‘There Will Be Grave-Dancing’: Why Not All Data Center Operators Stand To Benefit From AI Boom

The ongoing explosion in artificial intelligence will drive more demand to colocation data centers, but some operators may not be in a position to capitalize on the opportunity. 

The world’s largest tech companies are pouring money into the development of AI and the infrastructure needed to support it, an arms race with significant implications for the colocation data center industry that has evolved around the needs of hyperscalers like Microsoft, Google, Amazon Web Services and Meta.

There is a broad consensus among digital infrastructure providers that the data center sector is at an inflection point amid the earliest ripples of an imminent wave of AI-driven demand that will fundamentally change the data center ecosystem. 

Experts say colocation data center firms will remain an integral part of the emerging AI infrastructure landscape, even as these new technologies change their relationships with the hyperscale cloud giants who are their largest tenants. But some colocation providers may be positioned better than others to take advantage of this expected paradigm. 

Significant capital expenditure is needed to design data centers that can support AI computing, and not every company will be able to make those large upfront investments today for an expected demand surge that is only just starting to materialize. But experts say this may well determine the winners and losers as AI rewrites the data center map.

“It comes down to those who have capital,” Michael Ortiz, CEO of data center developer Layer 9 Datacenters, said at Bisnow’s DICE Southwest event last month. “There will be grave-dancing, and there will be opportunities for M&A. There will be opportunities to seize the moment and take advantage of those who don't have access to capital.”

Since the emergence of ChatGPT late last year, the largest cloud and social media providers have all made hard pivots toward artificial intelligence, pouring resources into incorporating generative AI technology into products and services across their different business lines. This has meant cranking up spending on the infrastructure needed to support these technologies. In short, adding more data centers.

At the same time, the AI craze has fueled the growth of an emerging enterprise sector: companies developing an array of AI products, tools and services with their own proprietary AI models or using existing cloud-based models offered through services like Google Cloud and Microsoft Azure. These companies are also hungry for computing power. 

But the equipment and infrastructure required for most AI applications differs significantly from the servers most data center providers have been hosting for years.

In general, AI requires faster, more powerful processors and networking equipment that use much more electricity and create more heat than the equipment for which most data centers are designed. Because of this, data centers to support AI have to be designed differently, incorporating fundamentally different electrical infrastructure and liquid-based cooling systems that most data centers aren’t designed for. These systems also weigh significantly more than traditional IT hardware, meaning the data center building has to be structurally stronger. 

“This completely changes how data centers are built and operated,” Ortiz said. “Try building a multistory data center supporting that kind of gear.”

Companies like Meta and Microsoft plan to build out some of these new data centers themselves instead of leasing from third-party colocation providers. While the hyperscalers have always developed some of their own facilities, some experts predict they will do so more frequently in the years ahead. 

But even if hyperscalers build their own infrastructure to support their AI models and tools, Cloud providers like AWS, Microsoft and Google could be at a disadvantage to colocation owners when it comes to providing massive amounts of computing power to other companies looking to develop their own AI models.

That kind of high-performance computing as a cloud-based service is far less profitable for hyperscalers than traditional cloud products, meaning it isn't something the tech giants are likely to pursue aggressively, said George Slessman, founder and CEO of AI-focused data center provider CR8DL.ai.

Indeed, from the perspective of the growing number of AI startups building and training their own AI models, colocation is a cheaper option than using the cloud.  

“Silicon Valley VCs are now pushing their guys to not use the cloud now," Slessman said. “It’s a great sign for colo operators.”

Even as the colocation industry’s leasing numbers taper off after three years of record growth driven by demand from hyperscalers, many within the colocation space are gearing up for what they see as the next data center boom fueled by AI. 

This week, Macquarie-owned data center firm Aligned — a wholesale colocation provider primarily focused on leasing to hyperscalers — announced an investment and strategic partnership with QScale, a Canadian firm focused exclusively on providing high-performance computing for AI. The agreement effectively adds AI capabilities to Aligned’s portfolio while funding QScale’s expansion.  

“While the hyperscalers have throttled back and have been reducing their demand, the forward-thinking people are looking at the increase in AI need,” Kurt Lindorfer, founding principal at data center specialist Paradigm Structural Engineers, said at DICE Southwest. “Those who are ahead of the curve are thinking, 'We're going to need to provide space and power and services for that.'”

There is evidence to suggest that AI-driven leasing is already on the upswing. Executives at colocation REITs Equinix and Digital Realty highlighted increases in demand from AI-focused tenants on calls with analysts this month, pointing to their existing infrastructure to support AI workloads for hyperscalers and enterprise tenants. 

“We've closed several key AI wins over the past few quarters and are seeing a growing pipeline of new opportunities,” Equinix CEO Charles Meyers said. “ChatGPT has created a media frenzy around AI, but the reality is we've seen AI-related opportunities in our pipeline for the last several years.”

Yet amid talk of the industry’s AI transformation, the reality is that demand for colocation space to support AI is just now beginning to slowly accelerate. Colocation operators across the industry report that AI accounts for just a tiny fraction of their overall capacity. 

But with a broad consensus that a wave of AI-driven demand is fast approaching, this creates a potentially existential question for companies developing data centers today: Do you foot the vastly higher development cost needed to redesign data centers for AI demand that might not materialize for years?  

“It requires another couple billion dollars per company to refactor their data centers. … Everything that matters needs to be engineered into that system — it's a very challenging transformation,” Slessman said. “I'm wholly convinced at this point in time the data center is built prior to this year and not going to be able to support what's coming.”

Slessman and others say that with data centers expected to be in use for decades, it is less a matter of whether data center providers want to design their facilities for AI than whether they have the financial backing to do so. And as development timelines lengthen to two years or more, it is the companies with existing inventory that will thrive when the expected demand wave hits.

While it may seem intuitive to design for AI now, not every company will have the capital available to do so, particularly those who are backed by private equity or fund development on a project-by-project basis.

With the rising cost of capital and tightening returns, investors are less willing to stomach higher development costs. And while those initial investments may drive higher revenue in the future, capital backers are often operating on a much shorter timeline. They need to generate returns quickly and may be unwilling to fork over millions more for returns that may not materialize until after they have exited the deal. 

“You're not going to be able to go back to your investors and say, 'Hey, give me another $300M so I can refactor this data center to support AI,'” Slessman said. “That's probably the biggest barrier to the legacy providers being able to keep up.”