By Dawn Lim CBRE Group Inc. executives have a warning for artificial-intelligence upstarts hunting for data centers: Find a serious partner to guarantee lease payments or face an uphill battle securing new sites.

Brokers at the world’s largest commercial real estate services firm told clients in big US markets that such a backstop is a must. Several said they’ll only enter serious talks when there’s a partner with strong credit ratings to stand behind tenants’ obligations.

Their message is targeted at so-called neoclouds, firms that rent out access to highly sought-after AI chips. Despite the market’s explosive growth, developers and banks are skeptical that new entrants with speculative credit ratings will stick around to pay all the dues for scarce power and land.

“It’s challenging enough for the biggest clients to find space,” said Pat Lynch, who leads CBRE’s data-center advisory services group. The firm doesn’t have a blanket ban on doing business with neoclouds, but its job is to educate them on market realities, he said. “If a neocloud has no credit or backing, it’s extremely challenging for them to get capacity.”The moves reflect an industry that’s filtering out the weakest players. To survive, newer AI cloud companies have to lean on the financial heft of deep-pocketed tech giants. They have no choice.

Vacancy in core markets fell to a record low of 1.4% in the first half of this year, according to CBRE,allowing data-center operators to be pickier. As wait times for power and permits grow, and opposition to the AI buildout intensifies, developers are looking to avoid pitfalls by sticking with blue-chip tenants.

Many neoclouds are former Bitcoin miners that took onhuge debts buying specialized chips to pivot to AI computing providers. Major players such as CoreWeave Inc. are racking up net losses even if they grow revenue.

A financial backstop from big tech firms and chipmakers like Nvidia Corp. can open doors and wallets. If a neocloud folds, the guarantors can promise to take over the lease or pay the rest of a contract or termination fee. The deals can draw players into intricate arrangements.

Take Fluidstack, a neocloud that provides access to AI chips from Alphabet Inc.’s Google. When Fluidstack signed a lease with landlord TeraWulf for a data-center campus in Western New York last year, Google offered a $3.2 billion guarantee, reducing risks if parties default on commitments. It also took warrants in TeraWulf, a crypto miner-turned-developer.

It wasn’t disclosed to investors at the time that Anthropic PBC would be the end-user of the chips, said people familiar with the matter. Google’s commitment was insurance, allowing TeraWulf to issue notes to finance the buildout at 7.75%, when it otherwise would have to pay 10% to 12%, according to one of the people.

It’s a playbook for how the industry can open up capital markets for new data centers tied to tenants with weaker credit profiles. The builder can use proof of a backstop to secure cheaper financing.

Neoclouds “have a right to play, but they just need to de-risk themselves as tenants,” said Asya Walters, a senior director at Alvarez & Marsal’s corporate performance improvement practice.

The need for strong contracts and commitments from an investment-grade guarantor is a change from the early days of AI, when the pool of tenants centered around big tech giants. “It used to be a handshake was OK,” Walters said. A recent high-profile backstop involved Nvidia swooping in to commit as much as $105 billion.

When OpenAI leased an Ohio site from SoftBank Group Corp.-backed SB Energy, Nvidia agreed to guarantee a minimum value if OpenAI doesn’t pay its lease. That the world’s most valuable company is standing behind what could be the planet’s biggest AI data center reduces the risks tied to the ChatGPT maker, which has forecast that it won’t generate positive cash flow for years and is pausing training of cutting-edge models. The site has been championed by SoftBank founder Masayoshi Son and the Trump administration, which touted its potential to “reinforce US leadership in the global AI technology race.”

Some in the industry worry the backstops are propping up weaker players that wouldn’t survive on their own. Investors are already nervous about how circular financings — when companies like Nvidia back the very companies driving business to themselves — will amplify industry losses in a downturn. The guarantees pile more shadow obligations on Big Tech.

The top five hyperscalers are projected to spend roughly $750 billion in 2026, and their capital expenditures are rising, according to a report earlier this year from CreditSights. Some big backstop providers to neoclouds have found creative ways to maximize bargaining power and minimize balance-sheet pressure.

They’re avoiding promises to pay all future lease payments, said Eamon Nolan, a project finance attorney at Vinson & Elkins.

For more flexibility,they’re pivoting to so-called residual value guarantees,committing that the data centers will retain some value if tenants bolt. “Even the nature of that backstop is a moving feast,” Nolan said. Many guarantees only get triggered if there’s imminent danger of default.

That’s what can keep these backstops off balance sheets for years. “If it’s a contingent liability, it doesn’t technically have to be recorded on the company’s balance sheet,” said David Trainer, a former member of an advisory committee for the Financial Accounting Standards Board who now runs research firm New Constructs.

“It can be tucked in the footnotes of quarterly and annual financial reports.”In the case of the Ohio campus, Nvidia isn’t guaranteeing the full cost of the site or all of the tenant’s obligations. The company will provide some backstop if the value of a replacement lease or sale involving the data centers fall below a certain level.

Its residual-value guarantees cover an initial 4.25 gigawatts, enough to power at least 3 million homes. The company said it has the option to provide support for roughly 3.75 gigawatts more in the future. The guarantees are triggered “only upon specified tenant insolvency or an uncured monetary default under an applicable lease,” according to a filing.