Digital Sovereignty

Ten Groups Built a Free Chatbot in Seven Weeks

Current AI holds 400 million dollars in committed funding, seeded by the French government. Its open Alpha Chat took seven weeks to build. Here is what that does to your next AI quote.

Digital SovereigntyBy Servola Tech Desk2026-07-205 min read

AI-assisted, edited by humans. | Editor-in-Chief: Leon Soliman | Editorial standards

Ten Groups Built a Free Chatbot in Seven Weeks

Seven Weeks, Ten Organisations, One Chatbot

In early July 2026, in Geneva, ten organisations switched on a chatbot that none of them owns. Alpha Chat was assembled in seven weeks by a coalition that included Hugging Face, Mozilla and the MIT Media Lab, and it is open source.

The assembly method is the interesting part. Each contributor brought a component rather than a cheque: language models from one, safety tooling from another, computing power from a third. Nobody had to build the whole thing, so nobody had to fund the whole thing. That is how a general-purpose chat product went from nothing to running in under two months.

Read that timeline twice before your next procurement meeting. A volunteer coalition, working part-time across ten institutions, stood up the category of software that has been quoted to European buyers for the last three years as a strategic capability. It is not the best chatbot in the world. It does not need to be. It only needs to exist and be free.

A French Cheque Anchors the Balance Sheet

The money behind this is public and philanthropic, not venture capital, and that changes what it is for. Current AI reports 400 million dollars in total committed funding, roughly 340 million euros, of which the French government seeded 100 million dollars, about 85 million euros. The Ford Foundation, the MacArthur Foundation, DeepMind and Salesforce are also funders.

Chief executive Ayah Bdeir, who previously led Mozilla's AI strategy, put the distinction plainly: "They're not investors; they're funders." The nonprofit was founded in February 2025 by Martin Tisne, and its stated aim is free, open AI infrastructure rather than a product with a margin attached.

That matters for a European buyer in a way a cheaper commercial competitor would not. Money with no return expectation has no reason to raise the price later, and no reason to close the artefacts once adoption arrives. The usual pattern of a free tier that hardens into a licence does not have a mechanism here.

A Pocket Device That Speaks 22 Languages Offline

The second artefact is hardware, and it makes the point harder to dismiss. Suno Sutra is a pocket-sized device built with India's Bhashini government AI language division. It runs AI in 22 Indian languages with no internet connection, and it has been open-sourced.

Offline and multilingual are exactly the two requirements that vendors price as bespoke work. A device that does both, released openly, removes the argument that these are exotic capabilities available only through a large integration project. The same logic runs through the first grant cohort, 3.2 million dollars, roughly 2.7 million euros, split across four organisations.

Those four are Masakhane in Kenya, building AI datasets for more than 50 African languages, the Institute for Worldmaking in Lebanon, working on Arab cultural history databases, Portal sem Porteiras in Brazil, building offline AI tools for Indigenous communities, and the African Internet Rights Alliance in Kenya, building AI audit tools. Current AI is also partnered with Tokyo-based Sakana AI on shared open-source infrastructure.

What This Does Not Do

None of this replaces a frontier model, and any piece that implies otherwise is selling you something. Alpha Chat is a commodity chatbot built fast by a coalition. It is not tuned, supported, indemnified or hardened to the standard a regulated European business needs before it touches customer data.

Nor is it enterprise-ready in the ordinary sense. There is no vendor to escalate to at two in the morning, no contractual liability if an output causes damage, and no security team whose job it is to answer your questionnaire. Those things have real cost and someone still has to be paid for them.

So the correct reading is narrower and more useful than a replacement story. A free working stack puts a price floor of zero under commodity capability. It does not touch the frontier, and it does not touch the operational layer. It changes only what you are allowed to believe about the middle of the invoice.

Split the Quote at Your Next Renewal

The practical move is to break your AI invoice into two lines and price them separately. One line is the model itself, doing chat, translation or summarisation. The other line is integration, support, security review and liability. Until this month the vendor could bundle both and call the total a model price.

Now the model line has a public comparator that costs nothing. You are not obliged to deploy the open stack to use it. You only need to name it, put it in the negotiation, and require the vendor to justify what sits on top of it. The answer may well be sound, and integration and indemnity are genuinely worth money. The point is that you now know which part you are buying.

The jurisdictional dimension belongs in the same conversation. The anchor funder here is a European state rather than a US platform, and the artefacts are open, which means a stack you can host yourself carries a different exposure profile from one you rent. That is the same exposure our earlier reporting on EU-region hosting already flagged.

Frequently asked questions

Can we actually run Alpha Chat in production?

Technically it is open source and you can. Practically, treat it as a benchmark first. It has no vendor support, no contractual liability and no security assurance behind it, so most regulated European businesses will use it to price a quote rather than to serve customers.

Does this mean our current AI vendor is overcharging?

It means you cannot tell from a bundled quote. Ask for the model line and the integration line separately. If the integration, support and liability work justifies the total, the vendor will be able to show it line by line.

Who controls Current AI and could the terms change later?

It is a nonprofit founded in February 2025 by Martin Tisne, led by chief executive Ayah Bdeir, who previously led Mozilla's AI strategy. Its 400 million dollars comes from funders rather than investors, and the artefacts released so far are open-sourced, so there is no return expectation driving a later price rise.

A capability stops being strategic the moment ten volunteers can build it in seven weeks. Commodity chat crossed that line in Geneva this month, and the only owners who keep paying frontier prices for it will be the ones who never asked for the invoice to be split.

Current AIOpen sourceAlpha ChatDigital sovereigntyProcurement

Sovereign Cloud Now Comes With a Score

AWS launched its European Sovereign Cloud in January 2026 with a 7.8 billion euro build. In June the Commission explained its scoring rubric for cloud sovereignty. The label era is over; the grade era has begun.

3 min read

Infrastructure

Nvidia's Real Lock-In Just Went Open Source

Alibaba's T-Head opened its SAIL software stack to challenge Nvidia's CUDA lock-in. For a European operator the useful lever is not the chip but your own framework layer.

4 min read

Law & Regulation

The EA Buyout Now Hinges on Brussels

The largest take-private in history has cleared EA shareholders and US antitrust. The last real gate is the EU - and not classic merger review, but the Foreign Subsidies Regulation, facing its first blockbuster test.