Forecasting

More Questions: Forecasting Data Centers

One afternoon, while talking with a utility executive, they mentioned a complaint about data centers. For the small utility, the proposed new data center will add roughly 10% to their system peak. After venting (for a bit) about risk, the executive asked, ‘How do you forecast them?’ and ‘What is everyone else doing?’ 

Since 2023, Itron’s Annual Energy Survey has asked utilities whether they are seeing data centers show up in their service territories. In 2023, 46% of respondents said ‘Yes, we are seeing them.’ In 2024, 55% said ‘Yes.’ With the pervasive inquiries about new data centers arising across North America, I’m getting bombarded with questions about the best forecasting practices.  

During Itron’s roundtable discussion at the 2024 Annual Energy Forecasting Meeting on this very topic, utilities provided a wide range of anecdotal information about best practices. While it is easy to draw upon these stories for best practices, I decided to begin quantifying what companies are actually doing.  In November 2024, I conducted a mini survey asking companies how they are forecasting data center additions. The survey was sent to 32 companies, all who responded ‘Yes’ in the 2024 survey, and it yielded 21 responses. The survey asked three questions: 

  1. How do you develop your data center forecast? 
  2. How do you incorporate the data center forecast into your long-term forecast? 
  3. How do you determine the likelihood (i.e., probability) of data center connections? 

Here’s a preview of the results. 

Question 1: How do you develop your data center forecast? 

The predominant response (86%) was that corporate account representatives provide information about potential projects. The remaining responses included additional methods paired with account representative information.  These methods include purchasing an external data center forecast and developing a model. 

Question 2: How do you incorporate the data center forecast into your long-term forecast? 

Just because a company has a data center forecast does not mean that they incorporate the entire forecast into the long-term forecast. Only 26% of respondents include 100% of the data center forecast into the long-term forecast. The remaining respondents discount that forecast based on the likelihood of connection. 53% of respondents discount the data center forecast based on a likelihood threshold. For instance, a company might only add data center projects that are more than 70% likely to connect. If the project does not exceed 70%, then the project is not added to the long-term forecast. 21% of respondents use a weighted average approach. This approach multiplies the project size with their likelihood to create an ‘expected value’ of data center additions. 

Question 3: How do you determine the likelihood (i.e., probability) of data center connections? 

The responses from Question 2 raised the question of how likelihood is determined. In this question, 68% confess that professional judgment is the primary factor in determining likelihood. 26% use a rubric that guides their judgement. This rubric might incorporate project development or contractual milestones. Only one response (5%) uses scenarios to manage the data center additions. 

What I told the utility executive is what everyone suspects: data center additions are primarily driven by expert judgement. Forecasters are supporting their judgement with mathematical probabilities (because that’s what forecasters do), but professional judgement still underlies these probabilities. 

I realize that professional judgement is not necessarily the answer we want to hear in the age of data science where we (rightly or wrongly) place faith in mathematical models that can tell us the future. More work needs to be done. So, I’m willing to concede that the survey shows us ‘common practices’ and not necessarily ‘best practices.’ But please be assured that we will continue to discuss practices at our annual conference next year in Savannah, Georgia, from April 8-11. Join the conversation by registering today, and be sure to answer the annual survey when it arrives in your mailbox next year to see how our practices evolve.  

By Mark Quan


Principal Forecast Consultant


Mark Quan is a Principal Forecast Consultant with Itron’s Forecasting Division. Since joining Itron in 1997, Quan has specialized in both short-term and long-term energy forecasting solutions as well as load research projects. Quan has developed and implemented several automated forecasting systems to predict next day system demand, load profiles, and retail consumption for companies throughout the United States and Canada. Short-term forecasting solutions include systems for the Midwest Independent System Operator (MISO) and the California Independent System Operator (CAISO). Long-term forecasting solutions include developing and supporting the long-term forecasts of sales and customers for clients such as Dairyland Power and Omaha Public Power District. These forecasts include end-use information and demand-side management impacts in an econometric framework. Finally, Quan has been involved in implementing Load Research systems such as at Snohomish PUD. Prior to joining Itron, Quan worked in the gas, electric, and corporate functions at Pacific Gas and Electric Company (PG&E), where he was involved in industry restructuring, electric planning, and natural gas planning. Quan received an M.S. in Operations Research from Stanford University and a B.S. in Applied Mathematics from the University of California at Los Angeles.


Region Selector Select a region and country for the best experience.