The public utility grid has officially been designated a stranded asset by the architects of the artificial intelligence boom. Driven by the exponential baseload requirements of sovereign AI and large language models, enterprise hyperscalers are systematically seceding from municipal power infrastructure. Fermi America has filed its fiscal year 2025 earnings report detailing a highly lucrative operational roadmap toward 17 gigawatts of private power generation. Concurrently, Meta announced the funding of seven new natural gas power plants to support a 27 billion dollar Louisiana data centre complex, expanding the site's capacity to 7.46 gigawatts. This private generation capacity is sufficient to power the entire state of South Dakota, yet it will be consumed entirely behind the meter. The capital logic here is flawless. Public electricity grids are plagued by underinvestment, regulatory gridlock, and the operational friction of consumer politics. By taking natural gas generation directly behind the meter, tech firms are shielding their compute clusters from the volatility of civic infrastructure. Natural gas is the only fuel source capable of delivering the uninterrupted, gigawatt-scale baseload reliability required to train next-generation models. The pivot entirely bypasses the hollow sentimentality of corporate climate pledges, proving that when forced to choose between carbon emission targets and uninterrupted algorithmic yield, enterprise capital will ruthlessly select the latter. This corporate energy secession creates massive arbitrage opportunities in natural gas equipment manufacturing and turbine financing, as seen in TotalEnergies' recent 5.1 billion euro acquisition of flexible gas-fired assets. Investors must immediately price the legal and lobbying defence required as cash-starved municipal utility commissions inevitably attempt to tax these private microgrids.