Seventeen months ago, at the Paris AI Action Summit, France announced the most ambitious public-interest AI vehicle ever attempted: Current AI, seeded with roughly $100 million from the French government and launched with over $400 million in total commitments from a coalition spanning the Ford and MacArthur foundations, the Omidyar-backed AI Collaborative, Google DeepMind, and Salesforce — with a stated goal of mobilizing $2.5 billion over five years and ten countries signing a Paris Charter on AI in the Public Interest.
This week it’s back in the news: a TechCrunch profile (July 19) of CEO Ayah Bdeir — the littleBits founder who led Mozilla’s AI strategy before taking the job in January — laying out the vision in its cleanest form yet: a “public option” for AI, open and free, modeled on the early World Wide Web. Her framing of the backers is the quote that matters: they’re not investors, they’re funders.
So this is the Reality Check the project has earned by surviving to month seventeen: is state-and-philanthropy money building sovereignty infrastructure Europe will actually control — or funding the world’s best-intentioned conference circuit? The honest answer requires taking both cases at full strength, because this one is genuinely undecided — and the piece ends with a test, not a verdict.
A public option for AI:
infrastructure or theater?
Current AI: ~$100M French seed, $400M+ committed, ten Paris Charter countries, a $2.5B five-year target — and, seventeen months in, $3.2M actually granted. Both steelmen at full strength; verdict deferred to a dated test.
Three verbs, three very different numbers
Bars to scale against the $2.5B target. The disbursement curve is the test of a funding vehicle — and every verb above is doing different work. (Fair note: the org’s own first six months were an explicit governance start-up phase; commitments were never claimed as disbursements.)
What has actually shipped
Funder list worth naming: the public alternative to Big Tech is part-funded by Google DeepMind and Salesforce — a governance question answerable only in artifacts, not charters.
Two European routes, same clock
Public route · Current AI
- ~$100M state seed → $400M+ committed → $3.2M granted in 17 months
- Output: governance framework, two open artifacts, ten charter signatures
- Ownership: everyone. Suno Sutra belongs to the commons.
Private route · Prior Labs
- €9M pre-seed → Nature paper + SOTA model in 18 months → €1B+ committed by SAP, closed in ten weeks
- Output: a frontier lab, shipping
- Ownership: SAP’s shareholders. Velocity’s price.
The velocity comparison isn’t as one-sided as it looks: for a public option, “who owns the result” is the metric — and only one route answers “everyone.”
- Disbursement: cumulative grants ≥ ~25% of the $400M, and a real second government tranche toward the $2.5B.
- Adoption: one load-bearing artifact — a dataset in production model cards, devices at population scale, a tool with a living developer community.
- Independence: at least one funded thing its corporate funders would prefer it hadn’t. The only observable proof a public option is public.
Pass two of three: the strongest answer yet to how Europe funds AI it controls. Fail two of three: €100M tuition for the lesson Prior Labs taught for €9M.
As an affiliate, we earn on qualifying purchases.
What has actually shipped
Hold the thesis to its outputs. In seventeen months, per this week’s coverage and the organization’s own communications:
One grant round: $3.2 million across four organizations (June 2026). Against $400M+ in commitments, that is a disbursement rate of under one percent.
Suno Sutra (February 2026, with Bhashini, the Indian government’s AI language division): a pocket-sized, offline device running AI in 22 Indian languages, no internet required, fully open-sourced for developer communities to build on.
Alpha Chat (July 2026, AI for Good Summit, Geneva): an open-source chatbot.
And, per the organization’s own May 2025 update, the first six months were explicitly a “start-up phase” — governance, strategy, legal, operational basics.
Two readings of the same ledger, both accurate. Reading one: a nonprofit that spent a year on governance and has granted 0.8% of its headline number is exhibiting the exact failure mode of public technology initiatives — the money exists as press release, the institution as process. Reading two: commitments were never claimed to be disbursements, standing up a multi-government funding vehicle legally and operationally in a year is roughly on schedule for the species, and the early artifacts are precisely chosen.
AI infrastructure development tools
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
The steelman for the public route
Three arguments deserve full weight.
The diagnosis is correct. Every frontier system belongs to a private company. If AI mediates public life the way the web does, the absence of any public-interest layer is a structural problem, not an aesthetic one — the same logic that gave the web itself its public, royalty-free foundations. You don’t have to like the execution to concede the diagnosis.
The data thesis is smart and cheap. Founder Martin Tisné’s original framing was that the bottleneck for public-interest AI is data, not compute — high-value datasets (health, government, low-resource languages) that markets can’t or won’t assemble, unlocked in privacy-preserving form. That is a lever where $400M is real money, whereas in the compute race it’s a rounding error. A fund that knows which race it can afford is rarer than it should be.
