The nationalization argument is really a vendor-continuity memo

Bruce Schneier and Nathan E. Sanders have argued, in an essay first published in The Guardian and reposted to Schneier's blog in its August 2026 archive, that if OpenAI and Anthropic fail in the public markets the United States should nationalize them and convert them into national labs operated under democratic control ^[19][20][9]. The policy is the headline; the premise is the part that belongs in a risk register, because the authors treat the June IPO filings as the point at which founder intent stopped being a governance mechanism at either company ^[1][4].

Their account of the origin story is unsentimental. Both labs were founded by developers who feared that unrestrained corporate development, at Google and Meta specifically, would steer the technology toward deleterious or catastrophically unsafe outcomes, and both sets of founders claimed their organizations could uniquely be trusted to act in humanity's interest ^[1][2]. According to Schneier and Sanders, each was then co-opted by the same market incentives and became a corporate behemoth guarding future investor value rather than the public interest ^[3]. The filings in June, met with buzz about trillion-dollar valuations, are where that becomes explicit rather than arguable ^[4].

The authors are dismissive of the softer remedies. Proposals to have the federal government seize a share of the stock for a sovereign wealth fund, or to redistribute revenues as a taxpayer dividend, are noted and then set aside in favor of public ownership and operation ^[5][18]. The essay text as supplied stops mid-sentence at that point, so the mechanism is asserted rather than designed; the same passage floats returning OpenAI to the nonprofit roots it fought to shed, which Anthropic's founders spurned, or reorganizing both as university research centers ^[21][17].

The financial case matters more to buyers than the constitutional one. Frontier models are expensive to train and depreciate within months once a newer model appears, which narrows the payback window; enterprise clients are getting smarter about minimizing token usage; the best models behave similarly enough to be commodities, which depresses prices; and open-source and Chinese competitors, lagging only a few months in capability, give away what the labs sell ^{[10][11][12][13]}. Many of those free models run locally, the large ones on private clouds and high-end servers and the small ones on a laptop or a phone, which is what puts the datacenter capital plans in question ^[14]. Meanwhile the news is datacenter backlash and a slumping Nvidia, and SpaceX's stock tanked weeks after its own IPO ^[6][7]. The authors' conclusion is conditional and blunt: if the market decides these firms cannot produce a growing return, they collapse ^[16][8].

That is the read to carry into procurement. The tail risk on a critical model dependency, on this argument, is not a price increase but a change of control, and every outcome the essay entertains, nationalization, nonprofit conversion, or academic absorption, is a change of control ^[9][17]. Schneier and Sanders also note staggering ongoing usage growth, which is another way of saying the internal switching surface is already wide ^[15].

Watch three things: whether enterprise token spend actually contracts as claimed ^[11]; whether your evaluations have ever been run against locally hosted open-weight models, since that is the substitution the commodity thesis implies ^[12][14]; and what your contracts say about assignment, change of control, and notice on model deprecation, given a stated depreciation cycle measured in months ^[10].