AI data center tenants now want more than $5 million a day for delays

The companies renting AI data centers have stopped asking nicely. Tenants are now demanding more than $5 million per day in penalties when landlords fail to deliver power or finished facilities on time.

Earlier this year, that figure sat at about $1 million per day. In other words, the price of being late has grown roughly fivefold in a matter of months, and the people building these campuses are the ones holding the bill.

The fine print got expensive

The tenants driving this shift are mostly hyperscalers and AI labs, the deep-pocketed buyers of compute that every developer wants on a long-term lease. Their leverage is showing up in the contracts.

The jump in daily penalties ran from approximately $1 million per day in early 2026 to more than $5 million per day by October of the same year.

Penalties are only one tool in the kit. Tenants are also pushing for rent abatements, liquidated damages and, in some cases, termination rights that are tied directly to when power and the building are actually ready.

The SB Energy and OpenAI arrangement for projects in Ohio shows how sharp these terms can get. Under the deal, OpenAI receives one free day of rent for every day of delay, and that penalty doubles once the delay passes 90 days.

The math compounds quickly. A one-year delay could lead to approximately 640 days of free rent, meaning a project that slips 12 months could hand back close to two years of income.

Oracle is also working the contract levers. The company issued a force majeure notice in September 2026 on its Project Jupiter campus in New Mexico, preserving its rights to defer rent if power delays affect the planned 2028 opening.

Why the power isn’t showing up

The core issue is simple to state and hard to fix. AI campuses need enormous amounts of electricity, and the system that delivers it is backed up at nearly every stage.

Permitting complications, equipment shortages and delays in grid interconnections are all feeding the problem.

Equipment is a particular choke point. Lead times for generators and transformers have stretched to 12 to 24 months, which means developers often have to order gear long before a project is fully locked in.

That pre-procurement requirement is reshaping lease negotiations too. Terms covering rent credits and termination clauses are being rewritten around the reality that a single missing transformer can hold up an entire campus.

Utilities have tightened their own stance as well. They are imposing stricter collateral requirements on large projects, asking developers to put more money at risk before the power gets connected.

The consequences have already shown up on the ground. Numerous large US projects were delayed or canceled in 2026 because of constraints on power, permitting and equipment.

What this means for developers and investors

For developers and landlords, the risk profile of an AI data center build has changed. The construction risk that used to be priced into budgets is now also priced into rent, and a schedule miss can erase revenue the project was counting on.

The OpenAI and SB Energy structure makes the stakes concrete. When a year of delay could cost approximately 640 days of rent, a developer’s real exposure depends less on how fast it can pour concrete and more on how fast a utility and its suppliers can deliver.

That shifts bargaining power in an interesting direction. Developers bear the penalties, yet many of the bottlenecks, including grid interconnection, permits and transformer queues, sit outside their direct control.

Oracle’s force majeure notice is worth watching as a test case. How that dispute plays out, and whether Project Jupiter holds its 2028 target, could shape how aggressively other tenants and landlords use similar clauses.

The broader signal is that AI demand has not slowed, but the physical world is setting the pace. Tenants want capacity badly enough to pay for it and to penalize anyone who makes them wait, and the electricity required to run that capacity is arriving on its own schedule.