Electricity didn’t ask permission to reshape the twentieth century. Neither did the internet. AI is following the same pattern — moving from tool to infrastructure faster than the institutions built to govern culture can respond. The question is no longer whether AI will change how made. It is who will own the system that makes it.
AI Is Already Infrastructure — Not a Tool
The framing still used in most policy debates — AI as a technology to be regulated — is already out of date. The World Economic Forum’s April 2026 report explicitly classifies AI infrastructure as critical infrastructure, placing it alongside power grids and telecommunications networks. The investment numbers reflect that reality. Meta has committed up to $72 billion in AI infrastructure for 2026 alone. OpenAI and Nvidia have formalised a $100 billion agreement. Goldman Sachs projects global data centre power consumption will expand from roughly 55 gigawatts today to 84 gigawatts by 2028.
Infrastructure, by definition, is what everything else runs on. As we explored in AI Runs on Power, the energy implications of this buildout are already reshaping European industrial policy. The cultural implications are less visible — but no less structural.
The Volume Shift in Cultural Production
The scale of AI-generated content is no longer marginal. An Ahrefs analysis of 900,000 new web pages in April 2025 found that 74.2% contained AI-generated content. A separate study by SEO firm Graphite, covering 65,000 articles published between 2020 and 2025, found that roughly half of all newly published articles since late 2024 were primarily written by AI — a share that held steady through 2025. In video, four of YouTube’s top ten channels by mid-2025 were using AI-generated material across their entire output. Projections suggest more than 60% of digital video will be at least partially AI-generated by 2027.
This is not a story about quality — it is a story about volume and velocity. The sheer quantity of AI-assisted content is already reshaping what gets seen, ranked, and consumed. As examined in Does AI Standardise Culture — or Expand It?, the democratisation of production does not automatically produce diversity of output.
The Copyright System Was Not Built for This
The legal framework governing cultural ownership is under acute stress. More than 70 AI copyright lawsuits were filed in 2025 alone, including the ongoing New York Times case against OpenAI and Microsoft, which entered discovery in January 2026 with a judge ordering OpenAI to produce its full log of anonymised ChatGPT conversations. In June 2025, two landmark rulings — Bartz v. Anthropic and Kadrey v. Meta — found that training on legally acquired works constitutes fair use, analogising the process to human learning. The same judge ruled separately that Anthropic’s use of pirated material was not protected. The distinction is important: legal sourcing matters more than the act of training itself.
The US Copyright Office has clarified that AI-assisted works retain copyright protection where human creative expression remains clearly present, while works whose expressive elements are determined by the machine do not. As of March 2026, the Supreme Court denied challenges to that position. In Europe, the Commission’s December 2025 consultation on text and data mining opt-out protocols represents an attempt to move copyright enforcement from the courtroom into technical infrastructure — standardising how rights-holders can limit AI scraping. The outcome of that process will shape the terms on which European culture enters AI training pipelines for years to come.
The Ownership Structure That Nobody Voted For
The deeper shift is structural. In the twentieth century, record labels and publishers owned the means of cultural distribution. In the 2010s, platforms — Spotify, Netflix, YouTube — took over that function, deciding what circulated and what didn’t. The current transition goes further: generative models and the infrastructure companies behind them are beginning to determine not just what circulates but what gets made in the first place.
Spotify decides which songs reach listeners. A generative model decides what a song sounds like. That asymmetry — between a distribution bottleneck and a production bottleneck — is qualitatively different. The companies building and controlling those models are simultaneously the largest investors in each other. Microsoft has invested $50 billion in OpenAI; OpenAI has committed $300 billion in cloud spending back to Microsoft. Similar circular arrangements link Google, Nvidia, and Anthropic. The “Blob” dynamic — where competition gives way to mutual dependency among a small number of firms — means that the infrastructure shaping cultural production is consolidating faster than any regulatory framework can track.
Europe’s Window, and What It Would Take to Use It
As noted in The Hidden Cost of AI: The Productivity Paradox, the economic returns from AI investment remain unevenly distributed and poorly measured. The cultural returns are even harder to account for — which makes them easy to leave out of policy calculations until the structure is already set.
Europe has leverage it has not yet fully used. The EU’s AI Act, the Commission’s TDM consultation, and the European Parliament’s calls for unified opt-out mechanisms and fair licensing models represent a coherent, if incomplete, framework. The question is whether those instruments can move fast enough to shape the infrastructure layer — not just regulate its outputs. The EU needs purposeful, open-source, and sustainable public AI — infrastructure that serves the public, not an AI-first race driven by dominant vendors.
The democratisation of cultural production and the concentration of cultural infrastructure are happening simultaneously. Generative AI has made it cheaper than ever to make things. It has also made it easier than ever for a small number of companies to determine the conditions under which making happens. Those two facts are not in contradiction. Both angles tell the same story.
Key Sources
- WEF AI Infrastructure report, April 2026: https://www.weforum.org/
- Ahrefs AI content analysis, 2025: https://ahrefs.com/
- US Copyright Office, AI Copyrightability Report Part 2: https://www.copyright.gov/ai/
- AI Multiple, Generative AI Copyright 2026: https://aimultiple.com/generative-ai-copyright
- Open Future, Public AI White Paper: https://openfuture.eu/our-work/ai-and-the-commons/
- European Commission TDM consultation, December 2025: https://ec.europa.eu/
Subscribe to EuroLuminant for independent European journalism.
