Economy & Markets
Eye On The Market podcast
Subscribe today and never miss an episode
Join Michael Cembalest, Chairman of Market and Investment Strategy, as he explores a wide variety of investment topics, including the economy, policy and markets. Your subscription, using your preferred podcast provider, will feature new episodes with the release of each Eye on the Market publication.
Listen on Apple Podcasts
Listen on SpotifyThe winter of our discontent
Good morning, everybody. This is Michael Cembalest with the November 2025 Eye on the Market podcast. This one’s called the “Winter of Our Discontent,” and the topic is generative AI disrupting the content mode in the entertainment industry. It’s going to be a very visually oriented presentation, so you might want to watch the video version of this podcast instead of just listen to audio one.
As a quick note, on December 2nd, in about a month, the, every two years we write the Eye on the Market Alternative Investments, which covers buyout venture, hedge funds, private credit and private real estate secondary funds, evergreen funds. So that comes out in the first week of December.
So the cover art for this piece is a picture of Richard the Third sitting on top of a crumbling content moat. Richard the Third was the one who uttered the phrase in Shakespeare’s play about him, “the winter of our discontent.” And actually, Richard, who died in the 1400s, was found his, there were his remains were found in a parking lot in Leicester, England, and in 2012. And they dug him up and gave him a proper burial at Leicester Cathedral. And that was interesting to note.
Anyway, let’s get started. So the prior decade was defined by disruption in content distribution. And the next decade is going to be defined by disruption in content creation, which creates a combination of risks and opportunities, maybe more risks for some of the legacy media companies. So the, the, originally, the content distribution business was a moat because cable fiber, satellites, all those things, were very expensive, and once they become, came unbundled from the infrastructure, that distribution moat fell, and the impacts were seismic.
Believe it or not, back in 2007, Netflix offered streaming as a free add-on, and has since become the most powerful company in Hollywood, and while the video-related profits of the big media companies fell by almost 40% after 2018. And what we’re going to be talking about is the impact of generative AI on the content production moat because parts of that are crumbling as well.
So, and here you can see the impact of the disruption in content distribution for the major studios. The five of them, suffering major direct consumer streaming earnings losses from 2019 to 2023, while Netflix is booking billions in profits. And here you can see, I thought this was interesting, if you look at the top 100 streaming titles in 2023, the Netflix share of originals went down, but Netflix’s overall market share and streaming in these titles stayed the same because other legacy studios were willing to license their best content to Netflix. It’s another sign of just how powerful Netflix has become.
Now, while the Netflix share of TV viewing time, you know, which includes streaming and then legacy TV plus broadcast cable. So while the Netflix share is growing, other kinds of streaming are now growing faster, and that would be obviously things like YouTube, and streaming as a whole now exceeds legacy broadcast plus cable for the very first time as a share of overall viewing. And the, the reason why the competition and the stakes are so high is the overall video pie is not growing. If you look at the revenue that comes from all of traditional TV box office, home entertainment and streaming, that pie has really not changed at all since year 2018. All that’s happened is that streaming is taking a larger share of it. And, you know, the, the disturbing signs for legacy film and TV, some of you may have seen this already. I’m an old person, I’m 63, but for people under the age of 50, just in the last 10 years, their viewing of legacy TV programing has, excluding sports, as absolutely collapsed. And for the for the 2-to-age-17 bracket, age 2 to 17, it’s down over 80%. Legacy TV, which again is referring to broadcast cable exports since 2015. And then, with respect to the box office, obviously the film industry was a pandemic-affected sector, but box office receipts and tickets sold, while they’ve rebounded, are still way below 2019 levels.
And so I put this chart together because I thought it was notable. These are all the recovery rates since 2019 of pandemic-affected sectors, and the film industry is at the bottom, they’ve been exceeded by theme park attendance, recovery, hotel occupancy, museum attendance, Broadway box office receipts in attendance. Airline passenger miles. Top 100 concert tour tickets. Restaurant revenues. So the film industry is at the bottom of the stack here in terms of a post-2019 recovery.
And all the factors that we’ve just talked about partially explains why most of the media stocks, with the exception of Netflix and Fox, but when you look at the stocks for Disney, Warner Brothers, Comcast and Paramount, they’re kind of flat or over the last three years, if not down. So there’s a lot of things working against some of these companies right now.
Now, I think we talked about the distribution moat having collapsed. It looks like the content moat is still pretty steep, right? I mean, we have some data here that shows the top grossing films at the box office still cost you somewhere between $100 and $250 million to produce, and that’s been the case for several years. And then TV shows like, I haven’t seen any of these, but WandaVision and Hawkeye and The Rings of Power and Stranger Things—I don’t even know if I’m pronouncing it right because I’ve never watched them. These routinely cost like $25 million an episode, so the traditional content moat looks pretty steep using this traditional lens. The issue is that there are new platforms that are competing for viewer attention that don’t require anywhere near that kind of investment, and the democratization of being able to produce high-quality content is growing.
