Updated Oct 23, 2023 - Economy

The big risk behind the AI investment boom

Illustration of Benjamin Franklin from a $100 dollar bill blowing a chewing gum bubble

Illustration: Sarah Grillo/Axios

Venture capitalists are pouring huge money into AI startups, with a dollar weighting toward those building foundational models.

Why it matters: The VC market may be making its biggest binary bet ever, without quite realizing it.

Behind the scenes: Many of those models are being built, in part, on data that is under U.S. copyright protection. Several of the copyright holders, including major music companies, have filed suit.

  • But venture capitalists don't seem to care

The risk: If some of these lawsuits are successful, they could eviscerate the cost structures of companies that have raised billions of dollars.

  • Such startups already are dealing with massive compute expenses, but also would need to pay rights fees before data even reaches the center. Camels crushed by countless straws.
  • "This copying to storage for use in AI training is a likely clear-cut case of infringement, and the courts will put a stop to it," Jason Peterson, CEO of IP right management and claims firm GoDigital Media Group, told Axios last week.

The response: Axios reached out to several VCs who are invested in some of the top LLM developers, and none sounded even the least bit concerned. Then again, no one wanted to go on the record, instead preferring to reply on background.

  • Several cited legal arguments about fair use, at least in regard to data on which the models had already been trained.
  • "AI training in most cases is not analogous to copying or plagiarism," one VC said. "These models for the most part do not plagiarize or copy data they were trained on, and they are not databases with lookup tables. As a result, they are not equivalent to the precedent set by Napster and others in the music piracy era."
  • Another VC added: "The models rarely regurgitate the exact thing that they trained on. They train on a lot of different things, and then they generate based on probabilistic models. Arguably, this is exactly what humans do.  So, then would you could sue every creative work that a human creates as well.
  • One seemed comforted by Big Tech's investment involvement. "I suspect they will have researched this with their in-house lawyers before writing a billion dollar (or multi-billion dollar) check."

The bottom line: AI is tech's next great platform shift, following web, mobile and cloud. And being on the ground floor of a platform shift can absolutely be an investment goldmine.

  • But right now it feels like everyone is enamored by the gold without recognizing the inherent risks of mines. Or that bigger wins could come from those who already own their own data (e.g., Meta), those that focus on foreign data (looser copyright laws), or on smaller, enterprise-focused data with more manageable licensing fees.
  • If things do come crashing down, it will look obvious in retrospect.
Go deeper