AI as Public Infrastructure: A Non-Profit Building the Internet for Everyone

An Indian rural woman takes a photo of a withered plant with her phone. She wants to know what disease the plant has, but she doesn't speak English. Why shouldn't she be able to access the world's cutting-edge knowledge in her own language?

This scenario was my first impression of that TechCrunch report. Behind it is a non-profit organization called Current AI, doing something that sounds crazy—building a free, open, accessible "World Wide Web of AI."

[!tip] The core reason this news is worth attention isn't how advanced the technology is, but that it redefines the ownership of AI. While tech giants are busy locking AI behind paywalls, a group of people want to turn it into a public library.

From "Digital Divide" to "AI Divide"

For the past decade, when we talked about the "digital divide," we meant whether you had internet access. Now, this divide is becoming the "AI divide"—whether you can access the best AI models.

Farmers querying plant diseases in their local languages is just the tip of the iceberg. Current AI's idea is: let people of any language and cultural background gain equal knowledge and capabilities through AI. This isn't a charity project, but an infrastructure-level ambition.

Looking back at history, the internet itself was originally a public network co-built by universities and research institutions. Later it was commercialized, but the underlying protocols—TCP/IP, HTTP—remained open. What Current AI wants to do is create a similar "protocol layer" for the AI era. They don't sell models or APIs; they build an open-source, decentralized AI ecosystem where any organization or individual can deploy, train, and use AI.

This sounds a lot like Linux for operating systems, or Wikipedia for knowledge bases. But the complexity and cost of AI far exceed those predecessors.

How Can It Be Free? Technical Path and Survival Logic

Anyone familiar with the AI industry would ask: Where does the money come from? Training a large model costs tens of millions of dollars, and inference costs aren't low either. Current AI's answer isn't simply "relying on donations," but a hybrid strategy of technology + community.

First, model miniaturization and distributed inference. They don't pursue the largest parameter count "omnipotent models," but focus on efficient, locally deployable models. Combined with edge computing and distributed networks, computing demands are dispersed across thousands of nodes. This is somewhat like the BitTorrent approach—everyone contributes a little computing power, everyone enjoys the benefits.

Second, data sovereignty and multilingual support. Giant models mainly rely on English internet data, offering poor support for low-resource languages (like Hindi, Swahili). Current AI emphasizes collaboration with local communities to collect and annotate localized data, ensuring models truly understand local contexts. This is not just a technical issue, but a cultural rights issue.

Third, governance of the open-source ecosystem. They don't just dump code on GitHub; they establish mechanisms for contribution, review, and incentives. Similar to the Apache Foundation or Python community, but adapted for the characteristics of AI models.

Will this path work? It's hard to say yet. But looking at industry trends, open-source AI is moving from "toys" to "productivity." Meta's Llama, Mistral, and other open-source models have proven that open models can catch up with closed ones in certain scenarios. Current AI's uniqueness lies in treating "openness" not just as a technical strategy, but as a social mission.

Openness vs. Monopoly: The Crossroads of the AI Era

If Current AI succeeds, what happens? Imagine a doctor in an African village training a diagnostic model in their local language without paying any US company; a South American artisan using AI to translate product descriptions to reach global markets directly; a teacher in a small Chinese county town using AI to generate personalized lesson plans covering all subjects.

But the resistance is equally huge. Existing AI giants have deep moats: computing power, data, user habits, capital. More importantly, the free model itself might be a "poison"—if everyone relies on free AI, where does the motivation for maintenance and upgrades come from? Wikipedia's predicament serves as a warning: non-profits require sustained social donations, while AI's operational costs far exceed Wikipedia's.

Another issue is security. Open AI models are easier to abuse and tamper with. Can Current AI's governance architecture effectively prevent malicious use? There is no answer yet.

[!quote] "Farmers should not be forced to query plant diseases..."

Original link: https://techcrunch.com/2026/07/19/nonprofit-current-ai-is-racing-to-build-the-world-wide-web-of-ai-free-for-all/