Data centres are enjoying a boom, with billions of dollars being pumped into new facilities to support increased demand for cloud and storage services, along with the emergence of AI.

But just as projects are exponentially, investors are having to navigate rising interest rates, economic uncertainties, and the looming fear of a market bubble.

Despite economic turbulence, investor appetite for data centres remains strong, with around 6.7 gigawatts of global capacity leased or absorbed in 2024, more than six times the amount in 2020.

As a result, investors are forced to leverage increasingly creative financial structures, diversifying funding sources and fostering partnerships in order to take advantage of the demand while navigating hurdles like high initial capital requirements, power grid limitations, and regulatory scrutiny over environmental impacts.

Sean P. McDevitt, partner and head of USA TMT Practice and global telecom transaction services – Arthur D. Little (moderator)
Sean Kennedy, Managing Director – AB Private Credit
Syed Ahmed, managing director and head of digital infrastructure – Apterra
Kush Urs, managing director for telecom and digital infrastructure, investment banking – Guggenheim Partners
Pim Rothweiler, head of telecoms and technology, Americas – Natixis
Ted Mocarski, senior partner, head of digital infrastructure – Novacap

With traditional capital routes under pressure, experts at Metro Connect 2025 outlined the more creative ways investors are exploring in terms of inventive strategies.

YieldCos, where a company is formed to own operating assets that produce a predictable cash flow, along with joint ventures and joint ventures are all gaining traction in the data centre space.

Another option that was highlighted was the concept of asset recycling models, where stabilised assets are spun off to fund new builds, as another route investors are increasingly turning to in order to get new data centre deals over the line.

Kush Urs, managing director for telecom, digital infrastructure and investment banking at Guggenheim Partners, noted that while some of these models look compelling on paper, the challenge for investors is structuring something the market is truly comfortable with, especially as YieldCos in adjacent sectors like renewables have had a “spotty history”.

Many of these models, like the YieldCo, have a mixed history in adjacent sectors like renewables, so execution remains a key hurdle.

“The market has to come up with something innovative because something's got to give,” Urs said. “The sheer breadth and depth of capital that's required is unprecedented.

“Common equity, or more straightforward debt, is a lot easier to set up and structure. But when you start to think of newer structures and forms, it might make sense on paper, but how do you effectuate that [and] structure something that works?”

One such creative capital structure highlighted during Metro Connect 2025 was over-equitising investments to maximise speed.

A proponent of such a play was Ted Mocarski, senior partner, head of digital infrastructure at Novacap, who admitted the firm is over-equitising investments on smaller data centres to get deals over the line at speed.

“When there's an interesting opportunity or a new customer that we want to lock down, you have to move very quickly,” Mocarski outlined. “Some debt products take time to execute. They can't execute overnight. The risk of the transaction not happening is something that we're concerned about.

“Once we close and deliver on some of the metrics that we lay out on the path of success, we’ll sit down and talk about our permanent capital structure and take the equity back and potentially recycle the equity proceeds into new opportunities.”

This phased approach, Mocarski explained, allows operators to close deals quickly while setting the stage for refinancing once growth metrics are hit. Debt still plays a role, but only after the initial risk is taken off the table.

He added that the speed of execution and execution of debt providers is often faster and more aggressive in companies with bigger portfolios.

“What we're doing at the smaller end, we're buying a market or two markets, so it puts a very fine point on doing the due diligence on that market because if you're wrong, that's your investment. I think the lenders are concerned about that.”

The surge in AI demand has been a game-changer for data centre investment, but it’s also introduced new layers of risk that investors are still learning how to price.

AI-centric workloads now require two to three times the compute power of traditional cloud workloads, resulting in significant shifts in how facilities are designed, cooled, and operated.

“AI data centres are very different from your regular cloud data centres,” said Syed Ahmed, managing director and head of digital infrastructure at Apterra. “There are questions around design, service level agreements (SLAs), and whether the facility has the right liquid cooling capabilities. You’re in untested territory, so the amount of diligence you have to do is considerably higher.”

One major concern is renewal risk — especially with newer AI tenants like CoreWeave or Lambda, who have gone from obscurity to headline capacity consumers in just a couple of years.

While many of these players boast high-profile clients (in CoreWeave’s case, a large share of revenue comes from Microsoft), their long-term viability remains untested.

“To get comfortable, people are looking at look-through credit, take-or-pay contracts, upfront capex contributions, and diversification strategies,” said Urs. “If one AI tenant is your only customer, that’s a very different risk profile than if they’re one of ten.”

Location plays a big role, too. Investors show far more confidence in Tier 1 markets where re-leasing risk is lower. “You can price that risk, but you need to know the market can absorb it if things don’t pan out,” Ahmed added.

The panel also touched on the long-term implications of relying on AI demand. If current growth levels plateau or AI use cases shift, that could reshape investor appetite. As Mocarski put it: “We believe in the tailwinds behind AI, but how it's ultimately used will dictate the strategy. A higher inference world versus a lower inference world calls for a totally different data centre model.”

As capital continues to pour into hyperscale and AI-specific builds, some investors are beginning to ask a hard question: are we heading toward overbuild?

The same week the panel took place, Microsoft cancelled several hundred megawatts worth of its US leases. The hyperscale giant repeated this just a few weeks later, deferring or cancelling leases in the US and Europe with analysts suggesting it was in an oversupply position.

Metro Connect panellists drew parallels with the fibre boom, where intense investment ultimately outpaced demand, leading to price compression and stranded assets.

“You can’t solve it today, but you can think about it,” said Mocarski, warning that today’s 7-year AI contracts may not look so attractive when they’re up for renewal in a saturated market.

Pim Rothweiler, managing director at Natixis, said the panic brought on by Microsoft’s cancellations was overblown, describing it as the “DeepSeek of the week”.

“Yes, there might be a shift from cloud to AI, and yes, they might ask for a design change mid-project,but if you’ve got a 20-year investment-grade offtaker contract, there is no termination for convenience, unless they pay up for the rents they still owe. That’s still a solid story when you talk to your credit officer.”

Sean Kennedy of AB Private Credit noted that as data centre growth slows, heavy capex requirements combined with high leverage could lead to what he called a “debt spiral” similar to what some fibre players experienced.

“We're not there today. We probably won't be there tomorrow, but at some point, as growth slows down, something to be mindful of,” Kennedy added.