So if you look now at, at the latest data from August of this year, of, of media companies’ share of television screen time, that includes streaming, cable and broadcast TV, this is just TVs. This isn’t even mobile. YouTube is already, YouTube is already ahead of Disney and Netflix and the rest of the studios. So again, this isn’t even mobile. This is just on your physical TV set in your home. YouTube’s already number one. Just kind of amazing. And then, even in news, I mean, this, this, this unfortunately explains a lot about why the country’s in a position it’s in, in my opinion. But even in news, more respondents to this Nieman survey said they get their news from social media and video networks rather than from, God forbid, print, online news sites or traditional TV. So that’s kind of amazing and shows you how much things have changed.
Now what, what’s happening that is really kind of dangerous but also enticing for the average legacy media company, is that acceptance of user-generated content is growing. People are getting more comfortable with less polished stuff. And one way that we look at this is to look at the creators’ share of global media revenue has doubled in in the last few years, from about 7%, 14%. And these were some interesting numbers that I, that I read that I thought were interesting. Hollywood put out about 15,000 hours of TV and film last year. There were 300,000,000 hours of content uploaded to YouTube. The vast majority of that is junk. But even if just 0.0 1% of all of that YouTube content were just as interesting to you as the, as what Hollywood put out, that’s twice as much content as Hollywood’s annual output. So it, it only takes a very tiny sliver of all this slop that’s put on YouTube to create a video competitor challenge for, for Hollywood.
And the same thing is happening in the music industry. Over the last few years, the share of Spotify music streams from people that were either independent creators or not signed to a major label doubled from like 13 to over 26%. So people are getting more comfortable with user-generated content. And here’s another one: Roblox again, another thing I don’t have much experience with, but Roblox users in the U.S. continue to grow even as the gamer, overall gaming is flat. So gamer shares can be measured in hours played or as a percentage of population. U.S. gaming is basically flat to a few years ago and is down since COVID, but Roblox you, you’re going up. Why is that important? Roblox is made up of mostly millions of lower-fidelity user-designed games. So another example of how people are getting comfortable, very comfortable with user-designed stuff alongside more polished stuff.
So then before we get into, this text to video, we have this image here. This is from Midjourney, and they are a text-to-image program, and just look at the progress they made from 2002 to 2004, in, in, in around two years. These are all pictures of somebody you’ll recognize if you’ve seen the, the movies. The first picture here is barely recognizable and reminds me of David Lynch’s Eraserhead. And then by the time you get to the picture from July 2024, you know exactly who you’re looking at. And what’s happened since then is our text-to-video programs that are remarkably faithful to life. So yeah, I’m interested to see how many of you that are watching this podcast recognize the face of that person I just put up. That is not a real young woman. That thing is an AI-generated actress named Tilly Norwood. And she’s currently actually seeking talent representation. It was, she was, she, it, I don’t know, it was created by Particle6, which is an AI firm, and is an example of text to video. And so these new tools like peak and runway stable diffusion, AI obviously with Sora are making it easier for new entrants to create really high-quality content for digital and TV platforms. And in 2024, these models were first able to match human capabilities in visual commonsense reasoning. And they, and they’ve been, you know, obviously getting better ever since.
And the same way that you see scoring benchmarks for language models like Gemini versus GPT versus Grok or whatever, there are now benchmarks for scoring the accuracy of text-to-video generation models. And one of them is called VBench. And we have this polar chart here. What’s interesting about the polar chart is just to look at all the different ways that they measure the quality of these programs. This one is measuring OpenAI’s Sora 1.0 against three different Chinese versions. And there’s motion rationality, camera motion, complex landscaping, understanding human interactions, multi-view consistency, clothing, human anatomy. These are all things that matter if you’re scoring the, the accuracy of a video generation program. And of course, there are closed-source versions of this, like OpenAI’s Sora, but there’s also open-source versions of these text-to-video programs, some of which now claim that they can create videos at comparable quality to OpenAI at 10% of the training cost.
Now the changes that are taking place here are very rapid. There are. And now there’s OpenAI Sora 2.0, LTX, VEO3.1, and these programs are eclipsing the capabilities of what were the leading-edge video generation models just a few months ago. And we include here from October a, the latest leaderboard, from users, of the of the best text-to-video program. So you can see that OpenAI and Google are on top.
Now, one of the things I was interested in as I was doing some research on this one, was do these text-to-video models actually understand physical principles, or they’re just engaging in simple prediction? And so the, the guys at Google DeepMind unsurprisingly decided to dig into this one and see how good are these models at really understanding physics, the physical world, like fluid dynamics, optics, engineering mechanics, magnetism, thermodynamics. And so they created a bunch of tasks for them to do, dropping, in video, like just text prompt, hey, let me see a claw dropping items of different weights into a pillow. Let me see what happens when you drop a match into a glass of water. Let me see what happens if you put a paintbrush against the glass and keep going eventually. How does that paint, the ability for the brush to put paint on the glass dissipate? And as you can see, the scores here are still low there. The, these models scored only a 10 to 30% in terms of their understanding of physical principles. But they’re optimistic about the future. And they think that as these models get trained on larger and more diverse video sets, their understanding of real-world physics will continue to get better.
