Interview, Fireside Chat, Conference Presentation
Wes Cummins, Founder & CEO of Applied Digital: Ellendale’s Purpose Built AI Factory in North Dakota
- The company anticipates delivering large-scale gigawatt-style campuses primarily for AI training and inference, with specific facilities currently under construction or operational in North Dakota, South Dakota, and other US states.
- Projected capacity additions include 100 megawatts for the CoreWeave project in North Dakota by October and November of the current year, an additional 150 megawatts by June of next year, and a third building adding 150 megawatts to reach a total critical IT load of 400 megawatts for the campus.
- Facilities are designed to handle 530 megawatts of utility power, though customers typically operate at 70 to 80 percent of this capacity to maintain safety margins.
- Near-term AI workload capacity will be leveraged from stranded power, specifically utilizing wind generation in North Dakota, the sixth-largest wind-producing state, where current grid infrastructure cannot transport all generated energy.
- Most viable stranded power sites in the Dakotas and the US have been secured by competitors over the last five years, though the company expects to lower regional power costs by utilizing unused infrastructure and spreading grid costs across additional megawatts.
- Power sourcing strategies focus on existing assets, new renewable assets, and gas assets buildable within the next one to five years, while Small Modular Reactors (SMRs) are viewed as long-term solutions rather than immediate priorities.
- Battery storage is identified as a potential solution for grid balancing during periods of renewable excess or low demand, though it is not yet considered economically feasible for widespread deployment.
- The company predicts a significant demand for larger data centers with higher power density to support inference workloads across Europe and globally as AI use cases expand.
- Europe is projected to be one to three years behind the US in AI infrastructure development, with growth expected in inference facilities rather than massive training facilities.
- Noise mitigation plans involve locating facilities at least one mile from populated areas to ensure high-frequency noise attenuates, distinguishing these sites from high-complaint regions like Northern Virginia.
- The company intends to continue development on US heartland projects extending to the Gulf and explore opportunities in Europe, maintaining a primary focus on the US market.
- Robust fiber connectivity with two to three diverse bandwidth paths running in multiple cardinal directions is required for the CoreWeave deal and similar projects.
- Operations have transitioned away from retrofitting Bitcoin mining facilities for AI use since 2022 due to divergent requirements for high-performance computing.
- AI workloads demand firm, continuous 24-7, 365 power availability, contrasting with the interruptible power acceptable for Bitcoin mining.
- Customers require "five nines" of redundancy and the ability to sustain 100 percent capacity on the hottest days of the year, with facilities managed like semiconductor clean rooms to protect high-value GPUs.
- The company believes that correct power utilization will reduce costs for the entire region rather than driving prices up, a sentiment supported by the expectation that the AI gold rush will make innovative clean energy ideas profitable.