In the data center business, lateness used to be a construction problem. A project slipped, the contractor paid liquidated damages, the opening moved a quarter, and everyone moved on. That world is ending. Tenants of new gigawatt-scale data centers are now asking developers for late-delivery penalties of $5 million or more per day, up from about $1 million as recently as this spring, according to reporting by The Information summarized by Dealroom on October 25. The fivefold jump in a matter of months is the clearest sign yet that schedule risk has become a financial product in its own right — priced, traded, and insured, with the bill landing on the party least able to carry it.
The reason is leverage. The tenants writing these terms are hyperscalers and AI labs, the deep-pocketed buyers every developer wants on a long-term lease. When compute capacity is the scarce input to an AI race, a month without a building is not an inconvenience; it is lost training runs, missed product windows, and revenue that cannot be recovered later. Tenants have begun to price that loss into the lease itself, and developers — competing for a small number of creditworthy tenants — have had little choice but to sign.
What $5 million a day actually means
The arithmetic is what makes the new terms so dangerous. A penalty of $5 million a day compounds to $150 million in a single month and $300 million in two months. Dealroom’s account of the reporting cites Occam Edge’s Christian Okoye for the estimate that two months of delay on a one-gigawatt shell could mean roughly $300 million in penalties plus lost rent — a figure that can exceed the equity in a thinly capitalized developer’s project.
The terms are getting sharper in structure as well as size. Under OpenAI’s arrangement with SB Energy for a 10-gigawatt campus in Ohio, OpenAI receives one free day of base rent for every day of delay, and that penalty doubles to two free days once the delay passes 90 days, according to the Wall Street Journal’s reading of SB Energy’s IPO filing. A full year late would cost roughly 640 days of free rent — close to two years of income handed back on a project that has not yet earned a dollar. Oracle, meanwhile, issued a force majeure notice in September on its Project Jupiter campus in New Mexico, preserving its rights to defer rent if power delays affect the planned 2028 opening, while saying the project remains on schedule. The notice itself is the tell: tenants are now lawyering for delay before the delay happens.
It is worth being precise about what these figures are. They are demands in new lease negotiations, reported by The Information and corroborated across several summaries of its reporting — not a published schedule of settled contracts. Not every tenant will get $5 million a day, and not every developer will agree to it. But the direction of travel is not in doubt, and the fact that brokers are building products around the number tells you the market treats it as real.
Why power delays, specifically
Not every delay is equal, and the penalties bite hardest where the delay is least within the developer’s control. A building can be built faster with more crews. A substation cannot be willed into existence, and a transmission upgrade cannot be expedited past permitting, land acquisition, and construction. As this desk has covered before, the gap between a data center’s desired energization date and the grid’s ability to serve it is the central tension of large-load interconnection — and it is measured in years, not months.
That is why power-access delays have become the trigger event for this new financial architecture. When a tenant’s penalty clock starts on the delivery of power and finished facilities together, the developer is effectively underwriting the utility’s construction schedule, the regulator’s permitting timeline, and the equipment supply chain, all at once. Goldman Sachs expects only about half of scheduled U.S. data center capacity to arrive on time over the next one to two years, according to summaries of its forecasts — a coin flip that makes a $5 million daily penalty look less like a deterrent and more like a lottery ticket the developer did not choose to buy.
The interconnection queue is the mechanism that turns a schedule slip into a penalty event. A project that filed early, studied cleanly, and secured its upgrades on time is insulated. A project that filed into a congested queue, watched its studies get redone as speculative projects ahead of it dropped out, and waited on a transmission upgrade is exposed — through no fault of its own construction management. The penalty does not distinguish between a developer that mismanaged its build and a developer that was queued behind a utility’s five-year transmission plan. That indiscriminate quality is exactly what makes it insurable, and exactly what makes it frightening to lenders.
The insurance industry’s answer
Where a new risk appears, brokers follow. Marsh, one of the two brokers Dealroom names as arranging most of this cover alongside Aon, had already been building toward it. Its Nimbus facility, launched in June 2025 for large data center construction, started with up to $1 billion in construction all-risks limits and $365 million for delay in start-up, according to Marsh’s own announcement and trade coverage. In January, Marsh raised Nimbus limits to as much as $2.7 billion, according to summaries of the reporting.
But Nimbus covers physical damage and the resulting delay in start-up. The newer product goes after a different bill: the late fee the developer owes its tenant. Mike Mathews, who runs Marsh’s 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, The Information reported. Within months, fresh deals had the ask at $5 million and up — and the policy concept had a market.
The distinction matters. Traditional delay-in-start-up cover pays when a fire, storm, or construction accident pushes back completion. Delay-in-delivery cover pays when the developer owes its tenant because the building or the power was not ready — whatever the cause. That is a much broader promise, and underwriters are pricing it accordingly. The cover is bespoke, arranged deal by deal, and its terms — triggers, waiting periods, exclusions — are where the real negotiation now happens. A policy that pays only after a 90-day waiting period is a very different instrument from one that pays from day one, and a policy that excludes utility-side delays is nearly useless for the risk developers actually face.
