AI, Data Centers and Electric Energy

AI driven electric demand is forcing utilities to rethink their approach to connecting new data center customers to the grid. Until quite recently, many utilities had generating capacity to sell and competed for new customers. That is changing. In regions that are attractive to data centers, like the Southeastern United States, data center developers are requesting capacity commitments that dwarf the amount of generation that utilities had planned to build before the AI explosion --they typically targeted less than 1% annual load growth. A single new hyperscale data center can represent ten years of load growth at that rate.  Multiple data centers compound the problem.

Major generation and transmission projects required six to eight years from permitting to commercial operation, raising the question of how to support systems reliably until new electric infrastructure is online.  New generating plants are built at current construction prices and have little accumulated depreciation to reduce their carrying cost. As a result, they can put pressure on rates for all customers, especially in states where electric utilities remain vertically integrated. 

Utilities are scrambling to winnow out real data center projects from speculative requests and are rationing capacity for customers seeking commitments in amounts that exceed specific megawatt (MW) thresholds. 

The winnowing out process can involve utilities requiring large, upfront deposits for the cost of the new substations and transmission upgrades needed to serve data centers. In many cases, these costs will eventually be covered (and refunded) from the revenue from electric rates charged to the data center.  But the requirement to pay these costs upfront is something of a new twist. 

Utilities may also winnow out projects by requiring binding rate and payment commitments early in the development process. This can mean that only projects with a committed end user can make the cut.

For even the most solid development project, utilities also may ration capacity by delaying firm supply commitments until new generation can be built. For vertically integrated utilities in particular, the generation planning cycle is governed by integrated resource plans or “IRPs,” which are often filed every three years with the state regulatory commission and are updated annually. Getting the load from a new data center included in the utility's IRP is an important step but by itself it may not unlock a capacity commitment from the utility. An IRP is not a guarantee that the necessary generating plants will come online when needed.  So, some utilities are now requiring that data center customers agree to take only interruptible service until the necessary new generation is online. 

So how should data center developers respond?  

1. Get in line early.  Begin the dialogue with the utility that serves the area as soon as possible and do all you can to get the utility to include your project's demand in its generation planning process.
2. Line up end-users as quickly as possible.  As utilities seek to push cost and schedule risks down to customers, it is important to have an end user willing to take those risks.
3. If interruptible service is all that is offered, try to limit exposure to interruption contractually and explore options for standby generation or battery storage.
4. Use local governments as allies.  They will benefit most from the tax base and economic spinoff from a data center.  They can be valuable allies.
5. Follow regulatory and legislative developments in your project's area.  These are novel issues and state regulatory commissions and lawmakers are only now beginning to weigh in on what the rules should be going forward.

As always, informed counsel can be of help on these issues.