Longer power contracts for artificial intelligence (AI) data centers in the United States are reportedly increasing developers' financial burdens.
In particular, as rules requiring data center developers to bear grid expansion costs become more widespread, analysts see a new profitability risk in arrangements requiring minimum electricity payments over long periods even if future AI demand falls short of expectations.
According to Ohio Power Company's data center tariff, Schedule DCT, as of November 6, new data centers with contracted demand exceeding 25 MW and developers undertaking major expansions of existing facilities must sign long-term power supply contracts lasting up to 12 years. The utility was formerly known as AEP Ohio.
The contract term consists of a phased power ramp-up period of up to 4 years, followed by 8 years. Developers must pay minimum demand charges calculated using contracted demand and other factors even when actual consumption is low.
Specifically, monthly billing demand is the higher of 85% of the highest billing demand during the preceding 11 months and a minimum based on contracted capacity, with the contracted-capacity minimum capped at 85%.
This means that certain costs associated with securing grid capacity may remain even if a data center's actual power demand declines.
The financial burden extends beyond minimum charges.
Ohio Power requires developers that do not meet specified credit rating and liquid asset requirements to provide a guarantee or collateral equivalent to 50% of total minimum charges over the full contract term.
The obligation to pay minimum charges continues even if a developer reduces power use or ceases operations during the contract term. Early termination after 5 years have elapsed following the end of the phased power ramp-up period requires a payment equivalent to 36 months of minimum charges.
Such long-term contracts are also spreading to other parts of the United States.
In January 2025, the Georgia Public Service Commission (PSC) imposed minimum charge conditions on new large electricity customers with demand of at least 100 MW and allowed contract terms to be extended from the previous 5 years to as long as 15 years.
In April, the Wisconsin Public Service Commission also approved utility We Energies' tariff for large customers, extending the minimum contract term from the company's proposed 10 years to 15 years and lowering the applicability threshold from the proposed 500 MW to 100 MW.
These measures are intended to prevent the costs of power infrastructure expanded for data center construction from being passed on to ordinary electricity consumers.
Delayed AI monetization adds to fixed costs as big tech reviews contracts
The problem arises if actual revenue generation lags behind aggressive investment in AI infrastructure.
Data center developers must continue paying minimum charges under long-term power contracts even if revenue declines because customer acquisition or server utilization falls short of expectations. This could weaken operating cash flow and extend investment payback periods.
In a report released on November 5, SK Securities analyst Shin Yoon-jung said grid expansion costs and reduced tax benefits increase initial investment costs, while minimum electricity consumption commitments increase fixed-cost burdens even when actual utilization is low.
Shin noted in particular that if AI demand falls short of expectations, individual data centers could become less profitable because costs would be difficult to reduce quickly.
Uncertainty in the data center business is already affecting the contracting strategies of major technology companies.
Bloomberg reported on November 5, local time, that Microsoft was increasing oversight of some data center projects operated by neocloud companies including IREN, Nebius and Nscale, and was also reviewing whether to enter new contracts.
Microsoft also withdrew from efforts to lease an Nscale data center in the United Kingdom that faced grid connection delays and decided not to exercise an option that could have more than doubled its contracted capacity at Nscale's Texas data center.
IREN also pushed back its target date for achieving annual recurring revenue (ARR) of more than $4 billion, approximately 5.4276 trillion won, from the end of this year to January 2027 because delivery schedules for some facilities were delayed by Microsoft's requests for network design changes and adjustments to supplier schedules.
Original Korean article: economytribune.co.kr
English brief by Korea Wind Intelligence, koreawindintel.com
