Bernstein Warns: A Wave of Neoclouds Will Go Bankrupt Within 1-3 Years

Bernstein Warns: A Wave of Neoclouds Will Go Bankrupt Within 1-3 Years

TREE NEWS reports: A senior data-center legal expert has issued a stark warning that many Neoclouds—the new breed of AI cloud providers—face a structural mismatch that will drive a wave of bankruptcies over the next one to three years. Jeffrey Moerdler, who has specialized in data centers and digital infrastructure since 1990 and handles hundreds of transactions annually, told Bernstein that Neoclouds are “WeWork 2.0.” The core problem: Neoclouds sign 15-year data-center leases while their customer contracts may last only days, weeks, or at best one to three years. If customers fail to renew, revenue collapses while lease obligations remain.

The Structural Mismatch

Moerdler explained that Neoclouds typically lease data-center capacity for 15 years but resell it to AI startups on short-term contracts. This duration gap creates a fatal vulnerability: when a customer leaves, the Neocloud still owes rent for years. “I am confident we will see bankruptcies. Not the biggest AI companies, but smaller AI startups in niche markets—say, AI for law firms or the clothing industry—that cannot generate revenue and will eventually collapse. Neoclouds will face the same fate,” he said.

Credit Guarantees Are Shrinking

The credit support structure in data-center contracts has quietly deteriorated. About 10 months ago, a hyperscaler changed market practice by refusing to provide full lease guarantees, instead covering only about 80% of construction loans rather than all rent obligations. Developers must now fund 40-60% of equity above the construction loan and bear 20% of the loan risk themselves. Moreover, guarantees typically amortize on a straight-line basis over the initial lease term—on a 10-year lease, the guarantee value is essentially zero by year seven or eight. Guarantors are also securing stronger step-in rights, allowing them to take over space and power resources if guarantees are called. This shift has spread to nearly all guarantors in the past 10-12 months.

Four Contract Landmines

Moerdler identified four often-overlooked contract traps:

• Termination-for-Convenience: Hyperscalers’ standard contracts almost always include this clause, often buried in fine print. Developers who accept it surrender long-term control.
• Insufficient or fast-amortizing credit enhancement: Guarantees may be worthless by the later years of a lease.
• Lease assignment erasing credit protection: If a lease is transferred without preserving credit enhancement, developers lose their key risk buffer.
• Excessive tenant control over future development: Expansion rights, contraction rights, and control over adjacent parcels can severely limit developers’ asset flexibility.

Bankruptcy Could Benefit Prepared Developers

For developers who secured credit enhancement upfront, a Neocloud bankruptcy may not be disastrous. Moerdler offered a math example: if construction costs $10 million per megawatt and the developer obtained 30% prepaid credit enhancement, then after a Neocloud bankruptcy and reclamation, the effective cost basis is $7 million per megawatt. “You now own a fully built data center with a $7 million per megawatt cost basis, ready to deploy immediately. That is a highly competitive asset,” he said. However, this logic depends on rapid reclamation—in markets like New York, recovering a property can take one to two years, eroding the advantage.

Power Is King

Data-center site selection has shifted from “location, location, location” to “power, power, power.” Moerdler outlined a tiering system: power available within 12 months is platinum; 12-24 months is gold; 24-48 months is silver; beyond 48 months, “don’t call me, I’ll call you.” He noted that ERCOT’s interconnection queue in Texas is four times the state’s existing grid capacity, much of it speculative. Regulators are screening out speculators via deposits and security requirements.

Demand Still Hockey-Stick

Despite skepticism about AI infrastructure spending, Moerdler’s frontline view is that demand has not slowed. “We are going up the hockey stick,” he said, citing a 3 GW data-center campus in Texas and a ~$2 billion sale of a 2+ GW Batch Zero campus. He expects a shift over the next 3-5 years toward smaller inference nodes in cities and suburbs, where supply of 250 kW to 2 MW local data centers is tight and pricing has risen 50-100% in two years.

Key Takeaways for Investors

• Neocloud credit risk is real: The duration mismatch between long-term leases and short-term customer contracts is a fundamental flaw that will likely trigger bankruptcies among smaller players.
• Watch credit enhancement terms: Shrinking guarantees and faster amortization mean developers and their financiers bear more risk.
• Power availability drives asset value: Data centers with near-term power access command premium valuations; speculative projects face financing hurdles.
• Opportunity in distress: Well-capitalized developers with strong credit support may acquire distressed assets at attractive cost bases.
• Inference demand emerging: The next wave of AI infrastructure demand may come from smaller, urban inference nodes, creating opportunities in niche data-center markets.