Finding the Exit – Dec. 1, 2025
Most stakeholders expect to exit their AI infrastructure investments in five to 10 years. The question is how they’re going to do it
The global appetite for data centers—from both a user’s and an investor’s standpoint—has grown so ravenous that 10-figure commitments hardly cause a stir. Now we’re seeing 11-figure projections becoming commonplace.
In the past few weeks, we have seen Alphabet pledge $40 billion for artificial intelligence and cloud computing infrastructure in Texas. Anthropic has announced it would spend $50 billion on custom-built data centers in Texas and New York, and Amazon last week announced a comparable investment to serve federal government customers of its Amazon Web Services platform.
Not to be outdone, Brookfield has launched a $100-billion global AI infrastructure program in partnership with Nvidia and the Kuwait Investment Authority. Brookfield said the program will encompass “investment across every stage of the value chain—from energy and land to data centers and compute.”
Sikander Rashid, head of AI infrastructure at Brookfield, said, “AI is creating one of the largest infrastructure buildouts in history, comparable to the formation of the modern power grid and global telecom networks, but unfolding at a far greater pace and significantly larger scale.”
However, it’s worth asking what happens after the gold rush, and PitchBook News’ Madeline Shi posed the question: what’s the exit strategy? “While less of a problem for corporates, the endgame is important to the institutional investors that back private equity and infrastructure funds,” she wrote.
To date, devising an endgame for AI infrastructure complexes that may span thousands of acres hasn’t gotten much attention, at least in terms of elucidating a concrete strategy. “Few investors are large enough to buy such mammoth companies or even an individual data center,” Tom Mannion, the global telecommunications leader at BDO Global, told PitchBook. “I mean, they are valued in the tens of billions of dollars.”
A survey conducted this summer by management consultancy AlixPartners found that most stakeholders expect to exit their data center investments in five to 10 years. The survey polled more than 400 private equity firms, venture capital funds, lenders, data center developers and operators, along with services and component providers.
“Exit horizons are complicated by the fact that no one can say with confidence whether the returns being underwritten today will hold up,” wrote Shi. “Overcapacity is a potential risk, with some warning that the industry’s frenetic spending on build-out may eventually result in more data centers than needed.”
Shi cited the post-pandemic surge in warehouse construction as a potential frame of reference for this risk. U.S. warehouse vacancies peaked at 7% in the second quarter of 2025 and stayed at that level in Q3, according to Cushman & Wakefield, which expects the vacancy rate to continue climbing.
It’s a prospect that Kipp deVeer, co-president of Ares Management, is well aware of—even as the firm raises its fundraising target for data center investments. Speaking to Bloomberg News at the Greenwich Economic Forum in Connecticut this past October, deVeer said, “If you look historically in areas like this over the past 20 or 30 years, typically when this much capacity comes online, some of it at the end of the day has to be marginal. These trends tend to lead to overbuilds in certain places, so us being selective and measured in what we build is important.”
The risk could intensify if AI model training becomes more efficient, or if workloads shift away from model training to inference, in which AI models “start to use their own learned knowledge to make decisions rather than relying on training data,” wrote Shi. “Either scenario will diminish the demand for these colossal data center sites designed exclusively for training AI models, which are difficult to repurpose.”
For investors in digital infrastructure, the public markets and M&A represent potential exit pathways, Andrej Danis, who heads AlixPartners’ Americas telecoms and digital infrastructure practice, told Shi. They’re not necessarily obstacle-free pathways, though: potential shareholders may balk at the amount of debt taken on by data center projects, and few buyers have access to the cash needed to acquire the properties outright, at least at current valuations.
Danis said creative ownership structures may emerge among PE funds, including a “loan to own” model for distressed properties. Yet he added, “For these mega data centers, it is not yet clear what will be the eventual vehicle to hold them. Everything here is as speculative as it gets.”