Current AI raises $400 million to build open public AI infrastructure

Nonprofit Current AI is trying to build a public alternative to today’s private AI platforms, and it is moving quickly: the group has secured $400 million in committed funding, launched open-source tools, and begun seeding projects in Africa, the Middle East and Latin America. Its goal is to make AI infrastructure that is multilingual, locally controlled and available for public interest use, rather than owned and governed by a handful of Silicon Valley companies.

The effort matters because Current AI is positioning itself as a “World Wide Web of AI” — an open stack that communities, developers and governments can use without being locked into commercial platforms or English-first systems. The nonprofit has already funded new work in 22 Indian languages, across more than 50 African languages, and in Indigenous and culturally specific communities that have historically been underrepresented in AI training data.

Founded in February 2025 by Martin Tisne, Current AI is now being led by CEO Ayah Bdeir, who joined in January after overseeing Mozilla’s AI strategy and previously founded littleBits, the STEM hardware company that sold to Sphero in 2019. In a conversation with TechCrunch, Bdeir framed the organization’s mission as both technical and political: build AI as public infrastructure, keep communities in control of their data, and ensure that the benefits of the technology are not captured only by a few private firms.

What Current AI is trying to change

Current AI’s central argument is straightforward: if AI will shape education, work, culture and public services, there should be a public option. Right now, the largest systems are controlled by private companies such as OpenAI, Google and Anthropic, which means access, model behavior and data practices are all influenced by commercial incentives.

Bdeir said the nonprofit wants to create an AI ecosystem that resembles the early web: open, widely accessible and useful to everyone, not just paying customers or platform partners. That means supporting models, datasets, tools and governance structures that can be reused by governments, nonprofits and communities with different needs.

The organization also sees a major gap in how AI systems are trained and deployed. Most frontier models are built around English and a relatively narrow set of globally dominant data sources. For Current AI, that leaves vast parts of the world outside the center of AI development.

Why language is at the center of the project

Language is the first place Current AI sees the consequences of that imbalance. Bdeir said the world’s linguistic diversity is under threat, with many languages at risk of disappearing, and she argued that current AI systems are not designed to preserve that diversity.

According to the nonprofit, an AI system that cannot understand a community’s language cannot fully support that community’s knowledge, memory or identity. In practical terms, that can mean poorer access to health information, weaker educational tools and lost opportunities to preserve local history and oral tradition.

Bdeir said that when major AI companies broaden language coverage, they are often doing so to expand market reach rather than to serve communities with consent and context.

She also warned that for some Indigenous languages, outside actors have already used religious translations or other legacy documents as training data before communities had the chance to establish rules for how their knowledge should be used. Current AI says its projects are designed to reverse that pattern by making consent and control part of the process from the start.

How Current AI is funded and structured

Current AI describes itself as a public-private partnership, but one where the usual incentive structure is flipped. Instead of investors expecting equity and returns, governments, foundations and companies contribute as funders to support public-interest infrastructure.

The French government seeded the initiative with $100 million. Other backers include the Ford Foundation, MacArthur Foundation, DeepMind and Salesforce. Together, those commitments have brought the nonprofit’s total pledged funding to $400 million.

Bdeir emphasized that these contributors are not traditional venture investors. Their role is to support a shared infrastructure layer that can be used broadly, including in places where commercial AI products are scarce, expensive or culturally mismatched.

What the first grants are funding

Current AI’s first cohort grant round totaled $3.2 million, split among four organizations working across Kenya, Lebanon and the Brazilian Amazon. The funding is relatively small compared with the company-scale budgets associated with frontier AI, but the nonprofit argues that the point is not to compete on sheer size.

Instead, the grants are aimed at building local capabilities and tools that can survive outside a corporate platform.

Kenya: language datasets and accountability tools

One project involves Masakhane, a community-driven effort in Kenya building datasets for more than 50 African languages. The work targets practical uses in health, farming and education, where access to language-appropriate tools can affect daily life.

Another Kenyan group, the African Internet Rights Alliance, is developing audit tools to examine how AI systems behave across the continent. The idea is to give researchers and advocates a way to test systems for bias, gaps and harms rather than leaving evaluation entirely to vendors.

Lebanon: digitizing cultural history

In Lebanon, the Institute for Worldmaking is turning Arab cultural history and contemporary practice into machine-readable databases. Current AI says those databases are meant to remain under community control, rather than becoming raw material for outside companies.

That approach reflects one of the nonprofit’s broader priorities: treat cultural data as something to steward, not simply extract.

Brazil: offline tools for the Amazon

In the Brazilian Amazon, Portal sem Porteiras is building offline AI tools with Indigenous communities. Because many remote areas have limited connectivity, the project is designed to work without a reliable internet connection and to keep data within the territory.

