How Much Should AI Giants Pay Hollywood? What Insiders Say Has Stalled Any Licensing Deals
In this article
Why haven’t any of Hollywood’s major studios pulled the trigger yet on a licensing deal with the tech giants looking to train their AI models?
With that question only growing louder among industry observers since OpenAI first unveiled Sora in February, Variety VIP+ spoke with several sources with direct knowledge of how studio execs are approaching the option as other content owners strike their own deals (see updated chart for full list) or otherwise bring lawsuits (see updated chart for full list).
It turns out there are many reasons — VIP+ has an in-depth exploration detailing each reason in a special analysis published in tandem with this report. Factors range from considerable legal risks to fears of repeating a grave strategic error Hollywood made in its first dealings with Netflix.
But one of the biggest concerns is a simple one: price.
Studios simply have no concept of how much their content is worth to AI companies for this kind of use, not even a ballpark sum or range, said one source who has advised studios on the prospect of licensing. Another source argued for transparency on pricing strategy among studios because the first deal would conceivably set the ceiling for subsequent ones. VIP+ has an analysis of how one dataset provider is negotiating the first licensing deals for film and TV content.
“With each of new wave of tech transformation, there’s just been enormous reticence by studios to be able to calculate the value of what they may be giving up. That’s a factor into the slowness [to license],” a source told VIP+. “The studios aren’t just sitting there doing nothing; they’re trying to understand what’s the best course of action.”
How to price studios’ film and TV content for AI training received mixed responses from those VIP+ spoke with, which suggests how little consensus and how much confusion there is on this question. Reps for the major studios either declined comment or did not respond to inquiry for comment.
One source argued the price should be set at the maximum value developers are willing to pay in the market for AI training data, in part because an AI model is ultimately indifferent to whether, for example, a live-action shot of a dog in a field came from a blockbuster movie or an indie film.
“If a film was produced for $50 million or if a film was produced for $500,000, the models are not discerning; they’re just learning what the world looks like,” a source told VIP+. “All that’s needed, according to these technology companies, is variety and volume.”
VIP+ Analysis: Licensing Your Movie & TV Content for AI Training — Could You? Should You?
Other sources argued that letting the market (developers) dictate the price for video to train AI — reported to range from $1 to about $6 a minute — would risk undervaluing the true worth of their content for what appeared to be a big number. They noted that doing the work to define the “true worth” for this purpose should happen in the context of existing distribution windows (e.g., broadcast, pay TV, SVOD) — or, better yet, according to a calculation of how much value studio content stands to bring to the developer’s product, both now and in the future.
That calculation might consider the current total addressable market of data licensing as 5% or 10% of the total amount that every AI company is spending on development, which one source noted was a sum between $17.5 billion to $35 billion.
“It goes back to the question, do the studios really know how much value they’re sitting on here? Are we witnessing a repeat of the Napster and early YouTube era of these companies just ostensibly stealing copyrighted material and publishers being paid a pittance relative to what value was created over time?”
Studios could conceivably break down their film libraries into a series of tokens, or the units of data processed by AI models. But, said the source, this sort of construct for properly valuing a content library for a generative AI context is completely foreign for studios.
Not knowing how to properly value content amid digital disruption could lead to undervaluing. “It’s just a fundamental lack of understanding of value. Know what it’s going to be worth to the company, not just this year but five years from now.”
An upfront lump sum for data may also not be enough. How a developer intends to build and distribute products or services off of the model in the future should also likely factor into the valuation. In that sense, publishers should probably also consider terms to perpetually share in ongoing revenues from any future AI products that benefit from their data in the training of the underlying model, a source told VIP+.
VIP+ Analysis: Fine-Tuning Video Models Getting Early Interest From Film & TV Studios
Studios may also have problematic blind spots about generative AI development that could lead them to undervalue their content. One source referred to the expected role of synthetic data as one such blind spot.
Though some disagree, prominent AI voices believe developers will eventually be able to train very good models entirely on high-quality synthetic (AI-generated) material. That potentiality should also likely be factored into the pricing for publisher content as training data — effectively the risk that AI developers may only need to license studio content in the short-term until synthetic data becomes good enough to replace licensed “human-made” data.
“With that mental framework, the question one should be asking is, didn’t you create synthetic data by using all of our IP and NIL data?” said a source. “And therefore, there should probably be a perpetual royalty paid and protections on how synthetic data can be used.”
Price could be the linchpin for a deal to happen. Ultimately, one source said, studios won’t bother to do a deal unless the price is high enough to make it worth a studio’s while, including taking on risks from exacerbating competition and industry tensions over gen AI up to breaching preexisting contracts.
“What’s the market clearing price for this? And second, for what? What are you going to give up to get money? It’s got to be enough to matter,” a source told VIP+. “[Studios are] not going to do this for a few bucks a minute. It doesn’t move the needle. If there are piles of money, that will lead people to find solutions. All these issues are not worth fighting through if you’re just dividing up like 10 bucks. If you’re dividing up 100 million bucks, it’s worth it to figure it all out.”
Other pieces on this topic: