How Big Tech Cash Is Shaping the AI Data Licensing Market

On the eve of its initial public offering in March, Reddit revealed that it had received a letter from the Federal Trade Commission seeking information about its then-recently announced deal with Google to license its vast archive to the tech giant for training generative AI models. Since then, the FTC’s focus on such deals has reportedly widened to other rights owners and other AI companies.

The timing of the Reddit disclosure was notable because of its pending IPO, but the FTC’s interest was no surprise. In comments last year to the U.S. Copyright Office, the FTC made clear its concern with the development of the competitive landscape for access to AI training data.

“Many large technology firms possess vast financial resources that enable them to indemnify the users of their generative AI tools or obtain exclusive licenses to copyrighted (or otherwise proprietary) training data, potentially further entrenching the market power of these dominant firms,” the antitrust agency wrote.

Last week provided a hint as to one way that market is developing. Wired reported that several leading publishers, including The New York Times, The Financial Times, Atlantic Media, Vox Media, the USA Today network and Condé Nast, are blocking Apple’s AI training bot from accessing their sites. As it happens, all of those publishers have signed data licensing deals with AI developers, but not with Apple, indicating something of a pay-to-play dynamic at work.

Pay-to-play, of course, is what licensing is all about. There is nothing wrong with publishers granting access to their content to AI developers willing to pay the freight while denying access to those that won’t. No one would question a publisher signing a deal — even an exclusive one — with a syndicator for the re-use of its articles and blocking others from doing so.

Apple, if fact, has shown itself willing to pay for access to training data for its AI tools, but for images, not text, having struck a deal with Shutterstock last year. Ultimately, we should welcome the emergence of a functioning data licensing market for AI training fodder. The question is whether that market will serve to entrench the market power of a few already dominant players, or facilitate broader competition.

In a forthcoming report, my colleagues at the Video Intelligence Platform and I list 30 AI data licensing deal that have been publicly announced or reported as of the beginning of August 2024. Of those 30 deals, six major technology companies account for 24 of them: Microsoft, OpenAI, Google, Amazon, Apple and Nvidia. One other deal, involving six named publishers, was announced by Perplexity.ai, which is backed by Jeff Bezos and Nvidia, among others, and one by Runway, backed by Nvidia, Google and Salesforce, among others.

Only four publicly announced data access deals involve licensees that don’t quite qualify as “major” AI companies.

Not all of those deals are exclusive, as the FTC warned against. But nearly all involve major publishers or rights owners with the kind of high-quality, professionally produced content increasingly in demand by AI developers as more publishers put their content beyond the easy reach of web scrapers. So far, at least, there’s not a lot of evidence of either smaller publishers or smaller AI developers landing deals.

The FTC is not the only agency concerned over the monopolistic proclivities of the big tech companies, however. Last month, the antitrust division of the Justice Department scored a landmark win when U.S. district judge Amit Mehta ruled that Google maintains an illegal monopoly over the online search business. Key to the ruling were the huge sums Google pays to Apple, Samsung and others to make Google the default search engine on their devices, allowing it to lock up 90% of the search market.

At a hearing on Friday before Judge Mehta , DOJ attorney David Dahlquist said the government now wants to probe Google’s AI strategy to prepare its recommendations for remedies in the case and requested additional discovery from the company.

The AI business is now shaping up like the rest of the tech economy, with a handful of very large, dominant firms able to overwhelm, buy up or otherwise prevent smaller startups from gaining a foothold in tech markets from which to challenge the giants. According to a study by CB Insights, Meta, Microsoft, Google, Amazon, and Apple collectively acquired 98 AI or machine learning companies from 2010 through the end of 2023. This year has seen a rash of partnerships, technology licensing deals and other arrangements between tech giants and startups that critics argue amount to acquisitions in all but name, presumably to avoid greater regulatory scrutiny.

Given the demonstrated — and now adjudicated — willingness of the tech giants to throw their cash around to protect their positions in both established and emerging markets, the FTC has good reason to scrutinize the development of the licensing market for AI training data.