Now, the, the implications for legacy, legacy film studios, who have to compete with all this new content, are not all negative. The, there was a Bain study last year at the, where the time and cost savings that the legacy studios themselves could achieve from more virtual AI-generated production depends on the kind of thing. For comedies, it was 5 to 10%, family dramas, sci-fi was closer to 15 to 20%. And there are, they can use generative AI and post-production colorization, motion capture, repetitive tasks like rotoscoping and background removal, effects like fires or crowds. And then one of the things I thought was really interesting was the use of these LED volume walls, which are essentially a large wraparound digital wall that you combine with actual physical set elements. And then the background is, is rendered in real time using a generative AI program. And we have a picture of an example of them. And look how lifelike that background is, set against the, you know, the physical space of the volume wall.
The challenge, though, is maybe the studios are able to shave 10 to 15% from their production costs for film and TV. But if viewers continue to migrate away from these legacy film and TV to alternative forms of content, that’s going to be a challenge. And one of the things that’s already happened is that digital video has overtaken traditional TV attention span of the average adult,
And amazingly social video alone, YouTube, TikTok and Instagram, now represent around 25% of total daily screen time. And I know for a lot of young people, it’s actually much higher than that. In other words, of all the video consumed in a day, how much is legacy TV? How much is, is either free or subscription video on demand? And how much is social media?
In terms of TikTok, I will say I am, I, I have a TikTok account and, you know how your feed gets curated. I have, there’s a picture on the screen here of a sad dog in the rain. I’m a sucker for these videos where people rescue these dogs in the rain, and then they show them getting a new home and they clean them up, and they give them medicine, and they’re playing in the yard. And then you click on it and there’s a link to an Amazon account where you can buy things for dogs. And Rachel is convinced that these are all scams, but I just can’t help myself. And I, I donate lots of money to rescue TikTok dogs.
Now the, the, the challenge again, another way of looking at this is that all these new platforms are deflationary. And this, this isn’t new, right? But this has happened with technology all along. But I just wanted to show you some and, some numbers on what I mean. So Doug Shapiro, who is, is, an excellent media analyst, has done some work on this. And based on his numbers, he calculates monetization rates per hour, whether it’s video gaming or music. So, for example, you make $0.54 an hour off legacy linear video. Netflix brings that number down to $0.37 and then YouTube to $0.19. And then look at music. A CD generates revenue of $0.67 per hour. Spotify cuts that by 90% to $0.06. So all of these things are deflationary as it relates to the amount of money that gets generated.
Now, just to wrap up, I did speak to a couple of people in the film industry and they reiterated that storytelling still matters. And I said, explain how? And they said, well, look at the amount of money that people are paying for these legacy film and TV libraries. Private equity firms have been involved, and there’s been a lot of strategic acquisition. And it’s true. Amazon, when they were acquired MGM, disclosed that it valued MGM’s film and TV library at $3.5 billion. Disney, which has paid over $80 billion for Pixar, Marvel, Lucasfilm and Fox, a lot of that went for the value of the film libraries. Viacom’s purchase of a stake in Miramax library act—was specific library acquisitions by Lantern, by Lionsgate, Raven. And earlier this year, Lionsgate actually reported an operating loss, but at the same time, record-high film library revenues. And now the latest news is that Netflix is, is exploring a potential acquisition of Warner Brothers. And that was libraries including Harry Potter, the DC comics and things like Game of Thrones. So storytelling still matters. And, you know, I was able to find evidence of that. If you look at the movie share of streaming revenues, they’ve gone up, interestingly, from 2022 to 2024.
Let’s use Amazon Prime Video as an example. Movies doubled from 30 to 60% of total streaming revenue. So that was interesting. And then, when you look at TV libraries, what are they worth? One way of thinking about it is what do people watch when they stream TV episodes? And so we have a table in here that shows the top 20 in the U.S., NCIS, Gray’s Anatomy, Squid Game. You know, I, I haven’t seen most of these shows, but Gunsmoke made the list. Little House on the Prairie made the top 20 list last year. So some of these legacy TV shows still have a lot of value in terms of getting eyeballs. And we show another analysis on a global level of, of the most popular streaming shows on a global basis. And I have to admit, I’ve never seen a single one of these shows: Gray’s Anatomy, Prison Break, Lost, Big Bang Theory, Dexter, the Resident, Gilmore Girls, Suits, Supernatural and Friends. Never seen it.
Okay, so just to wrap up, I do hope that the content moat survives a little bit longer. I, I’ve never seen any of those TV shows, but I’m a big, I’m a big film watcher, and I actually have a Letterboxd account, and I put a link to it in the, in the Eye on the Market file, the PDF or the HTML, whichever one you’re looking at. And I included a table of all the films that I ranked at a four-and-a-half or a five in the 21st century. There’s about 30 of them. The most recent one was a film called The Ballad of Wallis Island. Really good movie. And, so anyway, if you want to, if you want a list of, of good films to see in my, you know, following my recommendations, you could look at the table and, and when I looked at the ratio of worldwide box office receipts to the production budgets of these films, they were positive. Now, a lot of those receipts have to get paid to distributors. I understand that, but there does still seem to be a market for this kind of thing.