Dealroom’s account adds two details that show how far this market has already spread. Marsh is Meta’s broker on its El Paso project and found more than $5 billion of credit protection for one developer’s utility commitments — a reminder that the utility itself is now a counterparty whose promises need backing. And Hut 8’s Corpus Christi site carries 19 months of delay-in-startup cover arranged through FM via Aon, an unusually long tail that suggests underwriters and developers alike are planning for delays measured in years, not weeks.
A market growing faster than its own models
The penalty story sits inside a much larger insurance story. Global insurance premiums tied to data centers are expected to climb to $24.2 billion by 2030 from an estimated $10.6 billion in 2026, according to the Swiss Re Institute. S&P Global puts the 2026 figure at around $10 billion — roughly twice the size of global aviation insurance, which covers every commercial airline fleet on earth. A category that barely existed as an insurance line five years ago is now larger than one of the industry’s oldest.
That growth reflects genuine exposure, not just premium inflation. Swiss Re notes that a single site can cost more than $20 billion to build — doubling once servers and graphics processing units are installed — and that a single location could cause close to $10 billion in natural catastrophe losses. More than 40 percent of U.S. data center capacity sits in areas with significant tornado exposure, and more than a quarter in large-hail zones. During a Marsh press webinar covered by Reinsurance News, Mathews put the underwriting problem plainly: a $4 billion data center can hold $10 billion of equipment in a single facility. Spreading that risk across multiple carriers is, in his view, the trend — because no single carrier can take it whole.
The emerging risks Dealroom flags point to where the next products will be written. GPU residual value — what a rack of accelerators is worth if a tenant defaults or a technology generation turns over — is an underwriting question with almost no loss history behind it. Utility guarantees, the promises developers make to secure interconnection and capacity, are becoming credit instruments in their own right. And Marsh’s observation that labor shortages are uninsurable is a quiet warning: the industry can price a storm, a fire, and a late substation, but it cannot price a workforce that does not exist.
Where the risk actually sits
Strip away the insurance language and the picture is a three-way negotiation over who bears the cost of a grid that cannot keep up with demand. The tenants have the leverage and are using it: penalties transfer schedule risk to developers. The developers have the least balance sheet and the most exposure: a first-time developer with one gigawatt campus and $5 million a day in penalties is one bad interconnection study away from insolvency. KKR’s Waldemar Szlezak captured the dynamic in the line Dealroom quotes from the reporting — that late penalties “will bankrupt more developers than interest rates ever will.” It is a vivid claim, and it should be read as a warning from a lender’s perspective rather than a forecast, but it identifies the right failure mode: developers do not usually die from expensive debt, they die from a cash-flow cliff they cannot refinance their way out of.
The lenders, for their part, are responding the way lenders do: by demanding more protection. Credit protection for utility commitments, longer delay-in-startup tails, and insurance proceeds assigned to the loan are all ways of making sure that when the penalty clock runs, someone solvent is standing behind it. That protection costs money, and the cost lands in the project’s capital structure — higher insurance premiums, larger contingency reserves, more expensive debt — which in turn makes the next project harder to finance. It is a feedback loop: penalties raise financing costs, financing costs thin the developer pool, a thinner pool means more first-time developers taking bigger risks to win tenants, and bigger risks mean more penalty events.
There is a second feedback loop, running through siting. If penalties are priced by delay risk, then sites with ready interconnection — hosting-capacity-rich locations, phased energization options, flexible interconnection terms — become dramatically more valuable, and speculative sites in congested queues become dramatically less financeable. That is, in principle, a healthy market signal: it steers development toward grid-ready locations. But it also concentrates development further into the handful of regions that can offer speed, worsening the clustering dynamic behind community pushback, water conflicts, and ratepayer concerns.
What to watch
Several questions will determine whether delay-in-delivery cover becomes a standard product or a cautionary tale. The first is loss experience: the first time a major policy pays out on a utility-caused delay, the industry will learn what these contracts actually cover — and what they exclude.
The second is whether tenants keep escalating. If $5 million a day becomes $10 million, the insurability question flips: at some point the premium for covering the penalty exceeds the developer’s margin on the project, and the product prices itself out of existence. That ceiling is itself informative — it marks the point where the market admits that schedule risk cannot be transferred, only borne.
The third is regulatory. If penalty-driven developer failures start landing on ratepayers — through stranded utility infrastructure built for projects that collapsed, or through guarantee calls on utility commitments — regulators will notice. The interconnection reforms already underway, from first-ready-first-served queues to hosting-capacity maps, are attempts to reduce the delay risk at its source. If they succeed, the penalty market shrinks on its own. If they fail, the penalty market grows, and the question of who pays for a late gigawatt campus becomes a political one, not just a contractual one.
For now, the $5 million day is best understood as a price signal from the most informed buyers in the market. Hyperscaler tenants are telling developers, in the only language that reliably moves construction schedules, that on-time delivery of power is worth more than almost anything else in the deal. Developers are answering by buying insurance against their own lateness. Insurers are discovering that the hardest risk in the AI build-out is not a storm or a fire — it is a calendar date, set by a utility, that nobody controls.