Neoclouds have emerged as the fastest-growing niche within the data center landscape, with the sector’s revenues nearly doubling on an annual basis. But these AI cloud specialists also carry additional risks that data center developers, operators and investors are still figuring out how to navigate.
Neoclouds — AI-focused cloud providers like CoreWeave that offer on-demand access to the graphics processing unit computing critical for AI — have seen a compound annual growth rate in revenue of 82% since 2021, according to a new report from JLL.
The report described the neocloud segment’s growth as “unrivaled,” even within the booming cloud service and data center industry. It is being driven by seemingly insatiable corporate demand for AI that traditional cloud providers like Amazon Web Services, Microsoft and Google have been unable to meet alone.
But even as these startup AI cloud specialists offer many customers advantages over the established as-a-service giants, there are concerns that fundamental aspects of the neocloud business model bring risks that data center landlords and investors may not be comfortable with.
“Funding will be a major factor to translate the potential of neoclouds into a reality capable of handling the AI load,” Muhd Syafiq, director of JLL’s Asia-Pacific Data Center Research group, said in a statement. “Building GPU infrastructure is capital-heavy, and investors should have a clear vision for delivering a viable business model and support from key clients before undertaking an entry into the neocloud space.”
The nascent neocloud sector is dominated by a few major players, with CoreWeave by far the most prominent. Other key firms include Crusoe, Nebius, and Lambda. Additionally, the rapidly evolving ecosystem has approximately 190 smaller individual operators.
The industry’s steep growth rate isn’t the only indicator of its momentum. CoreWeave, despite a lukewarm entrance onto public markets in March, has seen its share price triple since the IPO. Meanwhile, over $10B in private investment was allocated to the industry last year, a trend that has continued into this year with upstarts like Nebius landing funding rounds of over $1B, according to JLL.
GPU cloud specialists have come to fill a market need as corporate giants clamber for AI computing power. Not only are GPUs expensive and difficult to acquire, but they also require specialized data centers with advanced cooling systems capable of handling high power densities. Firms with AI computing needs that either operate their own data centers or utilize colocation may not be able to support GPUs with their current infrastructure.
The major hyperscalers also have their own GPU cloud products. But neoclouds have emerged as a preferred option, or as a supplement to the established cloud giants, due to superior speed to market and lower pricing, according to JLL.
For firms that need GPU computing at a large scale quickly, neoclouds offer rapid deployment. While traditional hyperscalers add GPU capacity through large-scale data center buildouts that can take years, neoclouds can often have GPU infrastructure up and running within months.
But neoclouds’ primary competitive advantage is price. According to Uptime Institute, these GPU-specialty providers offer, on average, a 66% cost reduction compared to hyperscalers, the result of being able to locate computing in cheaper locations than traditional cloud firms due to fewer latency restrictions and avoiding infrastructure costs needed for non-AI cloud services.
Neoclouds also offer shorter contract terms tailored to the way companies generally use AI-specific computing. This makes them far more cost-effective compared to hyperscalers for firms whose GPU computing needs are variable or intermittent.
“Neoclouds have developed an advantage over traditional cloud providers by moving faster and pricing lower with flexible terms,” Andrew Batson, JLL’s head of data center research for the Americas, said in a statement. “As AI shows no signs of slowing, its success will rely on accessibility to GPU infrastructure, which neoclouds specifically cater to.”
Still, some of these advantages for customers present risks for data center developers, operators and investors with neoclouds as tenants, particularly the shorter contract terms compared to hyperscalers.
While traditional data center leases span 10 to 15 years, neocloud GPU contracts are typically just two to five years. In some cases, the contracts are far shorter than that. It’s a meaningful mismatch in an industry with asset payback periods of seven to nine years. One industry insider compared this to the fundamental flaw that damaged firms in the coworking office sector.
“I have a serious concern about those GPU-as-a-service companies,” Jeffrey Moerdler, a longtime data center attorney and member at Mintz, said at Bisnow’s DICE Capital Markets event in July.
“I’m worried that there’s a mismatch between their leasing space on a 10 to 15-year lease, but their customer-facing agreements range from by the hour to by the week to by the year. It’s kind of a WeWork mismatch.”
Because of these risks, performance guarantees and tenant creditworthiness assessments have become standard in neocloud lease agreements, with many operators requiring significant up-front collateral or a third-party guarantor. While the neocloud sector may be growing, data center firms used to anchoring facilities with credit-grade tech giants remain wary of an industry that has yet to separate the winners and the losers among its nearly 200 entrants.
“I expect we’re going to see a bunch of those companies, as well as a bunch of AI companies that don’t generate revenue, going bust in the next few years,” Moerdler said. “That’ll be a traumatic point for the industry.”