Current AI Aims to Build the World Wide Web of AI for Everyone

Picture a farmer in rural India. Her crops are struggling and she needs to identify a disease on a dying plant. She has a phone. She has access to a camera. She does not speak English, does not have reliable internet, and none of the major AI assistants that could help her exist meaningfully in her language. She is, in the eyes of the current AI industry, an edge case. For Current AI, the nonprofit that published its progress this week in a TechCrunch profile, she is precisely the person the organization was built for.

Current AI CEO Ayah Bdeir described this exact scenario in an interview with TechCrunch, using it to explain the core problem the nonprofit is trying to solve: building open, public AI infrastructure for people and communities that dominant AI systems have largely ignored. The comparison Bdeir reaches for repeatedly is the World Wide Web itself, the network protocol that turned privately controlled computing infrastructure into a public resource accessible to anyone with a connection. Current AI's thesis is that AI needs an equivalent, a shared infrastructure layer that no single company owns, that runs on shared standards, and that is free at the point of use for the communities that need it most.

Who Current AI Is and Where It Came From

Current AI was founded in February 2025 by Martin Tisne, operating as a public-private partnership that brings together governments, companies, and philanthropies to fund public interest technology. The French government provided the initial seed of $100 million, and that was joined by the Ford Foundation, the MacArthur Foundation, Google DeepMind, and Salesforce, bringing the total committed funding to approximately $400 million. Bdeir is careful to draw a distinction that matters for understanding how the organization operates: these are not investors. They are funders. Current AI is not building toward an exit or a liquidity event. It is building public infrastructure.

Bdeir herself joined in January 2026, bringing a background that is unusually well matched to the problem she is now working on. She previously founded littleBits, the STEM education company that put programmable electronics into the hands of millions of children and eventually sold to Sphero in 2019. Before joining Current AI, she led Mozilla's AI strategy, which gave her direct experience in what it looks like when a mission-driven organization tries to maintain an open alternative inside an industry dominated by corporate platforms. The combination of grassroots hardware experience, education-focused design thinking, and open-source organizational leadership is not a random resume for a nonprofit trying to democratize AI infrastructure at global scale.

The project seeks to establish shared AI resources that can be accessed by researchers, governments, universities, startups, and nonprofit organizations instead of relying on closed ecosystems. The organization describes the World Wide Web of AI as an interconnected network where different AI systems can exchange information using common standards. Similar to how websites communicate across the internet, the initiative aims to connect AI technologies through shared protocols instead of isolated platforms.

"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. Like the World Wide Web, available to anyone, for free."
- Ayah Bdeir, CEO, Current AI

The Language Problem That Current AI Is Actually Solving

The most visible and emotionally resonant part of Current AI's work so far is its approach to language, specifically the languages that the AI industry has systematically underserved. Half the world's spoken languages face extinction, and with English driving the largest language models and AI systems, a bulk of the world's cultures and communities are being left behind. That is not a rhetorical flourish. It is a statistical reality about where training data has come from and which communities have had the resources to contribute to that data in ways that major AI companies have been willing to use.

The distinction Bdeir draws between what Current AI is doing and what large commercial AI companies are doing on the multilingual front is worth taking seriously. When asked about Big Tech's multilingual push, Bdeir drew a sharp distinction. "Big tech builds multilingual models to expand their market," she said, "regardless of consent or context." The consequences are concrete. "For Indigenous languages, missionary Bible translations become training data before communities have set any rules," she said.

The consent and context point is not abstract. When a commercial AI company trains a model on data in an Indigenous language, the communities whose language and cultural knowledge are embedded in that data typically have no say in how that data is used, what commercial products are built on it, or whether the resulting AI accurately represents their culture and knowledge rather than distorting it. The business model of expanding into new language markets has nothing inherently to do with preserving or accurately representing those languages. It has to do with making a product useful to more paying users.