That local-first model is one of Current AI’s defining ideas. The organization says models and datasets should be stored locally whenever possible, with community experts involved before work begins and consent mechanisms built in so communities can stop the process if needed.

How Suno Sutra works in India

Current AI’s most visible early deployment is Suno Sutra, a pocket-sized, offline device that can run AI in 22 Indian languages without an internet connection. It was built in partnership with Bhashini, the Indian government’s AI language division, following work unveiled at the India AI Summit in February.

The name means “listening chronicles” in Hindi, and the product is designed to serve users who may not speak English, have reliable connectivity or want to rely on cloud-based systems for basic access to information.

For Current AI, India illustrates the scale of the language gap. The country has hundreds of languages and dialects, and Bdeir argues that many of them are missing from mainstream AI because existing models were not built with that diversity in mind.

By releasing the device as open source, Current AI is hoping developers can adapt the platform for other contexts, communities and languages rather than treating it as a one-off pilot.

What Alpha Chat signals about the broader strategy

Last week in Geneva, Current AI launched Alpha Chat, an open-source chatbot assembled in just seven weeks by a coalition of ten organizations. The contributors included Hugging Face, Mozilla and MIT Media Lab, with each participant supplying a piece of the overall stack — including language models, safety tooling and compute resources.

The speed of the build is important because it shows the nonprofit is not only funding ideas; it is also trying to prove that open AI infrastructure can be assembled quickly through collaboration. In a sector dominated by vertically integrated companies, that is a meaningful statement.

Alpha Chat is also a test case for Current AI’s broader thesis: if the building blocks are shared, communities do not need to wait for a major vendor to decide whether their language, culture or use case is commercially important enough to support.

Who is Current AI partnering with next?

Current AI is also working with Sakana AI, a Tokyo-based startup that has promoted what it calls Sovereign AI. The two groups plan to build an open-source AI stack that supports Japanese language and culture while also serving communities in the Global South.

The partnership suggests Current AI wants to avoid a simple north-south model in which “public” AI only means Western institutions setting terms for everyone else. Instead, the nonprofit appears to be trying to build a network of regional collaborations that can share technical components while preserving local autonomy.

Bdeir said the model should not be judged only by whether it scales like a venture-backed startup. In her view, a successful outcome could be a tool built in one community helping another community preserve knowledge in its own language.

How big can $3.2 million really be?

On paper, $3.2 million split across four organizations is modest in a field where frontier labs spend far more on compute and talent. But Current AI is intentionally challenging the assumption that larger budgets are always better.

The nonprofit argues that the value of its work comes from what gets unlocked at the local level: datasets that do not exist today, tools that can function offline, and governance models that give communities veto power over how their information is used.

In that sense, Current AI is making a different kind of bet. It is investing in infrastructure that may not produce a headline-grabbing consumer product immediately, but could change who gets to participate in AI development over time.

• Open-source tools can be adapted by local developers.
• Offline systems can reach communities with weak connectivity.
• Community-controlled data can reduce extraction and misuse.
• Language coverage can support health, farming and education.

Why data ownership is the hardest question

Data governance may be the most difficult issue Current AI faces, because it sits at the intersection of technology, culture, law and power. Bdeir said there are many possible models for community ownership, but the one thing she rejected was letting a major tech company in Silicon Valley make those decisions for everyone else.

That stance reflects a growing global backlash against extractive AI development, where large datasets are collected and repurposed with limited transparency. For communities that have already seen their language, art or history used without consent, the question is not abstract.

Current AI’s response is to embed consent and local control into the development pipeline. That means involving community experts early, keeping data close to where it originates and making it possible to pause or stop projects that no longer align with community preferences.

What the project means for the AI race

Current AI is not trying to beat OpenAI or Google on model benchmarks, and it is not positioned as a direct commercial rival. Its challenge to the industry is more structural: it argues that the AI economy should include public-interest institutions that build shared infrastructure, preserve local languages and protect community rights.

If the nonprofit succeeds, its influence may be less visible than a consumer chatbot launch but more durable. It could shape the norms around multilingual AI, data stewardship and public funding for foundational technology.

For now, the organization is still in its early stages. Its grantees have not solved the hardest questions yet, and the nonprofit is openly admitting that. But it has already begun to define the problem differently: not just how to make AI smarter, but how to make it more accountable, more inclusive and less centralized.

That makes Current AI one of the more distinctive new entrants in the global AI landscape — not a startup chasing growth at all costs, but a nonprofit trying to build common digital infrastructure for communities the market has often ignored.

And if it succeeds, the result could be more than a chatbot or device. It could be a public foundation for AI that many languages and many communities can call their own.