Executive Summary
Epoch AI's data reveals the AI infrastructure buildout is accelerating faster than consensus believes, with Anthropic surprisingly leading the gigawatt race. Market Consensus: 'Energy bottlenecks will slow AI scaling.' Variant Perception: Energy constraints are pricing problems, not physics problems - companies are paying 2x for power but still scaling because GPU costs dwarf infrastructure costs. The guests' satellite imagery analysis shows Microsoft's Fairwater project (half of NYC's power consumption) and Anthropic's Amazon partnership (Indiana state capitol-level power) coming online by January 2026. While NVDA trades at reasonable valuation (PEG 0.68) with massive cash generation ($77B FCF), pure insider selling signals management confidence concerns. The real alpha lies in infrastructure 'picks and shovels' - power management (Eaton), cooling systems (Vertiv), and utility-scale power providers who benefit from 2x power pricing without the AI execution risk. This is a capacity constraint arbitrage, not a technology bet.
Key Insights
what David Owen and Yafah Edelman (Epoch AI) said“The most likely candidate to have the first gigawatt scale data center is Anthropic, which would not have been my pick, but Anthropic, Amazon's new Carlyle project rain year development seems on track to come online in January”
This is a preview. Log in to see the full analysis including investment opportunities, risks, catalysts, and detailed insights.