"Language is how knowledge, tradition, memory and identity get carried from one generation to the next. So when a technology can't speak your language, it can't hold your culture either," Bdeir said. That framing moves the language access question from a technical problem, how do we add more language support to our model, into a cultural sovereignty question: who gets to decide how a language and its associated knowledge are represented in AI systems that will increasingly mediate how information is stored, accessed, and communicated.

Suno Sutra: AI in 22 Indian Languages, No Internet Required

The most concrete proof of concept that Current AI has shipped so far is a device called Suno Sutra, which translates from Hindi as listening chronicles. In February at the India AI Summit, Current AI teamed up with Bhashini, the Indian government's AI language division. The result became Suno Sutra, a pocket-sized, offline device that runs AI in 22 Indian languages, with no internet required. "In India, there are hundreds of different languages and dialects, and right now AI is not representing them," Bdeir said. The device is open-sourced, available for developer communities to build on.

The offline requirement is not a limitation but a design decision that reflects a realistic understanding of where the people who most need this kind of tool actually live. Rural India does not have reliable broadband. A device that requires a stable internet connection to function AI capabilities is, in practice, a device for urban educated users who already have alternative ways to access information in their language. An offline device that stores models locally and runs entirely on the hardware in a pocket is something that a farmer in a village with intermittent mobile connectivity can actually use when she needs it.

The open-source approach is equally deliberate. A device built around a closed proprietary system gives the developer community no ability to extend it to the additional languages and dialects that the initial version did not cover. An open-source device gives local developers, government agencies, and civil society organizations in India the ability to add new language models, improve existing ones, and build specialized applications on top of the base hardware. The goal is not to provide a finished product for India's underserved languages. It is to provide a foundation that the people closest to those languages can build on.

The First Grant Cohort: $3.2 Million Across Four Continents

Current AI's first cohort grant round, announced in June 2026, deployed $3.2 million to four organizations across Kenya, Lebanon, and the Brazilian Amazon. The geographic spread of the initial grant cohort is meaningful, because it reflects a deliberate strategy of working across multiple regions simultaneously rather than picking a single focus area. Each of the four projects addresses a different dimension of the same underlying problem: communities whose languages, cultures, and knowledge systems are underrepresented in the AI infrastructure that is being built primarily in English, Mandarin, and a handful of other commercially viable languages.

• Masakhane, Kenya: Building AI datasets across more than 50 African languages specifically for health, farming, and education use cases, the exact contexts where access to accurate information in a local language can have the most direct impact on people's lives and livelihoods
• Institute for Worldmaking, Lebanon: Digitizing Arab cultural history and contemporary practice into machine-readable databases that communities rather than technology companies control, addressing the data ownership question at the level of cultural heritage rather than just current language use
• Portal sem Porteiras, Brazil: Building offline AI tools with Indigenous Amazon communities and explicitly keeping data within the territory rather than uploading it to external servers, operationalizing data sovereignty as a design requirement rather than an aspiration
• African Internet Rights Alliance, Kenya: Developing audit tools to hold AI systems accountable across the African continent, addressing the problem that even when AI systems claim to support African languages, the quality, accuracy, and cultural appropriateness of that support is difficult to assess and challenge without dedicated evaluation infrastructure

When asked about how much progress can be made with a $3.2 million budget split across four organizations, Bdeir pushed back on the framing. "Scale is not always the measure. That is the Big Tech paradigm," she said. "This could look like an Indigenous elder in the Brazilian Amazon using a tool built in Kenya to be able to pass down ecological knowledge in their own language." That is a different metric for success than monthly active users or revenue growth, and it is worth taking seriously as a genuine alternative framework rather than dismissing it as naive about the realities of scale.

Alpha Chat: An Open-Source AI Built in Seven Weeks by Ten Organizations

The most recent project Current AI has launched publicly is Alpha Chat, an open-source AI chatbot that it debuted at the AI for Good Summit in Geneva in July 2026. Alpha Chat was assembled in seven weeks by a coalition of ten organizations, including Hugging Face, Mozilla, and MIT Media Lab. Each contributor brought a piece of the stack, including a language model, safety tooling, and computing power.

