BREAKINGVIEWS-AI data crunch speeds towards Napster moment

The author is a Reuters Breakingviews columnist. The opinions expressed are his own.

By Sebastian Pellejero

NEW YORK, Dec 19 (Reuters Breakingviews) - The free lunch will come to an end for artificial intelligence in 2026. Over the past decade, developers from Google to Alibaba have largely been helping themselves to the internet buffet, devouring copyrighted material without permission or payment. Make no mistake, however: the bill is coming soon.

Consider it AI’s Napster moment. More than 50 copyright lawsuits are under way, with major trials on the docket. Getty Images, The New York Times and NBCUniversal are among those queueing up to collect on their digital files being fed into large language models. Training on authorized data will raise costs, steering companies into licensing deals and partnerships, where access, capital and size promise to create a bigger industry divide.

The backlash is worldwide, from an entertainment and media industry estimated at some $3 trillion by consultancy PwC. Britain has scrapped plans to exempt text-and-data mining after publishers revolted and is instead shifting toward a permissioned regime. The European Union’s AI Act already mandates such disclosures for machine learning. China goes even further: data used for training must be fenced, licensed and ideologically screened before use.

In the United States, legal boundaries are fuzzier. Courts have so far split over whether training on pirated book collections qualifies as fair use, ensuring the fights will persist. The U.S. Copyright Office notes dozens of pending cases, a litigation overhang already shaping business plans.

Some companies are capitulating, if for no other reason but to firm up predictability. Anthropic, for example, paid $1.5 billion in September to settle claims over roughly 500,000 books, or about $3,000 apiece, and agreed to destroy its pirated datasets. More than 70% of the datasets used to train major AI models lacked any specified license, according to the Data Provenance Initiative. Settling may be cheaper than testing that number in court.

The music business, scarred from pirating a quarter century ago, has jump-started the new order. Universal, Warner Music WMG.O and Sony Music Entertainment sued Udio for using copyrighted songs to produce AI-generated ones. As of late November, two of the three had reached agreements on partnerships that force the startup to retrain its model from scratch. They’re now co-developing tools using licensed catalogs, evidence that rebooting is becoming the price of legitimacy.

Cooperative ventures are multiplying, but selectively. Universal, led by Lucian Grainge and home to artists such as Lady Gaga and Jon Batiste, is also working with SoundPatrol on digital fingerprinting and with Stability AI on music creation. Photography giant Getty Images, which sued Stability AI over image generation, has licensed its archive to chatbot engine Perplexity. The New York Times rejected overtures from ChatGPT-owner OpenAI and sued, while the Associated Press and German media conglomerate Axel Springer opted for multimillion-dollar deals instead.

Such clashes have a recognizable beat. Peer-to-peer file-sharing pioneer Napster wound up settling with songwriters for $26 million and went bankrupt, but only after the firm and its copycats had gutted CD sales. Yahoo’s “free news” experiment with Reuters, the parent company of Breakingviews, set an early precedent that even Chapter 11 could not erase. AI should break the pattern of destruction, as ideas about free content are reconsidered in real time.

Timing, as ever, favors the incumbents. Technology titans cultivated their foundational models when scraping the internet was still free. ChatGPT now boasts 800 million weekly users, Anthropic earns over $7 billion in annualized sales and Google’s Gemini serves some 650 million users monthly. With deep pockets and established audiences, they can afford to pay for forgiveness after having not sought permission. Latecomers face a market where both data and distribution are increasingly spoken for with exclusivity arrangements.

Resetting the rules will not democratize AI; it will consolidate the fledgling field. Litigation raises barriers, rewarding those with capital, computing power and proprietary archives. The music business offers a lesson: thousands of labels shrank into three giants that control 80% of the market, as streaming economics rewarded sheer size more than creativity.

Images are moving in a similar direction. A London judge in November largely sided with Stability AI in its fight with Getty Images. The court ruled that training was not considered infringement because the model did not contain or store Getty’s material. It also decided that allowing British users to access a model hosted abroad does not amount to importing images into the country. The sole blemish was narrow. Early models sometimes processed Getty-style watermarks, a snag since fixed.

These outcomes are starting to affect the way investors think. Owners of large, structured collections – publishers, music producers, stock-photo houses – now sit atop scarce feedstock. Provenance authenticators and rights audits will become standard line-items in AI budgets, complete with renewals and royalties.

Costs won’t be as onerous as the 70% of revenue that Spotify doles out to song-rights holders, but the expenses probably will erode several points of gross margin after totting up the many mouths that demand feeding. A shift toward royalties linked to usage is also likely to accelerate. Financial engineering will be right behind: future data outlays seem ripe for securitization, echoing the transformation of future streams into tradable assets.

The internet age frayed the link between creators and compensation. AI is stitching it back, imperfectly but at a large scale. What was once indiscriminately picked over without charge will soon be logged, licensed and renewed, turning rights into rent and data into capital. The upcoming feast will be a movable one.

Follow Sebastian Pellejero on LinkedIn.

This is a Reuters Breakingviews prediction for 2026.

(Editing by Jeffrey Goldfarb; Production by Oliver Taslic)

((For previous columns by the author, Reuters customers can click on PELLEJERO/Sebastian.Pellejero@thomsonreuters.com))