Anyway, thank you very much for listening. And I will, I will see you all again in early December for the biennial Eye on the Market Alternative Investments. Thanks for listening.
(DESCRIPTION)
Title card: JP Morgan, Eye on the market. JP Morgan. Presentation. Title: November 2025. The winter of our discontent. Generative AI disrupts the entertainment industry content moat. Image: A man wearing a golden crown and a brown robe sits on the edge of a crumbling stone wall, resting his chin on his hand. Part of the wall has collapsed into rubble below. A video box on the right. Michael Cembalest has short hair, glasses, and wears a dark zip-up jacket with light blue trim. He sits in front of a virtual modern shelf that holds vases, gold decorative items, books, and geometric objects.
(SPEECH)
[SWOOSHING]
Good Morning, everybody. This is Michael Cembalest with the November 2025 Eye on the Market podcast. This one's called The winter of our discontent, and the topic is generative AI disrupting the content mode in the entertainment industry. It's going to be a very visually oriented presentation, so you might want to watch the video version of this podcast instead of just listening to the audio one.
As a quick note, on December 2, in about a month, the every two years we write the Eye on the Market alternative investments review, which covers buyout venture, hedge funds, private credit, and private real estate, secondary funds, evergreen funds, so that comes out in the first week of December.
So the cover art for this piece is a picture of Richard III sitting on top of a crumbling content book. Richard III was the one who uttered the phrase in Shakespeare's play about him, the winter of our discontent. And actually Richard, who died in the 1400s, his remains were found in a parking lot in Leicester, England, in 2012. And they dug him up and gave him a proper burial in Leicester cathedral. That was interesting to note.
Anyway,
(DESCRIPTION)
Slide: The two entertainment moats.
(SPEECH)
let's get started. So the prior decade was defined by disruption in content distribution, and the next decade is going to be defined by disruption in content creation, which creates a combination of risks and opportunities, maybe more risks for some of the legacy media companies. So originally, the content distribution business was a moat because cable, fiber, satellites, all those things were very expensive.
And once they become came unbundled from the infrastructure, that distribution moat fell, and the impacts were seismic. Believe it or not, back in 2007, Netflix offered streaming as a free add on and has since become the most powerful company in Hollywood while the video-related profits of the big media companies fell by almost 40% after 2018. And what we're going to be talking about is the impact of generative AI on the content production mode because parts of that are crumbling as well.
And
(DESCRIPTION)
Slide: Disruption in content distribution. A line graph compares Netflix net income and major studios’ direct-to-consumer losses from 2019 to 2023. Netflix shows steady growth above zero, while studios’ losses deepen until 2022 before improving slightly in 2023. Text: Source: The Wrap, 2024.
(SPEECH)
here you can see the impact of the disruption in content distribution, the major studios, the five of them suffering major direct to consumer streaming earnings losses from 2019 to 2023, while Netflix is booking billions of profits. And here you can see-- I thought this was interesting.
If
(DESCRIPTION)
Slide. A stacked bar chart tracks viewing time of the top 100 streaming titles in 2023, divided into Netflix originals, exclusive acquisitions, nonexclusive acquisitions, and other non-Netflix platforms. Netflix’s share declines from 34% in January to 20% in December. Text: Source: The Hollywood Reporter, January 2024.
(SPEECH)
you look at the top 100 streaming titles in 2023, the Netflix share of originals went down. But Netflix's overall market share in streaming in these titles stayed the same because other legacy studios were willing to license their best content to Netflix. It's another sign of just how powerful Netflix has become.
(DESCRIPTION)
Slide: While Netflix share is growing, all other streaming is growing faster. A stacked area chart shows the share of US TV viewing time from January 2023 to July 2025 divided into Netflix, all other streaming, legacy broadcast and cable, and other. Streaming categories grow while legacy broadcast and cable decline. Text: Source: Netflix, Nielsen, Q3 2025.
(SPEECH)
Now, while Netflix share of TV viewing time, which includes streaming and then legacy TV, which is broadcast cable, so while the Netflix share is growing, other kinds of streaming are now growing faster. And that would be obviously things like YouTube. And streaming as a whole now exceeds legacy broadcast plus cable for the very first time as a share of overall [INAUDIBLE].
And
(DESCRIPTION)
Slide: Fierce competition and high stakes, the overall video pie is not growing. A stacked bar chart compares US video revenue from 2011 to 2024, dividing totals between streaming and traditional TV, box office, and home entertainment. Streaming grows steadily each year while traditional revenue declines. Text: Source: Doug Shapiro and others, May 2025.
(SPEECH)
the reason why the competition and the stakes are so high is the overall video pie is not growing. If you look at the revenue that comes from all of traditional TV, box office, home entertainment, and streaming, that pie has really not changed at all since year 2018. All that's happened is that streaming is taking a larger share of it.
