THE SITING WALL : On the Datacenter Moratorium Movement, the Infrastructure Constraint the S-1 Names as Material, and the Architecture That Has No Siting Problem

On October 26, 2026, Anthropic filed a Form S-1 with the U.S. Securities and ExchangeCommission. The filing discloses existential risk at length. It also discloses somethingnearer, and the contrast between the two is the subject of this paper. Among the risk factors:"Opposition to data center development has also become more organized and coordinatedacross jurisdictions and our public profile may draw particular attention to projectsassociated with us." And: "[We are vulnerable to] regulatory changes, moratoriums on the facilities and power failures." The filing places this in the same risk section as bioweapons and loss of control over AI. Areader who expects the nearer risk to be the smaller one will be surprised: the S-1 identifiesthe anti-datacenter movement as the more immediate threat to the business. This paper examines that constraint. It has three parts. First, the arithmetic. The filing discloses a datacenter program measured in hundreds ofbillions against a physical supply chain measured in transformer lead times of 128–144 weeks and interconnection queues measured in years. The constraint is not capital. Capital is abundant. The constraint is siting — land, power, water, and permission. Second, the political economy. The opposition is not partisan and not anti-AI. It is proximity.A survey reported in late October 2026 found 63% of registered voters opposed to adatacenter within ten miles of their home, and 13% in support. Voters who broadly support AIdevelopment oppose the infrastructure it requires in their own communities. That split is not a contradiction. It is the standard structure of concentrated cost and diffuse benefit, and it is the most durable form of American local opposition that exists. Third, the alternative. An architecture exists that has no siting problem. Not a smallerdatacenter — no datacenter. A deterministic inference architecture that runs atmicrowatt-to-milliwatt envelopes on commodity microcontrollers, draws no water, requires no grid interconnection, and can be deployed on a factory floor without a public hearing. The paper does not argue that this architecture replaces the frontier stack. It argues something narrower: the class of deployment that can be certified is also the class that can be built without a fight. The filing names the fight. The filing does not name the class.