On 11 May 2026 the Estonian government approved a supplementary budget that carved out €10.98 million for Eesti.ai, the national sovereign-AI program launched in January, even as officials openly conceded the money would not pay for itself this year. Project lead Kirke Maar told ERR that 2026 is "still the launch phase, making it unrealistic to assess the initiative's impact on the overall economy in monetary terms by year-end." In most capitals, a flagship technology line item that volunteers "no immediate return" would be a political liability. In Tallinn it is closer to honest accounting — and the reason it works is sitting underneath the program, not inside it.
The infrastructure is the moat
Eesti.ai is not a from-scratch industrial policy. It is a software layer poured onto X-Road, the secure data-exchange backbone Estonia has run since 2001. According to e-estonia, the government's own digital-state portal, X-Road now connects roughly 52,000 organizations, carries about 2.2 billion transactions a year, and underpins more than 3,000 public e-services, saving the state an estimated 1,345 working years annually in eliminated paperwork.
That distinction matters enormously for how the spending should be judged. The headline Eesti.ai projects — eAmbulance AI to cut paramedics' documentation load, and new AI tooling for the PAI family-physician information system — are not greenfield builds. They are automation bolted onto registries that already interoperate, already authenticate, and already log every access. The €1.055 million routed to the Justice Ministry for legislative and public-sector automation, and the €2.022 million for healthcare AI via the Health Insurance Fund, buy models and workflows on top of plumbing that took two decades and prior strategies to build.
Steelman the skeptic
The strongest case against the allocation is real and deserves stating plainly. €10.98 million is public money spent in a year of fiscal consolidation — the same supplementary budget booked €29.3 million in cuts elsewhere and prioritized €17 million for eastern-border construction. Sovereign-AI programs have a poor global track record of converting launch enthusiasm into measured productivity, and "no return expected this year" can quietly become "no return measurable ever." A government that funds AI from money it is simultaneously cutting from other services owes taxpayers a hard metric, not a vision.
That critique is fair. But it argues for measurement discipline, not for withholding the spend — and Estonia's structure makes measurement unusually achievable.
Why a no-ROI year is defensible here
The pro-innovation case rests on three features that distinguish this from typical state AI splurges.
- Proportionality. €10.98 million is small. It is a fraction of the €85 million Estonia committed across 2024–2026 to its broader Data and AI white paper, which itself targets a 70% cut in administrative burden by 2030. This is iterative, bounded spending — not a moonshot demanding a multi-billion-euro vindication.
- A real demographic forcing function. Estonia's working-age population is shrinking and aging. Prime Minister Kristen Michal has framed the program bluntly: "If the population is falling and we want to keep taxes low, then wider use of technology is unavoidable." The stated goal — to double the value of Estonian work by 2035 and grow the economy roughly 25% within five years — is a productivity response to a structural labor shortage, the textbook proportionate use of automation.
- Built-in accountability. Because the AI runs on X-Road, every query against a health or legal registry is signed, logged, and auditable. Notably, €250,000 of the package goes to the Chancellor of Justice specifically for AI legal oversight. The infrastructure that makes the automation cheap is the same infrastructure that makes its misuse traceable.
The governance model worth watching
Eesti.ai is coordinated from the Government Office under an advisory board chaired by Bolt founder Markus Villig, with the explicit aim of training at least 100,000 people in practical AI use. That public-private structure — entrepreneurs steering, the state providing rails and oversight — is the open-internet model applied to government: build on open-source interoperable infrastructure, invite the private sector in, keep the audit trail public. Advisory member Kaspar Korjus put the choice candidly: "between doing this quietly and insecurely, or openly and safely."
That is the right frame. The risk of sovereign AI is not that governments automate too much; it is that they automate opaquely, on bespoke stacks no one can inspect, with no rollback. Estonia is doing the opposite — incremental automation on transparent, interoperable, logged infrastructure, with a named oversight office funded in the same budget line.
The verdict
A launch year with no measurable economic return is not a red flag when the spending is small, the demographic logic is sound, and the infrastructure makes both the upside and the oversight cheap. The honest test will come later: Estonia should publish concrete deltas — documentation hours saved in eAmbulance, queries handled by PAI tooling, administrative burden reduced — against its own 2030 targets. If it does, this is a model for proportionate state AI. If "no return this year" hardens into permanent unaccountability, the skeptics will have been right. For now, building AI on X-Road rather than around it is exactly the bet a digital state should be making.