Current AI Raises $400M to Build Public AI Infrastructure

A nonprofit is trying to do for artificial intelligence what the early web did for information: make it a shared public utility rather than a walled garden. In a report published July 19, 2026, Current AI laid out its plan to build open, public AI infrastructure — models, data, and tools that are free for anyone to use and build on — backed by $400 million in committed funding and a mission its leaders compare to the founding of the World Wide Web.

The pitch is a direct counterweight to a market in which the most capable systems are controlled by a handful of well-capitalized labs. Current AI’s argument is that if the technology is as transformative as its builders claim, there needs to be a version of it that no single company owns.

Who is behind it

Current AI was founded in February 2025 by Martin Tisne and formally launched at France’s AI Action Summit in Paris, where governments and philanthropies pledged initial support. The organization operates as a public-private partnership, pooling money from states, companies, and foundations to fund what it calls “public interest” AI.

Its chief executive is Ayah Bdeir, who joined in January after leading AI strategy at Mozilla and founding the electronics company littleBits. Bdeir has framed the effort in civic terms. “If AI is truly a transformative technology, if it’s going to change every aspect of everyone’s life, there has to be a public alternative,” she said.

The money

The $400 million in commitments comes from an unusually mixed roster of backers. The French government seeded the effort with $100 million, and it has been joined by the Ford Foundation, the MacArthur Foundation, Google’s DeepMind, and Salesforce. That blend — a national government, two large US philanthropies, and two technology companies — is the model Current AI is betting on: no single funder large enough to dictate direction, and enough combined capital to build things at scale.

The figure is modest next to the roughly $700 billion in capital spending the largest hyperscalers have guided toward for 2026. Current AI is not trying to match that. Its wager is that a small amount of patient, non-commercial money aimed at the gaps the market ignores can produce infrastructure the private sector has no incentive to build.

What it is actually building

The clearest illustration of the strategy is a project called Suno Sutra — Hindi for “listening chronicles” — developed with Bhashini, the Indian government’s AI language division, after the two teamed up at the India AI Summit in February. Suno Sutra is a pocket-sized, offline device that runs AI in 22 Indian languages with no internet connection required.

The device pairs a camera, screen, microphone, and speaker with three models running at once: a vision model, Bhashini’s automatic speech recognition and machine translation system, and a text-to-speech model. It is open-sourced, so developer communities can adapt it for their own languages and use cases. The design choices matter as much as the technology. By running locally on cheap hardware, it sidesteps the cloud dependency, connectivity requirements, and per-token costs that put frontier chatbots out of reach for much of the world.

That focus on running capable systems on constrained hardware is exactly the territory where small language models have been closing ground on their larger cousins, and where open-weight models are narrowing the gap with the proprietary frontier.

Suno Sutra is one piece of a broader grant program. Current AI allocated $3.2 million in grants last month across four organizations, and last week launched an open-source AI chatbot at the AI for Good summit in Geneva.

The “World Wide Web of AI”

The organizing metaphor is deliberate. When Tim Berners-Lee released the web’s protocols without patent or license fee, he created a common layer that anyone could build on, and an entire economy grew on top of it. Current AI wants to establish the equivalent for AI: open datasets, especially for underrepresented languages and communities; open models that are genuinely inspectable and modifiable rather than merely “open” in name; and shared compute and tooling so that building useful systems does not require a hyperscaler’s balance sheet.

The emphasis on languages is strategic. Today’s leading large language models are trained overwhelmingly on English and a handful of other high-resource languages, leaving hundreds of millions of speakers with tools that barely understand them. Underserved languages are both a genuine public need and a domain where commercial labs see little near-term return — precisely the kind of gap Current AI was designed to fill.

Where it fits in the policy landscape

Current AI arrives at a moment when governments are actively contesting who sets the terms for AI. The effort’s public-interest, multi-government framing sits alongside a crowded field of governance initiatives: the United Nations’ new global AI governance dialogue, the European Union’s enforcement of the AI Act’s rules for general-purpose models, and China’s newly formalized WAICO cooperation organization. What separates Current AI from those is that it is not a rulemaking body. It is trying to build the thing itself — to make openness a fact on the ground rather than a policy aspiration.

What it means

Current AI is a test of whether public-interest infrastructure can survive in a field defined by extreme capital intensity. The economics are stark: $400 million is roughly what a single frontier training run and its associated compute can cost, and the labs Current AI hopes to counterbalance are spending orders of magnitude more every quarter. It cannot win by matching that spend, so it is choosing its battles — offline, multilingual, low-cost systems for communities the market underserves.

The likely near-term winners are those communities, and the developers who get open building blocks they could not otherwise afford. Governments gain a hedge: a way to reduce dependence on a small number of foreign AI vendors without standing up national labs from scratch. The open-source ecosystem gains a well-funded, non-commercial anchor tenant.

The risks are equally clear. Public-private consortia can move slowly, and coordinating a national government, two foundations, and two tech companies around a single technical roadmap is hard. Funding measured in the hundreds of millions has to be spent with unusual discipline to matter against private budgets in the hundreds of billions. And “open” is contested terrain — the difference between genuinely inspectable systems and models that are open in branding only will determine whether the “World Wide Web of AI” becomes a real common layer or a well-meaning demo.

What to watch next is execution: whether Suno Sutra ships beyond a prototype, whether the grant portfolio produces reusable infrastructure rather than one-off projects, and whether more governments write checks. The web analogy is powerful precisely because it worked once. Whether it can be repeated — deliberately, and on a compressed timeline — is the question Current AI has $400 million to answer.