(Welcome to the “Most Important Story of the Week”, my bi-weekly strategy column analyzing the most important (but often not buzziest) news story of the last two weeks. I’m the Entertainment Strategy Guy, a former streaming executive who now analyzes business strategy in the entertainment industry. Please subscribe.)
Well, sometimes, big news stories break and you have to pause your editorial calendar (mainly delay getting to the best-performing TV shows and films as part of the EntStrategyGuy 2025 Viewership Report) to cover said big news stories. This week…
- OpenAI shut down Sora 2, its LLM-powered video creation tool.
- Meta and YouTube were found liable in two landmark trials.
Those stories you might have heard about. Less notable, but just as important, was a third story:
- The Supreme Court overturned a verdict finding Cox Communications liable for not stopping piracy on its cable lines.
Wowza! That’s a lot of very important news. Luckily for me, I have an ongoing series, my “Four Horsemen of the Media-pocalypse” that two of those stories (Sora 2 and Cox) neatly fit into. Plus, Q1 is about to end anyway, so it’s a good time to look at theaters and cord-cutting. (Meta, YouTube, we’ll get to you later; those stories are not going away.)
So that’s the plan for today, we’re covering…
- Sora 2 and what this means for Hollywood and AI.
- The landmark Supreme Court decision on piracy.
- Some very good news for theaters (and some bad news).
- The continued good news for cord-cutting.
Let’s dive right in!
AI – OpenAI Shuts Down Sora 2. What This Means for Hollywood
Listen, I love a good Xanatos Gambit. For those not familiar, that’s a Gargoyles reference where someone sets up a situation where whichever choice their opponent makes, they still win either way. The inverse is, of course, the Kobayashi Maru, a Star Trek reference to a test Starfleet Academy members take where, no matter what they do, they lose.
I do hate it, though, when pundits use this line of thinking.
A few weeks back, I mentioned that I often read takes that come across as, “Heads Netflix wins, tails traditional entertainment loses”. But that’s not because Netflix set things up that way, but because Netflix super fans just interpret every news story that way.
Well, you can go ahead and lump AI into that dynamic as well.
Some folks basically treat every decision by OpenAI, Anthropic and their VC funders as genius (the Xanatos Gambit), and any move by Hollywood as hopelessly tired and doomed to failure (the Kobayashi Maru). Even when Disney invested in OpenAI last December, some AI boosters were upset that it implied the AI party was about to end.
And this week, the party did indeed end, at least for now.
Specifically, OpenAI shut down Sora 2, their text-to-video LLM, and that meant Disney ended their licensing deal with OpenAI. (A deal, by the way, which never actually closed, joining a long line of OpenAI announced-but-not-closed deals.)
I think it’s worth splitting this news event into two pieces. Today, let’s discuss what the closure of Sora 2 could mean for the future of AI-generated video in Hollywood. Then, in a future article, I’ll analyze the impact on Disney.
Note: This Article is About LLMs That Make Videos
One key thing to note before I go on: when I use “AI” in this article in broad terms, I do NOT mean machine learning models. I’m referring to large language models, or LLMs, but even more specifically, LLMs that turn text into video outputs. Sora 2 (and Nano Banana and SeeDance 2.0) are a specific type of LLM with specific outputs, namely video clips.
Often, articles about AI refer to companies that are not LLMs, but most people use “AI” to refer to LLMs and their ilk. Most common folk are hyped by LLMs and the feats they pull off, while they don’t interact much, if at all, with machine learning.
It’s a pinch of stolen AI valor, if you will. Machine learning has many applications in Hollywood, with about a tenth of the ethical issues or threats to labor. Heck, Weta uses an Avatar-sized amount of machine learning and has for years. This topic deserves its own deep dive in a future issue.
The One Absolute Key Issue: Costs
If I had one complaint about AI coverage since 2024, it’s that too few articles grappled with the costs associated with this technology. Even today, I read articles about AI companies (again, mostly LLMs), and, because they’re startups, no one mentions costs or profitability, with a general “Well, that’s for the future” handwaving.
Yet Sora 2 shows us that costs really, really matter!
Sora 2 was likely eye-wateringly expensive. The lone estimate comes from Forbes, who estimated that, at its peak, it cost OpenAI $15 million a day, at about $1.30 per video.
But even those numbers—again, the only ones I’ve seen—may be low. They rely on the $1.30 per video to be true, and not more expensive. Some reporting indicates that the quality of videos decreased over time on Sora 2, implying OpenAI had reduced the token spend to drive down costs.
