Hyperscalers are breaking capex records and committing hundreds of billions of dollars to the data center rollout – largely in service of training AI models and preparing for a wave of AI inference that they hope will provide a massive return on investment.
This wave of investment dominates news cycles. Breathless stories about GPU farms and the race to build gigawatt campuses in regions severely lacking in the infrastructure needed to support them. The sheer magnitude of money being moved around is both awe-inspiring and concerning, as the potential for a record-breaking return in investment collides with ever-present fears of an AI bubble wreaking havoc on the world’s economy.
Building the chips to power quantum computers
The current rate of forecasted data center capacity growth demands an additional $3 trillion in total investment through 2030, according to statistics from JLL, resulting in $1.2trn in real estate asset value creation. It is true that this level of investment is largely driven by AI workloads, which JLL says could represent half of all data center capacity by 2030. In this scenario, inference will overtake training workloads by 2027 and likely redistribute demand from the centralized clusters we’re seeing today to distributed regional demand centers.
The vast majority of this investment will come from hyperscalers, who are already putting their money where their mouth is. But for the data center market writ large, particularly when it comes to private equity, the gap between this narrative and actual capital allocation is a more nuanced story.
The case for cloud
Private investors and limited partners are arriving, excited about AI, only to find that cloud-focused ventures are a more long-term, stable, and attractive investment.
Many private equity firms believe that cloud is still the workhorse through 2026, as returns on AI investments – particularly training – are not as attractive as steady cloud investments. With power availability challenges coming to bear, forcing investment and development into secondary and emerging markets, many of which are now taking the lead and becoming primary markets in and of themselves, eager investors have more to consider when it comes to acquisition and development than ever before. But how do you manage investor expectations when the headline technology isn’t necessarily where the safest returns are?
Kevin Kujawski, partner, president, and COO at Menlo Digital, says his firm is putting “a little less” emphasis on AI.
Menlo Digital, the data center investment and development arm of Menlo Equities, has a pipeline of around 880MW. With $9.7 billion of assets under management (AUM), the firm counts between 50-60 percent of that AUM as being in the digital space. Five years ago, the digital portion of the business made up 10 percent or less of its AUM.
Kujawski says much of this meteoric growth is coming from low-latency and cloud demand in primary markets, such as Virginia and Silicon Valley. And while AI may not be the driver of investment here, it is having a significant impact on Menlo’s investments. Kujawski explains that the presence of AI has caused a significant knock-on effect, driving up lease rates and sucking up power availability, which, if you can secure power for your site, gives you a “great opportunity.”
“We see this as more of an infrastructure type investment that, regardless of what AI does and the pace of rollout, will be essential facilities to facilitate the movement of data,” he adds, “So, even though some of the top hyperscalers have asked us to fit out for potential AI deployment with closed-loop liquid cooling, we’re still air cooled and our deployments are cloud-focused.”
For private equity, there’s also a certain degree of caution that comes with investing in AI workloads. These investors have a different set of pockets than the hyperscalers, who are able to commit country-level spending to a project with, as yet, unproven returns. Private equity firms must balance risk and reward in a way that the large tech companies can simply bulldoze over.
Kujawski says that investing into AI is a risky endeavor. There will be “winners and losers” in the arms race, and it’s as yet unclear who those will be.
“We’re real estate investors at heart, that’s where the DNA of the firm is, and we have to think about the long-term value of that investment,” he explains. “We just think there’s a greater uncertainty as to how this will all play out, and whether you could have orphaned facilities. It’s a different risk profile that, at least from our perspective at this point, is not as compelling on a risk-adjusted basis.”
However, Kujawski says that, regardless of the outcome of the “AI arms race,” the company’s plans center around securing low-latency connectivity in primary markets, which have historically panned out for it regardless of macroeconomic factors.
“Our leases never got thrown out because of who’s in the building,” he adds.
Michael Hochanadel, managing director and head of digital at Harrison Street Asset Management (HSAM), takes a similar view to Kujawski, stating that HSAM has “shied away” from investing in the “classic AI training campuses.”
“In our view, those sorts of campuses have a different exposure to residual value risk than the things we focus on,” he says.
HSAM has $108bn in AUM, with around $6.5bn in data center and digital infrastructure assets, growing from $35 million in 2020. The vast majority of its data center investments serve cloud workloads.
Hochanadel says the firm focuses on primary data center markets first, with some investments in secondary markets that have high development factors. A core component for HSAM, Hochanadel explains, is versatility. While cloud and conventional workloads continue to be the vast majority of underlying compute, Hochanadel says that the key is that those markets can “support any of the underlying use cases, including AI.”
As AI workloads shift from training to inference, Hochanadel says that AI inference will “live in these markets,” as well as major metropolitan areas, requiring different attributes to the training campuses.