(DESCRIPTION)
Slide: Disturbing Signs for Legacy Film/TV. A line graph tracks viewing of legacy TV programming excluding sports from 2015 to 2023 by age group. Viewing declines across all groups, with the steepest drop among ages 2–17 and the smallest among ages 50+. Text: Source: Doug Shapiro, Nielsen, MoffettNathanson, May 2025.
(SPEECH)
And the disturbing signs for legacy film and TV, some of you may have seen this already. I'm an old person. I'm 63. But for people under the age of 50, just in the last 10 years, their viewing of legacy TV programming has, excluding sports, has absolutely collapsed.
And for the 2 to age 17 bracket, age 2 to 17, it's down over 80% legacy TV, which, again, is referring to broadcast and cable, exports since 2015. And
(DESCRIPTION)
Slide. A dual-axis line graph tracks US box office receipts and tickets sold from 1995 to 2025. Both metrics rise until the late 2010s, drop sharply around 2020, and partially recover afterward. Text: Source: Nash Information Services, 2025 figures are annualized.
(SPEECH)
then with respect to the box office, obviously the film industry was a pandemic-affected sector, but box office receipts and tickets sold, while they rebounded, are still way below 2019 levels.
And so I put this chart together because I thought it was notable.
(DESCRIPTION)
Slide. A horizontal bar chart compares recovery rates since 2019 across industries. US restaurant revenues and hotel rates exceed 100%, while film industry metrics and NYC mass transit remain below 80%. Text: Source: JPMAM, 2025.
(SPEECH)
These are all the recovery rates since 2019 of pandemic-affected sectors, and the film industry is at the bottom. They have been exceeded by theme park attendance recovery, hotel occupancy, museum attendance, Broadway box office receipts and attendance, airline passenger miles, top 100 concert tour tickets, restaurant revenues. So the film industry is at the bottom of the stack here in terms of a post-2019 recovery.
And
(DESCRIPTION)
Slide: Underperformance of most media stocks. A line graph tracks stock price performance of major media companies and market indices from 2022 to 2025. Netflix rises sharply above all others, while Disney, Fox, Comcast, Warner Brothers Discovery, and Paramount remain lower. Text: Source: Bloomberg, JPMAM, October 24, 2025.
(SPEECH)
all the factors that we just talked about partially explain why most media stocks, with the exception of Netflix and Fox, but when you look at the stocks for Disney, Warner Brothers, Comcast, and Paramount, they're kind of flat or over the last three years, if not down. So there's a lot of things working against some of these companies right now.
Now,
(DESCRIPTION)
Slide: The content moat still looks steep using a traditional lens. A line graph tracks median budgets of the top 10, top 20, and top 50 highest-grossing US films from 2000 to 2025. Budgets rise over time with peaks around 2010 and 2019, then drop slightly. Text: Source: Stephen Follows Film Industry Research, October 25, 2025.
(SPEECH)
we talked about the distribution mode having collapsed. It looks like the content moat is still pretty steep. We have some data here that shows the top grossing films at the box office still cost you somewhere between $100 to $150 million to produce, and that's been the case for several years. And then TV shows like-- I haven't seen any of these-- but WandaVision and Hawkeye and The Rings of Power and Stranger Things-- I don't even if I'm pronouncing it right because I've never watched them-- these routinely cost like $25 million an episode.
So the traditional content moat looks pretty steep using this traditional lens. The issue is that there are new platforms that are competing for viewer attention that don't require anywhere near that kind of investment, and the democratization of being able to produce high-quality content is growing. So
(DESCRIPTION)
Slide. A bar chart compares media companies by share of total television screen time in August 2025. YouTube leads at about 13%, followed by Disney, Netflix, and NBCUniversal, with smaller shares for Fox, Paramount, and others. Text: Source: Nielsen Media Distributor Gauge, August 2025.
(SPEECH)
if you look now at the latest data from August of this year of media companies' share of television screen time-- that includes streaming, cable, and broadcast TV.
This is just TVs. This isn't even mobile. YouTube's already ahead of Disney and Netflix and the rest of the studios. So again, this isn't even mobile. This is just on your physical TV set and your home. YouTube is already number one, which is kind of amazing.
(DESCRIPTION)
Slide: News. A line graph tracks US news consumption by source from 2013 to 2025. Social and video networks rise to surpass TV, while online news sites decline slightly, and print, radio, and podcasts remain low. Text: Source: NiemanLab, Reuters Institute, June 2025.
(SPEECH)
And then even in news, this unfortunately explains a lot about why the country's in position it's in, in my opinion. But even in news, more respondents to this Nieman survey said they get their news from social media and video networks rather than from, God forbid, print, online news sites, or traditional TV. So that's kind of amazing and shows you how much things have changed.
(DESCRIPTION)
Slide: Acceptance of UGC is growing. User-generated content. A bar and line chart compare corporate and creator media revenue worldwide from 2019 to 2023. Corporate media dominates, while creator media and its share rise steadily to about 14%. Text: Source: Doug Shapiro, May 2025.
