'WeWork 2.0': Booming AI Startups Concern Some Data Center Execs

Artificial intelligence-focused computing providers called neoclouds represent the fastest-growing segment in the booming data center market.

But some data center executives are skeptical about the business models underpinning this startup-dominated subsector. And the riskier tenants have created financing challenges for an industry in which development funding traditionally requires a lease from one of the world’s largest tech firms.

As data center development has surged across the U.S. — first driven by cloud computing and more recently by AI — the capital sources funding this nearly $2T building boom have remained risk-averse.

Given the massive scale of capital required and the narrow, long-term returns data centers deliver, the institutional investors fueling the sector’s expansion have largely backed only the safest of projects: those with investment-grade Big Tech tenants like Amazon, Microsoft, Google or Meta already under contract. 

While these hyperscalers still snap up the lion’s share of new data center capacity, the riskier neocloud tenants account for a rapidly growing slice of the demand pie. 

With major players including CoreWeave, Core Scientific, Nscale and Lambda, these AI-specific cloud providers offer on-demand access to the graphics processing unit computing critical for artificial intelligence. The sector posted 223% year-over-year revenue growth in the fourth quarter, according to Synergy Research Group.

The neocloud ecosystem has more than 190 operators, according to JLL, most of them startups. And despite the segment’s rapid expansion, even the largest of the neoclouds, publicly traded CoreWeave, is far from the kind of investment-grade tenant that has long been the benchmark for data center investment.

Rising demand from neoclouds has created significant challenges for a sector where the bulk of the capital flooding into the market is looking for a sure bet, industry leaders said at Bisnow’s Data Center Capital Markets Summit, held July 16 at the New York Marriott Marquis. 

To make neocloud-anchored projects financeable, data center providers and developers have increasingly turned to creative lease structures and service agreements, often involving investment-grade guarantors or significant upfront collateral.

At the same time, a new class of investors and lenders has emerged with a greater tolerance for this higher-risk segment of the market. 

As a result, there is a growing pipeline of data centers tailored to neoclouds. But longtime industry leaders remain wary, warning that facility owners must fully understand the risks involved before tying their future to companies that are carrying far more uncertainty than the industry has traditionally been willing to tolerate. 

“If you try to focus on some of the neoclouds, you have to really start peeling the onion and make sure you don't start crying,” said Phillip Koblence, co-founder and chief operating officer of data center provider NYI. “How confident are you that they're going to be there?”

He was one of multiple data center executives at the Bisnow event who expressed skepticism about the fundamental sustainability of the neocloud business model. 

Many neoclouds operate like coworking firms: They sign long-term leases for data center capacity, then rent AI computing power to customers on much shorter, more flexible terms — sometimes by the hour.

This leaves them particularly vulnerable if AI demand softens or customers shift workloads to their own AI infrastructure, said Jeffrey Moerdler, chair of the data center and digital infrastructure practice at Haynes Boone.

“The problem is, to me, that neoclouds are WeWork 2.0,” Moerdler said, referring to the coworking operator that was reduced to a fraction of its former value after business-model concerns led to a failed initial public offering and a bankruptcy filing. 

“There is a mismatch. They're signing 15-year data center leases, and their customer-facing agreements are by the hour, the day, the month, max one to three years. What happens when those agreements expire?”

Moerdler said it is particularly alarming that many of the largest neocloud customers are hyperscalers using them as a temporary source of GPU capacity while they build their own AI infrastructure. As that infrastructure comes online, he fears demand will shift away from neoclouds, potentially triggering bankruptcies among companies that fail to replace the lost business. 

Not all industry leaders share such a bleak view of neoclouds’ future. Other panelists argued neoclouds will continue to play a critical role in the fast-evolving AI infrastructure landscape.

Yet even neocloud bulls acknowledged that the sector’s infancy, marked by short track records and uncertain long-term demand, has made leasing and financing data center projects significantly more complex. 

“We’re big believers in many of those businesses, but this introduces challenges,” said Fentress Boyse, a member of management at Partners Group’s private infrastructure practice. 

The biggest challenge is evaluating the creditworthiness of neocloud tenants, panelists said.

Unlike hyperscale tenants with long operating histories and massive balance sheets, neocloud firms require developers and lenders to conduct far more detailed and comprehensive analyses of their business models, revenue streams and default risks. 

As a result, leases and service contracts have become more customized. Developers are increasingly structuring deals with protections designed to reduce risk for both the landlord and lenders, making projects financeable despite uncertainty surrounding some neocloud tenants. 

“The risk analysis has become a critical component of getting a deal done now, as has creativity and finding an avenue that works,” Moerdler said. “Every deal is bespoke these days.”

As with any tenant considered a credit risk across the commercial real estate market, some neocloud lease agreements are backed by hyperscalers or institutional investors with strong balance sheets, which either guarantee the tenant’s obligations or agree to assume the lease in the event of a default. 

Other deals are being secured with the AI chips themselves. Because high-end GPUs from companies like Nvidia remain in short supply, some lenders are willing to accept them as collateral to backstop a loan. In March, CoreWeave closed an $8.5B GPU-backed financing that became the first such debt issuance to receive an investment-grade rating. 

Many large-scale, neocloud-anchored developments rely on an investment-grade credit “wrapper,” in which a hyperscaler guarantees some or all of the debt supporting the project. The guarantee effectively upgrades the credit quality of the financing. This structure has become increasingly common as campus-scale developments are backed by structured debt. 

Frequently, these deals involve a hyperscaler committing to purchase compute capacity from the neocloud, with payments passed through directly to the landlord or the hyperscaler assuming the lease if the neocloud defaults. This type of structure has underpinned some of the sector’s largest deals, including Microsoft’s $33B agreements with CoreWeave and Nebius, deals that guaranteed the financing of the data centers built to serve those contracts. 

As neoclouds account for a larger share of demand, the execution of these deals is also becoming more sophisticated and complex, with a growing degree of nuance and variation in every deal, said John Day, chief commercial officer at CleanArc Data Centers.

Common sources of complexity include whether a wrapper covers all of a tenant’s obligations under a lease or could just cover the debt obligation, or whether the guarantor is able to take equity in a project.  

“Every deal is very different and takes a lot longer,” Day said. “These things aren't grooved like they used to be. You used to be able to do a deal with Amazon, and they’ve done 700 of them, so they just check them off.”

While developers are structuring deals to make neocloud data centers more palatable to the industry’s traditional capital sources, some panelists highlighted an emerging segment of lenders and equity investors focused specifically on this riskier end of the data center market.

These firms are willing to accept higher risk for higher reward. Rates on neocloud debt financing are around 275 to 325 basis points higher than for investment-grade, said Primary Digital Infrastructure President Dave Ferdman.

According to Ferdman, the data center financing landscape is bifurcating, with one set of lenders and equity investors focused on projects for investment-grade tenants and the other set focused on those below investment-grade, which he said is harder to execute. 

“We have a tale of two worlds — it's binary,” Ferdman said. “We have two different universes of investors and two different types of credit. … I think those universes don’t come back together.”