To be super blunt on this, it’s possible that every single video made by an LLM costs a couple of dollars. For ten seconds of video. Ten seconds. Like you could buy a cup of coffee at Starbucks per video. The longer the video, the more it costs. Would twenty second videos cost, possibly, five times more than that? The more context required to make the video—for example, if you use a prior video and say make the actors look the same—the costs could go up even more. And yes, if you need to re-do an image and tweak it, that ups the costs even more.
These are the blunt costs on LLM-produced videos that Hollywood has to grapple with.
This isn’t, by the way, theoretical. Coca-Cola made this year’s holiday commercial with AI, and headlines put out that it required 100 people to generate 70,000 clips. At five dollars a clip, yeah, that’s a $350K commercial. (To be clear, I don’t know the actual costs of the commercial, but again, I doubt it was cheap.)
Now, do I think Coke paid that? No. Why not? Because a lot of AI use is subsidized by the AI companies trying to get into video, and they want folks to use it before raising prices. But the key is getting those costs under control.
(By the way, this likely extends to plenty of AI companies aggressively seeking to expand their business in Hollywood. This is why “these studios are using AI” lists should be taken with HUGE grains of salt.)
The challenge is that LLMs can’t seem to get costs under control. An LLM’s cost is a relatively simple formula: cost per token multiplied by the number of tokens required. The good news is inference costs seem (again, we have terrible data on it) to be coming down. Say 5% per year. The bad news? To improve models, the companies need to drastically increase the number of tokens used per answer. But not in percentage terms, but orders of magnitude. So instead of using 1,000 tokens, new answers take 10,000 tokens.
Do that quick math. (Just for illustration.) Say a million tokens cost $1, and to answer most questions, it requires the model to spend 10 million tokens. (These are just generalities, but this article has some token costs.) So, $10. Then assume costs go down to $0.095 per million. Great, 5% cost reduction! But the model requires an order of magnitude more tokens, so you use a billion tokens. That’s $95! A 9.5X increase. Thus, while the costs of inference head down, the total LLM costs go up. Based on Ed Zitron’s reporting, I think that’s happening.
Shutting Down Sora 2 Implies OpenAI Is In Trouble
I have to emphasize that my jaw dropped when I heard OpenAI shut down Sora 2 for good. Yes, I get that they need to cut costs to potentially IPO this year, but I still thought the reputational damage would stay their hand.
It also begs the question: did the service have a lot of users or not? Because then OpenAI was choosing between two, still bad options:
- The service had a ton of users, and they’d be very upset that it got shut down. or…
- The service had very few users, implying a ton of the coverage from 2025 was pure hype.
Neither is a good look!
Or it could be a combination third option: after the initial burst of interest, few folks used the Sora 2 app. But it was still excessively expensive. So it was both unpopular and a cost center.
Crucially, OpenAI also removed the capability to make videos from ChatGPT, and that really speaks to the likely cost problems for these models. Plus, customers have so little brand loyalty to AI companies that, if anyone raises prices even remotely, users will abandon them.
This also implies something is wrong with the reporting on OpenAI’s usage and fundraising.
OpenAI claims to have 900 million weekly average users AND that they raised $120 billion in funding in their latest fundraising round. Even if Sora 2 cost $15 million a day to run, they could easily afford to keep spending that kind of cash if they really do have $120 billion on hand and 900 million weekly customers to try to sell it to.
But OpenAI likely doesn’t have enough cash on hand to deal with Sora 2’s daily cash burn. That’s crazy for a company that claims to have raised $120 billion.
This all ties back to the Disney deal. Unfortunately, OpenAI has a history of announcing huge deals—including notably a massive investment from NVIDIA that didn’t happen—and then pretending they didn’t. And yeah, it happens a lot! I wrote an entire article about the Disney/OpenAI deal because I assumed it was real and would happen. Normally, I ignore stories about rumored deals, because they often don’t happen. This deal was announced as if it had closed, but like many OpenAI announcements, it was fictional.
Frankly, I don’t trust any of OpenAI’s (or Anthropic’s) numbers leaked to the press, and I won’t until they publish audited financial statements. If you work in Hollywood, you shouldn’t either. This Disney-OpenAI deal should convince all of Hollywood of that.
