$400 Million Bet Pits Current AI Against Big Tech's Grip

Current AI has $400 million committed to build a public-interest AI stack that its CEO compares to the open web, and its first real test is serving people whose languages barely register in today’s dominant models.

That matters most for communities outside English-first AI. A farmer in rural India photographing a dying plant shouldn’t need English or broadband to ask for help. That is the problem Current AI says it wants to solve, according to TechCrunch.

The nonprofit’s pitch is simple: AI should have a public option. Not every model, dataset, chatbot, or device should sit behind corporate platforms designed around the largest markets.

“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,” Current AI CEO Ayah Bdeir told TechCrunch. “Like the World Wide Web, available to anyone, for free.”

The hard part is everything after that sentence. Free access raises expensive questions about compute, AI safety and governance, localization, consent, AI bias claims, and quality control. Current AI is moving fast, but it still has to prove that public-interest AI can work outside demos, summits, and grant announcements.

Current AI’s public option targets users Big AI misses

Current AI is a nonprofit founded in February 2025 by Martin Tisne. Bdeir joined in January after leading Mozilla’s AI strategy and previously founding littleBits, the STEM education company sold to Sphero in 2019.

Its structure matters. Bdeir describes Current AI as a “public-private partnership” that brings governments, companies, and philanthropies into public-interest technology funding. The French government seeded the project with $100 million, with the Ford Foundation, MacArthur Foundation, DeepMind, and Salesforce among the funders. Current AI says it has more than $400 million committed and is mobilizing $2.5 billion over five years, according to its own site.

So what is Current AI trying to build? A shared layer of AI resources that developers, nonprofits, researchers, public agencies, and communities can build on without starting from zero.

That stack could include:

• Chat interfaces: Open-source AI assistants for public use.
• Models and tooling: Language models, safety systems, and compute contributed by partner organizations.
• Datasets: Community-controlled language and cultural data.
• Device support: Offline or local AI tools where internet access is weak.
• Interoperability standards: Shared pieces that make projects easier to connect.

The live question: can a nonprofit coordinate this without becoming another gatekeeper?

Builders get an open AI stack, but not a blank check

Current AI has already moved beyond concept slides. Last month, it allocated $3.2 million in grants to four organizations across Kenya, Lebanon, and the Brazilian Amazon.

The projects show where Current AI thinks private AI leaves gaps.

For builders, the appeal is obvious. A shared AI stack can reduce duplicated work. A language dataset built for one public-interest project could help another. An audit tool built in Kenya might inform work elsewhere.

Current AI also launched Alpha Chat, an open-source chatbot assembled in seven weeks by a coalition of ten organizations, including Hugging Face, Mozilla, and MIT Media Lab. Each contributor supplied part of the stack, such as a language model, safety tooling, or computing power.

XOOMAR analysis: this is the nonprofit’s strongest near-term path. It doesn’t need to beat OpenAI, Google, or Anthropic on every benchmark. It needs to make neglected use cases cheaper to build and easier to maintain. That is a narrower test, but still a serious one.

The constraint is funding discipline. $3.2 million split across four organizations can seed useful work. It can’t, by itself, sustain a global AI alternative.

For adjacent context on how governments are thinking about AI sovereignty and security, see our coverage of how the US blocks pushed South Korea toward a security AI model.

End users get offline language tools before polished mass adoption

The clearest user example is Suno Sutra, built after Current AI teamed up with Bhashini, the Indian government’s AI language division, in February at the India AI Summit.

Suno Sutra, Hindi for “listening chronicles,” is a pocket-sized offline device that runs AI in 22 Indian languages, with no internet required. It is open-sourced for developer communities to build on.

“In India, there are hundreds of different languages and dialects, and right now AI is not representing them,” Bdeir told TechCrunch.

This is where Current AI’s “World Wide Web of AI” phrase becomes more concrete. If AI only works well through cloud services, premium devices, and dominant languages, it recreates the same access problem the web was supposed to narrow.

How could this work on ordinary devices? The source does not provide benchmarks, chip requirements, or performance data, so any claim about speed or reliability would be premature. But the direction is clear: offline tools, local processing, and open interfaces can make AI more useful where connectivity is fragile or language support is thin.

A practical mini case is already visible in the India example. A user takes a photo of a sick plant. The tool needs to process the image, handle the request in a local language, and return usable guidance without requiring English or a live internet connection.

That does not remove the need for human oversight. Crop advice, health navigation, legal forms, and financial guidance all need source checking, privacy protections, and clear limits. Current AI’s promise is access. Accuracy still has to be earned.

Consumer AI is also moving onto everyday devices in other ways, as seen in Phones Spin Text Prompts Into Games With Roblox Build, but Current AI is aiming at a different problem: who gets represented when AI moves beyond English-first interfaces?

Big AI faces a consent argument, not just a feature race

Current AI’s critique of mainstream AI is not that large companies ignore multilingual systems entirely. Bdeir draws a sharper line around incentive and consent.

“Big tech builds multilingual models to expand their market,” she said, “regardless of consent or context.”

Her concern is cultural capture. Half the world’s spoken languages face extinction, according to the source material, and Bdeir argues that English-driven AI systems leave many languages, cultures, and communities behind.

The issue goes beyond translation. A culturally aware chatbot must understand idioms, local institutions, social norms, and community knowledge. It also has to know when not to ingest data.

“For Indigenous languages, missionary Bible translations become training data before communities have set any rules,” Bdeir said.

Current AI says its approach is to store models and data locally, bring in community experts before systems are built, and write consent protocols into the pipeline so communities can halt the process. Bdeir acknowledged that none of Current AI’s grantees have fully solved the data ownership problem yet.

That admission is important. Public-interest AI can’t claim moral authority just because it is nonprofit. It still has to answer who decides, who audits, who can object, and who can stop a project when cultural or privacy lines are crossed.

Funders are testing whether public-interest AI can survive contact with scale

The biggest signal around Current AI is not one device or chatbot. It is the coalition behind it.

Current AI has funding commitments from governments, philanthropies, and technology organizations. It has grants in multiple regions. It has an open-source chatbot built with major research and open-source names. It also has a deal with Sakana AI, a Tokyo-based startup known for work on Sovereign AI, to build a shared open-source AI stack supporting Japanese language and culture as well as communities across the Global South.

The risk list is long:

• Trust: Open systems still need safeguards against misinformation, privacy leaks, abusive automation, and unsafe advice.
• Funding: Free access does not make compute, security, localization, or support free.
• Governance: Communities need real power over data, models, and consent.
• Quality: Tools must work reliably in the languages and contexts they claim to support.
• Adoption: Developers need reasons to build on the stack instead of treating it as a grant-funded side project.

XOOMAR analysis: Current AI will not be judged by whether “World Wide Web of AI” sounds elegant. It will be judged by whether a clinic, school, farm cooperative, or local government office can use its tools affordably, safely, and in the language people actually speak.

The next practical test is whether projects like Suno Sutra, Alpha Chat, and the $3.2 million grant cohort produce reusable infrastructure rather than isolated pilots. If they do, Current AI becomes more than a counterargument to private AI. It becomes a working public layer.

Impact Analysis

• Current AI is trying to create a free public alternative to corporate AI infrastructure.
• Its focus on underserved languages could expand AI access for communities ignored by dominant models.
• The project must still prove it can handle compute, safety, governance, localization, consent, and quality at scale.