Late penalties on new AI data center leases have climbed about fivefold since spring. Tenants negotiating new leases are asking developers for more than $5 million for each day that power and buildings aren’t ready, The Information reported Sunday, and they justify the figure with lost-revenue estimates that run higher still.
The Information’s account starts at insurance broker Marsh. Mike Mathews, who runs its global digital infrastructure practice, had his team put together a custom “delay-in-delivery” policy this spring for a developer whose tenant wanted $1 million a day if the first phase of a gigawatt-scale campus missed its date. Within months, The Information reported, tenants in fresh deals had moved the ask to $5 million and up.
Why tenants expect the dates to slip
The schedules give them reason. In a May analysis built on Aterio’s facility-level tracking, Goldman Sachs Research estimated that about 60% of U.S. data center capacity scheduled for the next year would come online on time, and only about half of what’s slated for the next two years. Historically, about 72% of capacity scheduled for the following four quarters has opened when promised. Analysts Hongcen Wei, Daan Struyven and Samantha Dart named supply chain and labor shortages as the most common causes, and put a typical build at 18 to 24 months once permits are secured. Even after those haircuts, Goldman still expected 11.5 GW of new capacity in the final three quarters of 2026, following 2.2 GW in the first quarter. The same team expects reliability risks in the Mid-Atlantic, Mid-Continent and Northwest power markets, where planned generation trails incoming demand, and wrote that those regions may end up turning some future data centers away.
Power is the other choke point. Microsoft ended its agreement for Nscale’s planned site in Loughton, Essex after grid delays pushed electricity availability into the 2030s. Oracle went the other direction and started trucking compressed natural gas to sites in Utah and Texas, paying about four times the cost of piped supply to keep projects moving while pipelines catch up.
Financing costs have risen, too. Bonds issued by CoreWeave, the GPU cloud provider carrying about $50 billion in debt, were yielding close to 15% earlier this month. Its quarterly interest expense reached $640 million in the June quarter, up from $267 million a year earlier.
Insurers are writing new paper for it
Marsh had been building toward this kind of cover for a while. Its Nimbus facility launched in June 2025 for large data center construction projects, starting with up to $1 billion in construction all-risks limits and $365 million for delay in start-up, Business Insurance reported. In January, Marsh raised Nimbus limits to as much as $2.7 billion, including delay-in-start-up and business interruption coverage, for projects in the U.K., U.S., Canada, Europe, Australia and New Zealand. The broker said at the time that Nimbus was in use on numerous projects in the U.S., U.K. and the Netherlands.
The newer product goes after a different bill. The policy Mathews’ team arranged this spring, which The Information described as novel, covers the late fee a developer would owe its tenant.
Scale is the underwriting headache. During a Marsh digital infrastructure press webinar that Reinsurance News covered in March, Mathews said: “You have a $4 billion data centre with $10 billion worth of equipment in a single site facility. So, the size and scale is the challenge.” He said spreading that risk across multiple partners and carriers would be the trend.
The pipeline those policies would sit behind keeps growing. Goldman forecasts U.S. data center power demand rising from 31 GW in 2025 to 66 GW in 2027. On Aterio’s schedules, developers are due to add 36.3 GW of U.S. capacity in 2027, more than four times the 8.5 GW that actually came online in 2025.