Suno Sutra is exactly what markets don’t build. An offline, open-source, 22-language device for communities the commercial models ignore — no subscription, no cloud dependency, no data leaving the device. Readers of this site will recognize the architecture: it is local-first AI as public infrastructure, and it deserves credit on the merits, independent of who funded it. Bdeir’s stated counter-paradigm — that scale is the Big Tech measure, not the only measure — is at minimum a coherent alternative theory of value.
As an affiliate, we earn on qualifying purchases.
The steelman for the skeptics
Three arguments, equally full weight.
Commitments are the softest currency in technology policy. $400M “committed,” $2.5B “mobilizing,” ten countries “signed” — and $3.2M granted. Every one of those verbs is doing different work, and public-interest initiatives have a long history of living in the gap between them. The test of a funding vehicle is its disbursement curve, and this one is seventeen months in with a curve that barely lifts off the axis.
The funder list contains its own contradiction. The public alternative to Big Tech’s AI is funded, in part, by Google DeepMind and Salesforce — with Google and Salesforce named core partners at launch and Hugging Face, Instacart, and Sakana as supporters. There are honorable readings (hedge, conscience, talent pipeline) and structural ones: an institution can’t be the counterweight to interests represented on its funding roster. This isn’t an accusation; it’s a governance question the organization will have to answer in artifacts, not charters.
The comparison Wednesday’s Signal makes unavoidable. This same month, SAP closed its acquisition of Prior Labs — a Freiburg lab that turned €9 million of private pre-seed into a Nature paper and a state-of-the-art model inside 18 months, then attracted €1 billion+ in committed private scale-up capital, deal signed to closed in ten weeks. Two European routes to AI capability, running in parallel, on the same clock. One produced a frontier lab; the other produced a governance framework, two artifacts, and $3.2M of grants. The private route’s advantage isn’t ideology — it’s that its money is spent by people who lose it if nothing ships.
The counter-counterpoint, in fairness: Prior Labs’ output now belongs to SAP’s shareholders. Suno Sutra belongs to everyone. If “who owns the result” is the metric — and for a public option, it is — the comparison isn’t as one-sided as the velocity numbers suggest.
As an affiliate, we earn on qualifying purchases.
The test, written down now
Per this site’s house habit for genuinely undecided questions: no verdict, but a falsifiable test, dated. Revisit in July 2028:
Disbursement: has cumulative granting reached a meaningful fraction — say, a quarter — of the original $400M, with the $2.5B mobilization showing a real second tranche from a second government? Commitments that never convert are the theater diagnosis confirmed.
Adoption: is any Current AI-funded artifact load-bearing — a dataset cited in production model cards, Suno Sutra-class devices deployed at population scale, an open tool with a real developer community? One genuine hit justifies the model; a portfolio of pilots does not.
Independence: has the organization funded anything its corporate funders would prefer it hadn’t? That is the only observable proof a “public option” is public.
Pass two of three, and Current AI becomes the strongest answer yet to the question this site keeps asking — how Europe funds AI it actually controls. Fail two of three, and the €100 million was tuition for the lesson the Prior Labs route taught for €9 million.
Sources
- TechCrunch, “Nonprofit Current AI is racing to build the World Wide Web of AI, free for all” (July 19, 2026) — Bdeir interview; $100M French seed; Ford, MacArthur, DeepMind, Salesforce; $400M total; $3.2M/four-organization grant round; Suno Sutra and Bhashini partnership; Alpha Chat launch; founded February 2025 by Martin Tisné; “funders not investors”
- The Next Web, “Current AI is building a public alternative to Big Tech AI” (July 20, 2026) — €100M seed figure; $2.5B five-year mobilization target; public-option and early-web framing
- Current AI, “About” and “Building Public Interest AI — Current AI’s next chapter” (May 2025) — $400M+ committed, $2.5B target, ten Paris Charter signatories; self-described six-month governance start-up phase
- Fortune (via Yahoo, February 2025) — launch structure at the AI Action Summit: Omidyar/AI Collaborative role, Google and Salesforce as core partners, Hugging Face/Instacart/Sakana as supporters; Tisné’s data-not-compute bottleneck thesis and special-envoy role
- Philanthropy News Digest (February 12, 2025) — launch announcement; Omidyar Group, Ford, MacArthur, McGovern foundations
- AI for Good / ITU speaker profile (2026) — Bdeir biography, Mozilla AI strategy, littleBits/Sphero
- ThorstenMeyerAI.com, “SAP’s €1 Billion Bet Is on Tables, Not Chatbots” (July 22, 2026) — the parallel-route comparison
Seed figure reported as both $100M and €100M across sources; treated as approximately equivalent and noted. All commitment and mobilization figures are commitments and targets, not disbursements — the distinction is the article’s subject.
