非営利団体 Current AI、誰でも無料で使える「AIのワールドワイドウェブ」構築を急ぐNonprofit Current AI is racing to build the World Wide Web of AI, free for all

Current AI, a nonprofit initiative that positions itself as a public-interest counterweight to the handful of companies dominating artificial intelligence, is accelerating efforts to build what it describes as a freely accessible "worldwide web for AI." The stated goal is to make core AI capabilities available to anyone at no cost, echoing the open, decentralized ethos of the early internet and pushing back against a future in which advanced models sit behind corporate paywalls. For a technology increasingly woven into work, education, and public services, the question of who controls the underlying infrastructure carries significant economic and social weight.

The organization frames its mission around a straightforward concern: the most capable AI systems are increasingly controlled by a small number of well-funded firms, primarily in the United States and China. Training frontier models requires enormous amounts of computing power, large proprietary datasets, and specialized talent, resources that are difficult for smaller developers, academic researchers, and public institutions to match. Current AI argues that leaving this infrastructure entirely to commercial actors risks concentrating power, and it wants to establish shared, openly available alternatives before those dynamics become entrenched.

Current AI was launched publicly around the Paris AI Action Summit in early 2025 as a partnership spanning governments, philanthropies, and technology organizations. Backers reportedly included France alongside other national governments and philanthropic funders, with early financial commitments in the hundreds of millions of dollars and ambitions to raise considerably more over several years. The effort has been associated with public-interest technologist Martin Tisné, and its work appears to center on a few broad pillars: opening access to high-quality datasets, supporting the development of open models and tools, and funding compute and infrastructure that independent builders can draw on.

The "worldwide web" comparison is deliberate. In the early internet, common protocols and open standards allowed anyone to publish and build without seeking permission from a gatekeeper. Current AI is likely aiming to reproduce that layer for AI, providing openly licensed components that developers in different countries and sectors can adapt to local needs rather than depending on a few commercial application programming interfaces. That framing situates the group within a broader movement rather than as a standalone project.

Several adjacent efforts help illustrate the landscape. Hugging Face has become a widely used hub for sharing open models and datasets, while organizations such as EleutherAI and LAION have produced openly available training data and research. Companies including Meta, with its Llama family of models, and France's Mistral AI have released open-weight systems, though the degree of openness varies and some licenses carry restrictions. On the compute side, national and regional programs, including European public supercomputing initiatives, have sought to give researchers access to hardware they could not otherwise afford. Current AI's contribution appears to be an attempt to coordinate and fund these strands under an explicitly public-interest banner.

The obstacles are substantial. Building and serving large models is expensive, and sustaining free access at scale requires either continuous funding or a model that offsets costs over time. Assembling high-quality datasets raises questions about copyright, consent, and licensing, issues that have already prompted litigation across the industry. There are also governance and safety considerations: openly distributed systems are harder to restrict once released, which some researchers view as a benefit for transparency and others see as a risk. How Current AI balances openness against these concerns will likely shape its credibility.

It is too early to judge whether the initiative can deliver infrastructure that rivals commercial offerings in capability and reliability. Much depends on the scale of funding it secures, the partners it retains, and whether governments continue to support it amid shifting political priorities. Even so, the effort reflects a growing recognition that access to AI may become a matter of public infrastructure rather than purely private enterprise. Whether the "AI worldwide web" becomes a lasting foundation or remains an aspiration, it underscores an intensifying debate over who should own the tools that increasingly mediate digital life.