Episode Summary
Executive Summary: Chase Lockmiller argues AI’s power demand is not a side issue but the central constraint and opportunity in the next computing cycle. Crusoe is building climate-aligned data centers and cloud infrastructure by siting compute near low-cost, clean, abundant energy, while using modular design, grid-aware placement, and new cooling/networking approaches to scale AI sustainably.
Main Topics: Crusoe’s energy-first origin story (Priority: 5/5): Lockmiller’s background in AI, trading, crypto, and infrastructure led him to focus on the cost and emissions of computation, culminating in Crusoe’s mission to use wasted or stranded energy resources. AI demand as a power and infrastructure problem (Priority: 5/5): He argues AI growth is already hitting real limits in power supply, interconnection, transformers, and data center vacancy, making electricity the main bottleneck rather than chips alone. Siting compute near clean, abundant energy (Priority: 5/5): Crusoe’s strategy is to place data centers in locations like West Texas, upstate New York, and Iceland where energy is cheap, clean, and underutilized, including behind-the-meter and greenfield generation partnerships. Cloud product and climate-aligned computing (Priority: 4/5): The company’s cloud offering gives customers high-performance virtual compute, storage, and networking for AI workloads, with the differentiator that the underlying infrastructure is powered by clean energy. Engineering challenges of scale and density (Priority: 4/5): Lockmiller discusses rapid increases in GPU power draw, liquid cooling, modular construction, cabling complexity, and the need to deploy remote facilities faster and more cost-effectively. AI as a catalyst for broader energy and climate innovation (Priority: 4/5): He contends AI will accelerate breakthroughs in fusion, battery chemistry, weather modeling, direct air capture, and carbon capture, turning high power use into a net societal opportunity. Optimism about grid transformation (Priority: 3/5): Rather than viewing demand growth as a reason to slow AI, Lockmiller sees it as a chance for the ICT sector to shape future generation, storage, and carbon-managed power systems.
Key Arguments: AI energy demand is likely underforecast, not overhyped; the load will be much larger than many current projections suggest. Power is the main constraint on AI scaling today, more than chips, because data center capacity and grid access are both tight. AI can be co-located with low-cost, clean, abundant power because many workloads are latency-tolerant. Building near curtailed renewables, hydro, geothermal, or new greenfield generation can turn stranded energy into productive compute. The environmental challenge is not to halt AI but to shape its energy supply so the technology advances with lower emissions. Higher compute demand can catalyze innovation in climate tech, including battery materials, fusion, weather modeling, and carbon capture. Modularity, off-site fabrication, liquid cooling, and improved cabling are essential to deploying next-generation AI facilities quickly. Rising GPU power density forces major redesigns of cooling, water systems, and electrical infrastructure. AI clusters act like synchronized, breathing loads, creating difficult power-management dynamics that data centers must engineer around.
Data Points: Projected electricity demand growth: double in the next 3 to 5 years - Referenced as the view of IEA and Goldman Sachs for overall electricity demand Generative AI share of data center computational demand: one third - Morgan Stanley projection for 2025 Annual utility capital investment: $5 to $10 billion - Projected regulated utility investment needed annually to meet new power demand Power from half an H100 per U.S. adult: 250 gigawatts - Lockmiller’s illustration of how much power mass-market AI use could require U.S. data center vacancy rate, end of 2023: sub 2% - Used to show near-total absorption of existing capacity Current U.S. data center vacancy rate: sub 1% - Indicates even tighter market conditions now Crusoe pipeline: about 12 gigawatts - Development or advanced commercial discussions for future capacity NVIDIA A100 power draw: about 300 watts per chip - Last generation chip used as baseline for rising density NVIDIA H100/H200 power draw: about 700 watts per chip - Current-generation Hopper chips NVIDIA GB200 power draw: about 1,200 watts per chip - Next-generation Blackwell chip power density Cluster cabling: over 1 million strands of fiber - Scale of networking complexity in a large cluster deployment
Pivotal Quotes: "Power is the main constraint right now." — Chase Lockmiller: On what is limiting AI scaling versus chip availability "I’m a firm believer in AI’s potential to completely transform the entire human experience, and with that, it will require tremendous amounts of power." — Chase Lockmiller: On why he views AI energy demand as inevitable and manageable "What we do is, you know, we partner with renewable energy producers... and we leverage the existing substation infrastructure and actually set up a grid connection as sort of a backup resource." — Chase Lockmiller: Describing Crusoe’s behind-the-meter and renewable-powered siting model
Implications: AI growth will keep pressure on grids, utilities, and data center supply chains, but it also creates a major market for clean generation, storage, and advanced cooling. Winners will be builders that can pair compute with abundant low-carbon power and move fast.