Bernstein: Neocloud Lease Structures Have Fatal Flaws, Bankruptcy Wave Likely Within 1–3 Years
nashnova research
Bernstein cites data-center legal expert Jeffrey Moerdler's warning that a wave of Neocloud firms will go bankrupt within one to three years due to a fatal mismatch between 15-year leases and short-term customer contracts — yet developers with proper credit protections may profit from the fallout.
Why are Neoclouds being called "WeWork 2.0"?
The core problem is a term mismatch: Neoclouds sign 15-year data-center leases, but their customer contracts run from days to, at best, one to three years.
This means → once a customer leaves, revenue vanishes while the lease obligation stays — the cash-flow gap cannot be bridged.
Moerdler says Neoclouds "worry me — they are WeWork 2.0," and expects bankruptcies to concentrate among small and mid-size AI startups, not the largest AI companies.
How has the guarantee structure changed?
Over the past roughly ten months, credit guarantees have shifted fundamentally: one hyperscaler moved first, replacing full lease guarantees with coverage of only about 80% of construction loans.
In plain terms = landlords used to hold a "full insurance policy"; now they get 80 cents on the dollar — and it shrinks every year.
Guarantees typically amortize on a straight-line basis over the lease term. On a 10-year lease, by year seven or eight the guarantee is essentially worth zero.
Moerdler says every single deal he has handled in the past ten to twelve months involving hyperscalers, AI firms, or chipmakers has shifted to the limited-guarantee model.
What four "traps" are buried in the contracts?
Termination-at-will clauses: hyperscaler standard contracts almost always include them, often buried deep in the documents — yet some developers still accept them to win the tenant.
Credit enhancement that amortizes too fast: by late in the lease term, the guarantee may be worthless — protection is weakest precisely when it is needed most.
Lease assignment erases credit protection: when a lease is transferred, the original credit enhancement may not carry over, leaving the new landlord exposed.
Excessive tenant control rights: expansion rights, contraction rights, and control over adjacent-parcel development — all severely limit the developer's freedom over their own asset.
Could developers actually profit from Neocloud bankruptcies?
Moerdler lays out the math: assume construction costs of $10 million per megawatt. If a developer secured 30% upfront credit enhancement, reclaiming the facility after bankruptcy drops the effective cost basis to just $7 million per MW.
This means → the developer can re-lease at prices well below market and still earn more than when the Neocloud was in place.
The catch: reclaiming the facility must happen quickly. In jurisdictions like New York, recovering a property can take one to two years, significantly eroding the advantage.
Why are power timelines and inference nodes the next battleground?
Moerdler ranks power availability by asset value: available within 12 months is platinum, 12–24 months is gold, 24–48 months is silver, and beyond 48 months "has essentially no commercial appeal."
Pending power applications on Texas's ERCOT grid are four times the grid's current capacity, with a large share being speculative filings.
This reflects a structural shift over the next three to five years: as large-scale AI training demand plateaus, the market will tilt toward small urban and suburban inference nodes — pricing for local data centers in the 250 kW to 2 MW range has already risen 50%–100% over the past two years.
市场有风险，内容仅供研究参考，不构成投资建议。