The AI Power Shortage That's Reshaping Infrastructure Investment
17 September 2025 - A Weekly Publication by New North Ventures
AI data centers will consume 945 terawatt-hours by 2030—equivalent to Japan's entire electricity grid today. A single large AI training cluster now demands 500 megawatts of continuous power, roughly what a mid-sized city requires. These aren't projections anymore. They're procurement realities driving unprecedented infrastructure investments.
The AI power crisis reveals something profound about where generational returns emerge. Goldman Sachs projects 85-90 gigawatts of new nuclear capacity needed just for data centers—infrastructure spending that dwarfs traditional venture categories. Microsoft restarting Three Mile Island, Google committing to small modular reactors, Amazon investing across multiple nuclear partnerships—these moves signal a fundamental shift in how tech giants approach infrastructure constraints. Unlike previous scaling challenges solved through software optimization, AI's power appetite demands physical infrastructure at unprecedented scales. The opportunity extends far beyond nuclear itself. Power conditioning systems, specialized cooling technologies, grid interconnection hardware, and advanced materials all become critical bottlenecks. Early-stage companies in these categories achieve enterprise traction faster than traditional enterprise software because the alternative for customers is simply not scaling their AI capabilities. The urgency creates pricing power and shortens sales cycles in ways that software rarely achieves.
The infrastructure picks-and-shovels may prove more rewarding than chasing AI application layers. Companies solving power, cooling, and grid integration challenges for hyperscale deployments face tailwinds that software startups simply cannot replicate. Teams with deep domain expertise in power systems, nuclear engineering, or industrial automation represent increasingly compelling investment opportunities. The decade's best returns likely come from enabling AI infrastructure rather than building on top of it.
U.S. Army adopts “venture capital mindset”
The Army launched its "Fuze" program this week, marking Secretary Dan Driscoll's most ambitious attempt to inject venture capital thinking into military procurement. The initiative consolidates hundreds of millions in annual funding across existing programs like xTech competitions and small-business contracting, aiming to move technologies "from the garage to the front line" within 30-45 days of identification—a dramatic acceleration from the traditional 12-18 month contracting timeline.
The program represents a fundamental shift in military innovation strategy. Rather than the historical approach of identifying internal Army problems and then seeking solutions, Fuze flips the model to scout external technologies that could benefit military applications. Army labs and the 75th Innovation Command will handle vetting, while the program works alongside existing program executive offices rather than replacing them. The first live pitch competition occurs mid-October at the Association of the U.S. Army convention, focusing on electronic warfare, drone countermeasures, and energy solutions—priorities identified by the 25th Infantry Division and aligned with the Army's Transforming in Contact initiative.
The timing reflects broader frustration with defense acquisition bureaucracy that Driscoll describes as "calcified." Whether Fuze can overcome institutional inertia that has stymied previous reform efforts remains the central question, particularly as defense technology cycles continue accelerating faster than traditional government procurement can accommodate.
More links to explore:
HawkEye 360 continues expanding its RF detection constellation with nine new satellites across three clusters, demonstrating the scalability that made it an attractive investment. According to Military Aerospace, the Herndon-based signals intelligence company is upgrading Clusters 15 and 16 with real-time downlinking capabilities, enabling concurrent in-theater data collection—a significant operational advancement for maritime situational awareness and national security applications. The partnership with SFL Missions leverages the proven 30-kg DEFIANT platform while maintaining flexible production scaling options. This expansion reinforces HawkEye 360's position as the leading commercial RF geolocation provider, serving sectors from emergency response to spectrum mapping with increasing constellation density and enhanced real-time capabilities.
In this episode of Securing Our Future, host Jeremy Hitchock, explores the intersections of commercial innovation and national security. Join us as we engage in a thought-provoking discussion with Alex Lisle, CTO of Reality Defender. Alex shares insights from his 24-year career in cybersecurity, including his transition from a youthful hacker to a leading figure in the industry. We delve into the challenges posed by AI-generated deceptions, the role of cybersecurity, and the evolving landscape of synthetic content detection. This episode also touches on the regulatory and policy implications of AI advancements and the importance of maintaining digital authenticity in a rapidly changing world. Don't miss this in-depth conversation on the future of AI and cybersecurity.
Thanks for reading Securing Our Future! Subscribe for free to receive new posts and support my work.