The seven-week timeline and the coalition model are both deliberate demonstrations of what the World Wide Web of AI concept looks like in practice. A single company building a chatbot in seven weeks would be building on infrastructure it already owned, trained on data it had already licensed, evaluated against safety standards it had already developed. Current AI assembled an equivalent capability by connecting organizations that had different pieces of that infrastructure and establishing protocols for them to work together. The result is a chatbot where no single organization owns the model, the safety tooling, or the compute, and where the contributions from each participant are publicly documented and available for others to build on.

Alpha Chat is not intended to compete with ChatGPT or Claude on capability benchmarks. Its purpose is different: to demonstrate that capable, safe AI chatbots can be assembled from open-source components by coalitions of organizations with different expertise rather than requiring the concentrated resources of a frontier AI lab. If that demonstration is credible, it changes the range of actors who can participate meaningfully in AI development rather than simply consuming what the largest companies produce.

The Sakana AI Partnership and the Sovereign AI Framework

Current AI struck a deal with Sakana AI, a Tokyo-based startup known for its work on what it calls Sovereign AI. The two organizations plan to build a shared open-source AI stack designed to support the Japanese language and culture, but also communities across the Global South that dominant AI systems have largely ignored.

The Sakana partnership is notable because it connects Current AI's global south focus with a well-resourced AI research organization based in one of the world's most significant non-English language markets. Japan presents a specific version of the language sovereignty problem that is distinct from the African and Amazon contexts: Japanese has a large, highly literate, technologically sophisticated user base, but the cultural and linguistic specificity of the language, including its three writing systems, its extensive honorific grammatical structures, and the depth of its literary and cultural tradition, means that general-purpose English-primary AI systems represent it imperfectly in ways that matter to Japanese users.

Sakana AI's framing of its work as Sovereign AI, rather than simply multilingual AI, reflects the same underlying concern that Bdeir articulates for underserved Global South languages: the question of who controls the AI systems that represent a culture is a sovereignty question, not just a technical or commercial one. A shared open-source stack built in partnership between Current AI and Sakana that serves both Japanese and Global South communities creates an interesting alliance between a wealthy market with a sophisticated AI ecosystem and communities that have been largely excluded from AI development, connected by a shared commitment to AI infrastructure that communities rather than corporations control.

The Data Ownership Question That No One Has Solved Yet

One of the most honest and important parts of Current AI's public communications is the acknowledgment that even its grantees have not fully solved the data ownership problem that Bdeir identifies as central to the organization's mission. "There are different models and proposals for who owns data in various communities, but one thing is sure: it shouldn't be a company in Silicon Valley trying to make a select few thousand people wealthier," Bdeir said. The nonprofit's approach is to store models and data locally, bringing in community experts before anything is built, and writing consent protocols into the pipeline so communities can halt the process at any point. None of Current AI's grantees have fully solved it yet. But Bdeir sees that as the point. "Every one of them has built the question into their work," she said, "rather than accepting the usual default, where complexity becomes the excuse to let a government or a tech company decide for everyone."

That honest reckoning with the gap between aspiration and current implementation is one of the things that distinguishes Current AI's public communications from the typical nonprofit communications that overstate demonstrated impact to satisfy funders. Building data sovereignty into the design philosophy of each project, even when the specific mechanisms for implementing it have not been fully resolved, is a structurally different approach from the typical tech company approach where data governance concerns are addressed, if at all, after the product has already been built and deployed at scale.

Why This Moment Is the Right Time for This Project

Current AI's announcement and the attention it has received from TechCrunch and other outlets this week comes at a specific moment in the AI industry's development that matters for understanding why the organization's approach is more relevant now than it would have been even two years ago.