Describing his firm’s investments, Sebastian Dooley, senior fund manager at Principal Asset Management, points to similar investments in colocation assets, focused on being in cloud availability zones with established demand, and having an interconnectivity focus.
Dooley says that in discussions with institutional investors, he often finds they have been drawn to the data center industry by the perpetual chatter on AI, but end up “investing more in cloud-focused solutions.”
Investors want exposure to the once-in-a-generation technology shift that AI is promising to the world, but when they’re walked through the actual risk-adjusted returns, proven demand patterns stretching back years are simply more attractive, Dooley argues. AI training campuses in rural America are much harder to underwrite with confidence.
Like Hochanadel, Dooley sees opportunity in AI when it is further commercialized and there is an uptick in demand, but he explains that much of this opportunity, from his perspective, remains in the cloud, where AI will simply cause more data to flow through these sites. There’s a level of safety in the cloud that firms believe will insulate them from any uncertainty around AI commercialization.
“It’s very difficult to imagine a world where those assets aren’t really part of critical infrastructure,“ Dooley says.
With AI hype, there is often a misunderstanding of where the industry is with cloud deployment. The cloud infrastructure buildout is far from complete. Worldwide, investments in the cloud are not slowing. Cloud deployments are critical infrastructure for digital commerce, financial systems, enterprise software, streaming services, and the general use of the Internet. It’s unlikely that cloud assets will depreciate in value or lose their attractiveness as a safe investment. What’s more, all of these use cases are growing alongside AI.
In Europe’s mature cloud market, for example, digital infrastructure is continuing to rapidly grow, making it a highly attractive market for investors. The market remains dominated by the US hyperscalers, but it is becoming distinctively shaped by regulatory and security requirements.
Data sovereignty is the new buzzword in Europe, and it is driving a new wave of investment and offers local providers an opportunity to gain market share. European governments have made data sovereignty a core pillar of their agenda. Dooley says there is a “general push” across the continent for data secured within European borders, and this has led to an interesting dynamic for investors.
Europe is still facing similar challenges to the US, with land and power availability curbing further expansion in primary markets, driving investment into secondary and emerging markets. But there is a massive opportunity here, as Dooley says the sovereignty campaign is leading to a “specialized uptick” in demand and investors are keen to take advantage of this and get “a little bit of diversification in terms of their data center exposure.”
Similar to the US, these challenges are actually a net benefit for investors. Dooley says his firm is seeing “attractive risk-adjusted returns by going in on sites that have power secured.” In these cases, you accept you might pay a bit more for the tier one markets, but the returns can be worth it.
These issues can also allay fears of an AI-induced data center overbuild. Dooley explains that, the more difficult it is to replicate sites, “the lower propensity for that overbuild risk to come through.” This, he says, offers a much better chance for stronger value preservation.”
A compelling equation
AI has flipped the narrative in data center investment, drawing a record-breaking level of investment and attention to the industry. Kujawski, Hochanadel, and Dooley all describe substantial growth in their firms’ digital investments over the past few years, with much of that coming since the early reveals of large language models in 2020. Investors remain extremely bullish on data centers. Even with the risks of AI, the data center market is seen as a safe investment, more so now than ever, with cloud remaining the sturdy workhorse or the calm port in a storm.
Reports from various investment services and research firms point to green indicators across the board for the industry. JLL, for example, said that property metrics “do not indicate a bubble,” with analysis showing the sector maintains “healthy fundamentals with 97 percent global occupancy and 77 percent of the construction pipeline pre-committed to tenants.”
And, in a more recent report, the real estate services company said that it continued to see record-low data center vacancies. In the US, “frontier markets” like Texas are on track to become primary markets. Even those non-traditional data center markets are seeing record-high leasing rates, with capacity pre-leased long before construction is complete.
This is all positive news for the data center investor, but Kujawski notes that real estate investment has a cyclical nature, and being aware of the risks associated with this is extremely important.
“I’m old enough to know that real estate, in particular, goes through its cycles – it typically gets overbuilt at some point, and things come crashing to some realization of capacity relative to demand,” Kujawski says.
Hochanadel adds that it’s “impossible to maintain the current pace” of growth, but concludes that, “even if demand growth moderates, that doesn’t mean the demand for existing capacity goes away – it’s not going to decline. There’s no world, in my view, where the overall amount of compute capacity starts declining.”
He continues: “The growth will moderate, but if you have a campus with power in Northern Virginia, I think it will be needed in some form or fashion in perpetuity. There’s going to be a capex refresh along the way, and so deal structure and financing is important. Matching the duration of your asset with liabilities is important. But I think the long-term residual value of modern data center campuses with power, I think fundamentally, that’s a very compelling equation.