(SPEECH)
Now, what's happening that is really dangerous but also enticing for the average legacy media company is that acceptance of user-generated content is growing. People are getting more comfortable with less polished stuff. And one way that we look at this is to look at the creator share of global media revenue has doubled in the last few years, from about 7% to 14%.
And these were some interesting numbers that I read that I thought were interesting. Hollywood put out about 15,000 hours of TV and film last year. There were 300 million hours of content uploaded to YouTube.
Now, the vast majority of that is junk. But even if just 0.01% of all of that YouTube content were just as interesting to you as what Hollywood put out, that's twice as much content as Hollywood's annual output. So it only takes a very tiny sliver of all this slop that's put on YouTube to create a video competitor challenge for Hollywood.
And the same thing is happening in the music industry. Over the last few years, the share of Spotify music streams from people that were either independent creators or not signed to a major label doubled from 13% to over 26% So people are getting more comfortable with user-generated content.
And here's another one.
(DESCRIPTION)
Slide: Roblox is made up of 44mm lower fidelity user-designed games. A line and scatter plot compare the share of US gamers and Roblox daily active users from 2018 to 2025. Gamer share rises steadily to over 80%, while Roblox users increase to about 20 million. Text: Sources: Roblox, Matthew Ball, TechCrunch, 2025.
(SPEECH)
Roblox, again, another thing I don't have much experience with, but Roblox users in the US continue to grow even as the overall gaming is flat. So gamer shares can be measured in hours played or as a percentage of population. US gaming is basically flat to a few years ago and is down since COVID.
But Roblox is going up. Why is that important? Roblox is made up of mostly millions of lower-fidelity user-designed games. So another example of how people are getting comfortable, very comfortable with user-designed stuff alongside more polished stuff.
So
(DESCRIPTION)
Slide: Text to image came first. A sequence of six AI-generated portraits depicts the same fictional young man, Harry Potter, with round glasses evolving in realism and detail from February 2022 to July 2024. Each version becomes more lifelike and cinematic, with V6.1 appearing most refined. Text: J.P. Morgan.
(SPEECH)
then before we get into this text to video, we have this image here. This is from Midjourney, and they are a text-to-image program. And just look at the progress they made from 2002 to 2004. In around two years-- these are all pictures of somebody you'll recognize if you've seen the movies-- the first picture here is barely recognizable and reminds me of David Lynch's Eraserhead. And then by the time you get to the picture from July 2024, you know exactly who you're looking at.
And what's happened since then is our text-to-video programs that are remarkably faithful to life--
(DESCRIPTION)
Slide: Then text to video. Image: A woman with long brown hair wearing a dark top sits indoors and waves with a smile. Behind her, large posters with faces and text line a glass wall. Text: AI text-to-video tools like Pika, Runway, Stable Diffusion and Sora are making it easier for new entrants to create high quality content for both digital and TV platforms. In 2024, AI models were first able to match human capabilities on Visual Commonsense Reasoning tasks, and text-to-video models are building on that progress.
(SPEECH)
so I'm interested to see how many of you that are watching this podcast recognize the face of that person I just put up. That is not a real young woman. That thing is an AI-generated actress named Tilly Norwood, and she's currently actually seeking talent representation. She, it, I don't know, was created by Particle6, which is an AI firm, and is an example of text to video.
And so these new tools like Pika and Runway, Stable Diffusion and AI, obviously, with Sora are making it easier for new entrants to create really high-quality content for digital and TV platforms. And in 2024, these models were first able to match human capabilities in visual common sense reasoning. And they've been obviously getting better ever since.
(DESCRIPTION)
Slide: With new benchmarks for storing them. A polar chart compares video generation models OpenAI Sora 1.0, Hunyuan, CogVideoX-1.5, and Kling 1.6 across categories like human anatomy, motion rationality, composition, and diversity. OpenAI Sora 1.0 scores strongly across most areas. Text: Source: Open-Sora, HPC-AI Tech, March 2025.
(SPEECH)
And the same way that you see scoring benchmarks for language models like Gemini versus GPT versus Grok or whatever, there are now benchmarks for scoring the accuracy of text-to-video generation models. And one of them is called VBench, and we have this polar chart here. What's interesting about the polar chart is just to look at all the different ways that they measure the quality of these programs.
This one is measuring OpenAI's Sora 1.0 against three different Chinese versions. And there's motion rationality, camera motion, complex landscaping, understanding human interactions, multiview consistency, clothing, human anatomy. These are all things that matter if you're scoring the accuracy of a video-generation program.
And of course, there are closed-source versions of this, like OpenAI's Sora. But there's also open-source versions of these text-to-video programs, some of which now claim that they can create videos at comparable quality to OpenAI at 10% of the training cost.
Now,
(DESCRIPTION)
Slide: The latest leaderboard, changes are very rapid. A leaderboard table ranks text-to-video models by score, with Google’s Veo 3.1 and Sora 2.0 leading, followed by other versions from Minimax, Kling, and Runway. Text: OpenAI’s Sora 2.0, LTX, and Veo 3.1 are already eclipsing the capabilities of what were leading-edge video generation models just a few months ago.
