'Shifting In Our Favor': Data Center REITs Could Finally Get Their Piece Of Big Tech’s AI Pie

Public data center REITs have thus bar been left out of the artificial intelligence building boom. But their executives say that is about to change. 

Big Tech’s AI arms race has helped drive record leasing numbers and a wave of new development across the data center industry, with companies like Microsoft Google and Amazon snapping up inventory hundreds of megawatts at a time. Yet public data center REITs Equinix and Digital Realty, despite being the world’s largest third-party data center providers, haven't been among the major players building these increasingly large single-tenant facilities in U.S. markets. 

But leaders of these data center REITs insist they aren't missing the AI boat.

In third-quarter earnings calls this week, executives at Equinix, Digital Realty and American Tower have all pointed to a fundamental shift underway in the AI ecosystem they say will drive a larger share of AI demand to their existing portfolios.

And for the first time, Equinix and Digital Realty signaled interest in more aggressively pursuing the kind of massive AI-supporting campus developments that are driving the industry’s growth and expansion across the U.S.

“This doesn't mean that Digital Realty will be chasing large AI deployments far and wide, as we will continue to assess the longer-term opportunity set of remotely located single-tenant … data centers,” Digital Realty CEO Andy Power said on the company's earnings call Thursday. “We're being thoughtful in how we approach these opportunities.”

Generative AI has rapidly reshaped the data center landscape in just 11 months since the release of ChatGPT. The massive amount of data center capacity needed to support power-intensive AI computing, particularly AI training, has hyperscalers scrambling to pre-lease new facilities months or years ahead of delivery. AI has led to more and significantly larger data centers that are increasingly located outside traditional markets as developers chase available power. 

Driving the bulk of this new hyperscale development has been a growing number of private firms backed by deep-pocketed institutional investors: developers like Blackstone-owned QTS, DigitalBridge Group’s Vantage, and Compass, backed by Brookfield and a Canadian pension fund. 

When it comes to this kind of single-tenant development, major public REITs like Digital Realty, Equinix and even American Tower-owned CoreSite have been pushed to the margins.

Why have these firms missed out on the AI infrastructure gold rush, even as they posted strong revenue growth and reported rising demand from AI within their existing portfolios?

In part, these REITs have been handcuffed by public markets. While firms backed by private capital can take on significant debt loads or make large capital expenditures to build campuses that won’t produce a penny of revenue for years, public firms would see their stock price suffer. 

Perhaps more importantly, leaders of major data center REITs say they have shied away from pursuing AI-driven hyperscale projects, particularly outside of major markets, because they don’t fit with the portfolio development strategies they have employed for years.

While a colocation REIT like Equinix operates a huge range of different data centers, its core real estate strategy is focused on multitenant data centers in major markets. Their portfolios prioritize interconnection and low latency through proximity to population centers, hyperscale facilities and key fiber infrastructure.

This is a fundamentally different development calculus than what is required for hyperscale AI projects that follow power availability into new markets and where latency isn't always a major consideration. For companies like CoreSite, the safe bet is to stay away. 

“We continue to believe the majority of today's generative AI workloads will provide hyperscale opportunities that don't meet our investment criteria or fit within the CoreSite ecosystem,” said Tom Bartlett, CEO of CoreSite parent company American Tower.

But Bartlett also pointed to a significant shift in the AI marketplace that he — along with his counterparts at Equinix and Digital Realty — says is primed to drive a growing share of AI demand toward data center REITs. The change: a growing share of demand coming from AI inference, not AI training.

“We anticipate a dramatic acceleration in inference workloads,” Equinix CEO Charles Meyers told analysts Wednesday. 

AI adoption is still in its infancy. Data center infrastructure is primarily needed to support training of large language models and other applications that need massive clusters of high-performance compute, but for which latency isn't a major consideration. Few AI workloads today need such low latency that they have to be in the prime locations that companies like Equinix have set up their portfolios to serve.

But as AI is adopted and incorporated into businesses and organizations of all stripes, inference is going to account for a larger share of workloads. This means latency and interconnection are going to be increasingly important — exactly the kind of application the public REIT portfolios were designed to support. 

“This demand case is going to grow and change, and it feels like it’s shifting in our favor,” Digital Realty's Power said on the earnings call. “This is still early innings of what's going to be a vast build-out of required infrastructure. We're in a world right now where it's about large language model training, but the dawn of inference points to incremental location latency-sensitive needs.” 

Yet even as demand trends shift toward their portfolios, both Equinix and Digital Realty say they plan to at least explore developing the kind of single-tenant hyperscale facility for AI workloads that they have thus far shied away from.

Equinix, in particular, is aggressively targeting hyperscale AI deployments.

The company announced Wednesday that it will be building and operating single-tenant data centers for hyperscale AI in U.S. markets through its xScale JV development program. Until this year, Equinix had deployed the xScale product exclusively in international markets due to concerns over competing in an already-oversupplied market. But added demand from AI has changed the equation. 

“That dynamic is changing. We intend to meaningfully augment our xScale portfolio, including in North America, to pursue strategic large-scale AI training deployments with the top hyperscalers and other key AI ecosystem players, including the potential to serve highly targeted enterprise demand,” Meyers said on the earnings call. “We expect some builds will be tightly coupled with our retail campuses … while other builds will be larger-scale campuses in locations with access to significant power capacity.”

The shift in strategy is also intended to shore up relationships with the Big Tech cloud providers who are the firm’s most important tenants, according to Meyers, ensuring that Equinix can offer solutions across a range of deployment sizes and geographies. The increased scale also improves the company’s purchasing power with suppliers, ensuring its development schedules are prioritized by manufacturers as lingering supply chain woes plague the industry and create costly delays. 

“We said, ‘Look, we need to continue to have really well-developed and constructive relationships with the major players in the digital ecosystem, and obviously, the hyperscalers are at the top of that list,’” Meyers said. “It's also important that we continue to maintain our scale and relevance in the supply chain. I think we are very well positioned there.”