But this is the glimmer of good news for Tinsel Town. If OpenAI (and likely their ilk) can’t actually make viable products to rival Hollywood, then Hollywood productions will survive as the dominant form of feature length films in the near future.
The Solution to Bad AI Media Coverage: Insist on Knowing the Costs!
Going forward, the simple rule of thumb for any article you read that mentions AI is to ask yourself, “Did it mention the costs to run the service?”
Ignore any hype about how many customers an AI company claims it has, or how well the journalist using the (subsidized) model enjoyed it, and focus on costs. If the article ignores, hides or obscures how much it actually costs to use the model, then it isn’t grappling with the real issues with AI. If an article won’t say if customers have subsidized usage (meaning artificially low costs to hook them), then you shouldn’t care if the article says that lots of companies are using it.
If you read articles describing any company as “disrupting” Hollywood, and it doesn’t have costs, discount it. It’s vaporware.
Piracy – Traditional Studios and IP Rights Holders Lose Big
As I’ve written before, even more than AI, piracy is the biggest issue confronting the streamers and traditional Hollywood studios. If customers can easily find, download and watch pirated content, it directly costs the companies money.
Some countries have figured out a solution that does lower piracy, which is to just make it illegal for broadband companies to allow it. Will that work in the US?
——————————————————PAYWALL———————————————————
The Supreme Court has recently said no. Sony sued Cox Communications—a big cable provider—for allowing known pirates to continue to download pirated content. Sadly, the decision was unanimous, implying it wasn’t a close call.
Unfortunately, the problem is still likely Section 230 of the Communications Act, which holds that providers aren’t responsible for the behaviors of their users on the internet, but that’s a poorly written law from the previous century that needs to be updated for the current internet.
The solution is for Congress to actually pass laws protecting copyright and IP rights holders at the expense of Big Tech’s advertising dollars. So I’m not holding my breath.
Theaters – The Box Office Starts 2026 Strong
When it comes to the US box office, there’s good news and there’s bad news.
The first bit of good news is that Universal is lengthening its theatrical window! Their films will now stay in theaters for five weekends before heading to TVOD/PVOD.
My take? This is unequivocally good news. Shorter theatrical windows directly led to customers staying home from theaters. And yeah, if I ran a studio, I’d probably try to lengthen windows even further, especially SVOD windows. It used to take at least a year, if not longer, for movies to make it to pay cable, let alone basic cable. Studios did that for a reason!
In very, very recent news, Amazon-MGM’s Project Hail Mary, an original, non-IP film, as everyone (really, everyone?) pointed out, had a big opening weekend. This is good news on two levels. First, on the macro level, 2026 American box office returns are up 21% over 2025 and down just 20% from the late 2010s, which, echoing Sean McNulty, is a great start to the year considering the limited offerings in January and February.
I don’t like responding/overreacting to one-off events, but this successful launch should (fingers crossed) convince Amazon-MGM to send more films to theaters. True, they’re just batting one for three at theaters this year (after Crime 101 and Mercy flopped) and Apple Studios had a big opening with F1 last year, but they’re not sending any films to theaters this year. But Project Hail Mary could inspire Amazon-MGM to try to replicate this success in the future.
One other piece of good news. According to Pew Research, half of all Americans went to the movies in the year before. Half! That’s huge! For all the ‘movie theaters are dead” punditry, that’s a huge counterargument against it. (It doesn’t appear that Pew runs this survey over time, so I can’t compare it to the past.) Cinema United, formerly NATO, used to publish annual reports on the US box office, and I wish they started doing so again. We need more stats like this.
Okay, on to the bad news. It was a rough end of the year for movie theaters. Cinemark and AMC made more money in 2025 than 2024, but ticket sales were down. What explains this discrepancy? Well, ticket prices went up by 6% last year, which is one reason why less wealthy Americans aren’t going to theaters as much. AMC, in particular, posted a 10% decline in Q4. And luxury cinema iPic filed for bankruptcy.
Cord-Cutting
So…the rate of cord-cutting slowed down again in Q4 of 2025. To be clear, most companies had net losses, but those losses might be slowing down.
I’m still waiting on a full analysis from the experts who follow this space, which usually come out in April, but Comcast “only” lost 245K cable subscribers, down from over 300K in the two previous quarters. Charter’s losses were also down year-over-year.
But as I wrote last time, I’m waiting until Q2 of this year to see if these declines are actually slowing, mainly to see the impact of the NFL on cord-cutting.