(SPEECH)
the changes that are taking place here are very rapid. And now there's OpenAI Sora 2.0, LTX, VEO 3.1, and these programs are eclipsing the capabilities of what were the leading-edge video-generation models just a few months ago. And we include here from October the latest leaderboard of from users of the best text-to-video programs, and you can see that OpenAI and Google are on top.
(DESCRIPTION)
Slide: Do AI video models simply engage in pixel prediction, or are they learning the laws of physics? A bar chart compares video generative models by Physics-IQ score, with VideoPoet (multiframe) scoring highest and OpenAI Sora 1.0 lowest. Text: Google DeepMind, February 2025, notes that models trained on larger and more diverse videos will better understand real-world physics concepts like dynamics and thermodynamics.
(SPEECH)
Now, one of the things I was interested in as I was doing some research on this one was, do these texts-to-video models model actually understand physical principles? Or they're just engaging in simple prediction? And so the guys at Google DeepMind, unsurprisingly, decided to dig into this one and see, how good are these models at really understanding the physical world, like fluid dynamics, optics, engineering, mechanics, magnetism, thermodynamics?
And so they created a bunch of tasks for them to do, dropping in video, just text prompt. Hey, let me see a claw dropping items from different weights into a pillow. Let me see what happens when you drop a match into a glass of water. Let me see what happens if you put a paintbrush against the glass and keep going. Eventually how does the ability for the brush to put paint on the glass dissipate?
And as you can see, the scores here are still low there. These models scored only a 10% to 30% in terms of their understanding of physical principles. But they're optimistic about the future, and they think that as these models get trained on larger and more diverse video sets, their understanding of real world physics will continue to get better.
(DESCRIPTION)
Slide: The implications for legacy film studios are not all negative. A bar chart shows time and cost savings from virtual production across comedy, family, and sci-fi films, with larger budgets achieving up to 25% savings. Text: Source: Bain & Company Virtual Production Analysis, 2023, notes benefits from automation, motion capture, and LED volume use.
(SPEECH)
Now, the implications for legacy film studios, who have to compete with all this new content, are not all negative. There was a Bain study last year, where the time and cost savings that the legacy studios themselves could achieve from more virtual AI-generated production depends on the kind of thing. For comedies, it was 5% to 10%. Family dramas, sci-fi was closer to 15% to 20% And they can use generative AI in post-production colorization, motion capture, repetitive tasks like rotoscoping and background removal, effects like fires or crowds.
And then one of the things I thought was really interesting was the use of these LED volume walls, which are essentially a large wraparound digital wall that you combine with actual physical set elements, and then the background is rendered in real time using a generative AI program. And
(DESCRIPTION)
Image: A person walks across a large indoor studio toward a curved LED wall displaying a bright sky and mountain landscape. Overhead lighting rigs hang from the ceiling, and a camera on a dolly stands nearby.
(SPEECH)
we have a picture of an example of them, and look how lifelike that background is set against the physical space of the volume wall.
(DESCRIPTION)
Slide: Viewers may continue to migrate to alternative forms of content on platforms. A bar chart compares average daily viewing time of US adults from 2021 to 2025, showing digital video increasing steadily while traditional TV declines. Text: Source: Insider Intelligence, eMarketer, 2024, notes digital video has surpassed TV as AI-driven content improves efficiency and quality.
(SPEECH)
The challenge, though, is maybe the studios are able to shave 10% to 15% from their production costs for film and TV. But if viewers continue to migrate away from these legacy film and TV to alternative forms of content, that's going to be a challenge. And one of the things that's already happened is that digital video has overtaken traditional TV attention span of the average adult.
And
(DESCRIPTION)
Slide: A stacked bar chart shows 2024 daily video consumption by type, with legacy TV, subscription video, free ad-supported streaming, and social video. Social video accounts for the largest portion at about 25%. Text: Source: Doug Shapiro, Maverix Insights MIDG data, Nielsen, 2024. A small image shows a dog sitting on a wet road as cars approach.
(SPEECH)
amazingly, social video alone, YouTube, TikTok, and Instagram, now represent around 25% of total daily screen time. And I know for a lot of young people, it's actually much higher [INAUDIBLE]. Of all the video consumed in the day, how much is legacy TV? How much is either free or subscription video on demand? And how much is social media?
In terms of TikTok, I will say I have a TikTok account, and you know how your feed gets curated? There's a picture on the screen here of a sad dog in the rain. I'm a sucker for these videos where people rescue these dogs in the rain.
And then they show them getting a new home. And they clean them up, and they give them medicine. And they're playing in the yard. And then you click on it, and there's a link to an Amazon account where you can buy things for the dog.
And Rachel's convinced that these are all scams, but I just can't help myself. And I donate lots of money to rescue TikTok dogs.
(DESCRIPTION)
Slide: Alternative platforms are generally deflationary. A table compares monetization rates per hour across video, music, and gaming categories. Linear video earns the most at $0.54 per hour, followed by CDs, Netflix, and YouTube, while Spotify and gaming on PC or mobile yield the lowest rates. Text: Sources include Doug Shapiro and others.