The AI industry in 2026 is beginning to mature in ways that create both more urgency and more possibility for the kind of public infrastructure project Current AI is pursuing. On the urgency side, the patterns of who benefits from AI and who does not are beginning to calcify. The commercial AI companies have identified the markets they are building for. The training data they have used reflects the languages, cultures, and knowledge systems that were well-represented in English-language internet text. The communities that were not well-represented in that training data will find themselves interacting with AI systems that understand them poorly if those systems become the default infrastructure for accessing information, completing administrative tasks, and participating in economic life.

On the possibility side, the open-source AI ecosystem has matured enough that the gap between what a well-funded nonprofit can build on open foundations and what a frontier commercial lab produces has narrowed considerably from where it was in 2022. Hugging Face's model hub, the proliferation of smaller but capable open-source language models, and the growing ecosystem of safety and evaluation tools have all made it more realistic for an organization like Current AI to assemble credible AI infrastructure from open components rather than either building from scratch or depending entirely on commercial models whose terms of use they cannot control.

How Good Is the World Wide Web Analogy?

The World Wide Web comparison that Bdeir and Current AI use as their central organizing metaphor deserves examination rather than acceptance or rejection on its rhetorical appeal alone. The analogy is both illuminating and imperfect in specific ways that are worth understanding.

What it gets right is the structural claim: the early web was built on open protocols that no single company owned, and that openness is what allowed the internet to develop into a genuinely global resource rather than a collection of proprietary networks. HTML, HTTP, and TCP/IP are public goods in the sense that anyone can use them to build anything that communicates over the internet. The value of those protocols is inseparable from their openness: a proprietary protocol that only some organizations could use would not have produced the world wide web. It would have produced an AOL. Current AI's claim is that AI needs equivalent open protocols, shared model weights, shared safety standards, and shared evaluation frameworks that anyone can use as infrastructure.

Where the analogy is imperfect is in the resource requirements. A web server in 1995 required a modest amount of computing hardware and some knowledge of HTML. Training a frontier AI model in 2026 requires billions of dollars of GPU compute, petabytes of curated training data, and teams of research scientists. The open web was accessible to individuals and small organizations from the beginning. Open-source AI infrastructure at the scale that Current AI is describing is not yet accessible at anything like the same level, even with open weights and open protocols. What Current AI is more precisely aiming to create is not the equivalent of letting anyone build a website, but the equivalent of letting anyone access the network rather than requiring them to go through a gatekeeper, which is a meaningful but different kind of openness.

What to Watch as Current AI Grows

Current AI is still a young organization moving fast through an early phase of work where its most important achievements have been demonstrations of possibility rather than infrastructure at scale. The Suno Sutra device, Alpha Chat, and the first grant cohort all represent proof of concept for different dimensions of the World Wide Web of AI vision. The question the next twelve to eighteen months will answer is whether those proofs of concept can develop into something that actually changes who has access to AI infrastructure in a durable and scalable way.

The $400 million in committed funding provides enough runway to move significantly beyond demonstration projects if the organization makes good choices about where to allocate it. Whether Current AI uses that funding to expand its grant program to more languages and communities, to deepen the technical infrastructure underlying Alpha Chat and Suno Sutra, to build the shared standards and protocols that would allow different organizations' AI systems to interoperate in the way the early web protocols did, or some combination of all three, will determine what the organization actually looks like in two years rather than what it aspires to be.

The partnership with Sakana AI and the coalition that built Alpha Chat both suggest that Current AI understands its most powerful tool is not its own technical capability but its ability to coordinate organizations that have complementary capabilities toward a shared goal. If that coordination model scales, the World Wide Web of AI becomes something built by a distributed global community working on shared open infrastructure, which is exactly what the web analogy implies. If it does not scale, Current AI becomes a well-funded nonprofit that produced some genuinely useful tools for underserved communities, which is not nothing, but is not the structural alternative to commercial AI dominance that Bdeir's framing promises.

Related Topics: #CurrentAI #AIForGood #OpenSourceAI #AIAccessibility #MultilinguaAI #DataSovereignty #PublicAI #ArtificialIntelligence #Technology #AIEquity