a possibility.
Bustamante, who has worked in the data center sector since the late 1990s, said what is happening now with AI is “mind boggling. The growth rate in the data center sector is insane,” he said.
How institutional investors target AI
Mohammed Al-Sowaidi, CIO for the Americas region at the Qatar Investment Authority, said at the Bloomberg Invest conference in New York in June that the QIA decided AI was a long-term theme they needed to spend time on, and several years ago they started “investing heavily in things like data centers” and investing around infrastructure that will fuel AI.
Al-Sowaidi also gave the example of investing in a medical service business, Radiology Partners, that is using AI to make radiology assessments.
“In addition to finding new hardcore AI, we’re looking into other applications that could be empowered by opportunities or businesses that can be transformed with an element of AI,” he said.
Overall, it’s still early days for institutional investors and the next several years will be an evolution, said Jamshid Ehsani, a partner and head of global principal structured finance at Apollo Global Management and chairman of the ACRA group of insurance companies. But with trillions of dollars in investments coming over the next decade, it’s an area allocators “can’t ignore,” he added.
“In my opinion, as people start to better understand the whole ecosystem, what AI does, and how we are evolving the technology, be it on the compute semiconductor or memory semiconductor, or the way we are bringing the whole ecosystem together, I think people’s interest is going to increase, and you will have an allocation to this,” Ehsani said.
Understanding regional differences is also key. “You need to look at it from the perspective of the U.S. and the perspective of Europe. You need to have two separate strategies,” he said, pointing to data privacy regulations.
Investors are increasingly seeing AI and the digital ecosystem “as its own asset class, and some of the delineations between private equity, real estate and infrastructure are breaking down, and people are looking for ways to play these themes,” said DigitalBridge’s Smithen.
‘Long-term play’
The AI boom has been compared to the internet boom of the late 1990s, and it remains unclear which companies will come out on top.
“I think one of the advantages of infrastructure is we don’t know what uses will succeed and where,” CBRE’s Dowd said.
Dowd, who has invested in infrastructure for over 30 years, said data centers didn’t register at the beginning of his career and didn’t become viewed as part of infrastructure until it become more established with use cases.
Amid all the hype around AI, Ehsani cautioned investors need to be wary of overbuilding.
“It’s extremely difficult to know five years from now where we are going to be in AI. There’s one thing that is certain, whoever resolves the issue of power … power capacity will be dominating this sector,” Ehsani said.
He added, “This is not just a short-term play, this is a long-term play, and you have to be measured. It’s not a sprint, it’s a marathon.”
DigitalBridge’s Smithen thinks over the next five to 10 years, the AI boom will be revolutionary.
“The analogy is that there are always fits and starts with any technology cycle. But we do think this one is revolutionary, but it may play out after this early, early adoption phase … it may not always be just up and to the right, just persistent growth. There could be ebbs and flows temporarily, and then a reacceleration,” he said.
Consumers will also play a big role, Affinius’ Brown argued.
“The supply and demand fundamentals that we’re seeing in the sector is unlike anything that I’ve ever seen, the demand that we’re seeing from the users is really insatiable,” Brown said.