(SPEECH)
Now, the challenge, again, another way of looking at this, is that all these new platforms are deflationary. And this isn't new. This has happened with technology all along. But I just wanted to show you some numbers of what I mean.
So Doug Shapiro, who is an excellent media analyst, has done some work on this. And based on his numbers, he calculates monetization rates per hour, whether it's video gaming or music. So, for example, you make $0.54 an hour off legacy linear video. Netflix brings that number down to $0.37 and then YouTube to $0.19.
And then look at music. A CD generates revenue of $0.67 per hour. Spotify cuts that by 90% to $0.06. So all of these things are deflationary as it relates to the amount of money that gets generated.
Now,
(DESCRIPTION)
Slide: Storytelling still matters, which is why film/TV libraries are worth so much. A bullet point list.
(SPEECH)
just to wrap up, I did speak to a couple of people in the film industry, and they reiterated that storytelling still matters. And I said, explain how. And they said, well, look at the amount of money that people are paying for these legacy film and TV libraries. Private equity firms have been involved, and there's been a lot of strategic acquisition.
And it's true. Amazon, when they acquired MGM, disclosed that it valued MGM'S film and TV library at $3.5 billion. Disney, which has paid over $80 billion for Pixar, Marvel, Lucasfilm, and Fox, a lot of that went to the value of the film libraries. Viacom's purchase of a stake in Miramax, specific library acquisitions by Lantern, Vine, Lionsgate, Raven, and earlier this year, Lionsgate actually reported an operating loss but, at the same time, record-high film library revenues.
And now the latest news is that Netflix is exploring a potential acquisition of Warner Brothers. And those libraries include Harry Potter, the DC Comics, and things like Game of Thrones. So storytelling still matters.
And
(DESCRIPTION)
Slide: Movies aren't dead yet. A line graph compares movie streaming revenue shares from 2022 to 2024 for Disney+, Amazon Prime Video, Netflix, and Warner Bros. Disney+ leads with steep growth, followed by Amazon Prime Video and Netflix, while Warner Bros remains lowest. Text: Source: Video Week, Parrot Analytics, 2025.
(SPEECH)
I was able to find evidence of that. If you look at the movie share of streaming revenues, they've gone up, interestingly. From 2022 to 2024-- let's use Amazon Prime Video as an example-- movies doubled from 30% to 60% of total streaming revenue. So that was interesting.
And
(DESCRIPTION)
Slide: And TV episode libraries are worth a lot to companies that own them. A table lists the top US and global streaming programs, with Bluey, Grey’s Anatomy, and NCIS leading US viewership in H1 2025, and Grey’s Anatomy, Prison Break, and Lost topping global rankings in H2 2024. Text: Sources: Nielsen Streaming Content Ratings, Digital, 2025.
(SPEECH)
then when you look at TV libraries, what are they worth? One way of thinking about it is, what do people watch when they stream TV episodes? And so we have a table in here that shows the top 20 in the US-- NCIS, Gray's Anatomy, Squid Game.
I haven't seen most of these his shows. But Gunsmoke made the list. Little House on the Prairie made the top 20 list last year. So some of these legacy TV shows still have a lot of value in terms of getting eyeballs.
And we show another analysis on a global level of the most popular streaming shows on a global basis. And I have to admit, I've never seen a single one of these shows. Gray's Anatomy, Prison Break, Lost, Big Bang Theory, Dexter, The Resident, Gilmore Girls, Suits, Supernatural, and Friends, never seen it.
So
(DESCRIPTION)
Slide: I hope the content moat remains in place a bit longer. A table lists highly ranked 21st-century films excluding documentaries, showing each title’s year, ranking, studio, and box office-to-budget ratio. Films like The Ballad of Wales Island, Anatomy of a Fall, and Aftersun appear near the top. Text: Source: Michael Cembalest Letterbook account.
(SPEECH)
just to wrap up, I do hope that the content moat survives a little bit longer. I've never seen any of those TV shows, but I'm a big film watcher. And I actually have a Letterboxd account, and I put a link to it in the Eye on the Market file, the PDF or the HTML, whichever one you're looking at.
And I included a table of all the films that I ranked at a 4 and 1/2 or a 5 in the 21st century. There's about 30 of them. The most recent one was a film called The Ballad of Wallis Island, really good movie. And so anyway, if you want a list of good films to see in my following my recommendations, you can look at the table.
And when I looked at the ratio of worldwide box office receipts to the production budgets of these films, they were positive. Now, a lot of those receipts have to get paid to distributors. I understand that. But there does still seem to be a market for this kind of thing.
Anyway, thank you very much for listening.
(DESCRIPTION)
Title card: November 2025. The winter of our discontent. Generative AI disrupts the entertainment industry content moat. Image: A man wearing a golden crown and a brown robe sits on the edge of a crumbling stone wall, resting his chin on his hand. Part of the wall has collapsed into rubble below.
(SPEECH)
And I will see you all again in early December for the biennial Eye on the Market alternative tribute. Thanks for listening. Bye.
[SWOOSHING]
(DESCRIPTION)
Logo: J.P. Morgan.
