Doctorow: 'Sovereign AI' sells the wrong layer of the stack

Cory Doctorow, the longtime internet critic and former Electronic Frontier Foundation staffer who popularized "enshittification," called "sovereign AI" a misguided priority in a Pluralistic post published Wednesday. The Register ran an opinion piece amplifying the argument the same day under the headline "Sovereign AI is 'nonsense,' says Doctorow."

Doctorow's claim is narrower than the column's headline lets on. "Sovereign AI," the pitch for nationally controlled models, training data, and compute sold by governments and hyperscalers as digital independence, is, in his read, a label wrapped around the wrong layer of the stack. The lever for actual technological sovereignty, he argues, sits below the model: in the data a country can move, the applications it can replace, and the datacenters it can run on its own terms.

The test case Doctorow offers is blunt. Cutting off access to US-built AI chatbots would not measurably damage foreign economies, he writes. Disabling Microsoft 365 or bricking John Deere tractors, both real precedents he cites from prior Register coverage, would cause immediate, measurable harm. Sovereignty that cannot survive losing Office 365 is sovereignty in name only.

His alternative is operational rather than rhetorical. Build sovereign apps and datacenters that run everyday administrative, telecoms, and e-commerce software. Migrate ministries, companies, and households off US platforms. Let frontier models stay where they are; replicating them is the wrong fight.

He calls AI itself "the money-losingest technology in human history," one that "has consistently underperformed relative to its boosters' promises." That is a separate claim from the sovereignty argument, and a more contestable one. Hyperscaler capex disclosures and inference-revenue numbers tell a different story than Doctorow's claim. The Register opinion piece prints the line; the surrounding numbers do not.

Doctorow is responding to economist John Quiggin, whose July 16 Crooked Timber post argued that the Trump administration is systematically dismantling US diplomatic, military, financial, and technological sources of strength. Quiggin's own take on frontier AI runs closer to Doctorow's than to the vendor pitch: he predicts second movers such as DeepSeek and Mistral will run roughly a year behind US leaders at a fraction of the cost and without the datacenter footprint. That is a forecast rather than a measured outcome, but it is consistent with what The Register's systems desk has been documenting.

Tobias Mann's separate Register column on Chinese and open-weight model competitiveness lands in the same neighborhood. Cost-per-inference is no longer a moat for US labs, Mann writes. If Quiggin and Mann are right, the question for any "sovereign AI" procurement is not whether the national model will catch up. It is whether the country needs the model at all, or whether it needs the data, the apps, and the datacenter beneath it.

The Register's own editorial sign-off is worth separating from Doctorow's argument. There is a "clear and present danger" in US tech firms' refusal to resist political pressure, and Europe would be wise to invest in alternatives, the outlet writes. That is the Register's stance, not a Doctorow quote, and it covers a wider surface than the column's headline lets on.

The legitimate case for sovereign AI still stands and is not refuted here. Data residency, regulatory control over training data, reduced vendor lock-in, and domestic capability in critical sectors are real buyers' needs. Doctorow's argument is that none of those require owning a frontier model. They require owning the layer that survives a vendor's policy change. The falsifier is concrete: if a foreign economy genuinely collapsed the day a US AI lab went dark, Doctorow would be the one who is wrong. So far, no such test has been run.

The argument is the same point made operational in Doctorow's new book, "The Reverse Centaur's Guide to Life After AI," where AI is the "mindless horse's head" bolted onto a human body. Do not buy the head. Rebuild the body it pretends to lead.