Top VC: 'Power, Not Silicon' is AI's Future
Investment prodigy Leopold Aschenbrenner has reportedly dumped his chip stocks—Nvidia, TSMC, Broadcom—to bet on energy and infrastructure. He's moved over $1.6B into companies like Bloom Energy and CoreWeave, arguing the AI bottleneck is shifting from GPUs to power. His take: "The next moat is power, not silicon."
Leopold Aschenbrenner, a former researcher on OpenAI's "Superalignment" team, is the author of "Situational Awareness," an influential 165-page essay predicting Artificial General Intelligence (AGI) could be developed by 2027. His investment firm, also named Situational Awareness, is backed by tech luminaries including Stripe founders Patrick and John Collison and former GitHub CEO Nat Friedman. The shift in focus to power is underpinned by stark projections. The International Energy Agency estimates global data center electricity consumption could more than double between 2024 and 2030, reaching 945 terawatt-hours—an amount comparable to the current electricity demand of Japan. In the U.S. alone, data centers could account for over 7% of the nation's electricity use by 2028. Aschenbrenner's bet on CoreWeave is a direct investment in specialized AI infrastructure. Unlike general-purpose cloud providers, CoreWeave operates a GPU-centric platform purpose-built for AI training and inference, essentially acting as a "rent-a-GPU" utility with a usage-based pricing model. The company has secured long-term, multi-year contracts for 96% of its revenue, indicating sustained demand for its high-performance compute. The investment in Bloom Energy targets the power delivery bottleneck directly. Based in San Jose, Bloom Energy manufactures solid oxide fuel cells that generate electricity on-site through a non-combustion chemical process, using fuels like natural gas or hydrogen. This decentralized approach bypasses the strained electrical grid, where connection wait times in key regions like Virginia can stretch from four to seven years. This power crunch is already creating significant delays for data center projects. Beyond long interconnection queues, shortages of critical grid components like large electric transformers are driving up prices and pushing project timelines out by years. Aschenbrenner's thesis is that the ability to secure power will become a more significant competitive advantage than securing capital or even chips. In his "Situational Awareness" paper, Aschenbrenner outlines a future requiring "trillion-dollar clusters" of GPUs for advanced AI development. He forecasts that by 2030, these clusters could demand 100 gigawatts of power, intensifying the race to secure energy resources and making companies that provide power infrastructure central to AI's continued scaling.
