Yesterday, OpenAI continued its streak of announcing massive deals with fuzzy, circular math—this time with the chipmaker AMD. As part of the deal, OpenAI committed to deploying a total of 6 gigawatts of AMD G.P.U.s over the next several years, starting with a 1-gigawatt deployment in the second half of 2026. In exchange, AMD will grant OpenAI warrants for 160 million shares, or about 10 percent of the company, which will vest after certain milestones are achieved.

The warrants could be worth more than $30 billion, which should come in handy for Sam Altman’s still cash-strapped company. The shift away from Nvidia is also great news for AMD: At tens of thousands of dollars per G.P.U., the partnership could potentially deliver “tens of billions of dollars in revenue for AMD,” according to the company’s C.F.O. Of course, where that money is coming from—and how OpenAI is going to meet all its compute commitments given that the capacity does not currently exist—are questions for a different day, and ones Wall Street seems content to ignore. Anyway, shares of AMD spiked 23 percent on Monday and nearly 4 percent today—another example of OpenAI-adjacent stock market magic.

And now for the main event…

It didn’t take long for the latest version of Sora, OpenAI’s video-generation app, to become just about the only thing people in entertainment and media circles want to talk about. After all, Sora 2 was only released to the public last week following a brief, invite-only preview period. Over the weekend, the app topped the iOS App Store, created a wave of excitement among OpenAI’s feverish supporters, and freaked out basically all of Hollywood with its copyright law-oblivious A.I. TikTok successor. To wit: Charles Rivkin, the chairman and C.E.O. of the Motion Picture Association, implored OpenAI in a statement to “take immediate and decisive action” to “prevent infringement on the Sora 2 service.” Sam Altman, not one to downplay OpenAI’s goals and achievements, meanwhile proclaimed that a “Cambrian explosion” of creativity had arrived, while acknowledging the potentially addictive nature of an A.I.-optimized “slop feed.”

Of course, OpenAI has already had to address some of the more salient concerns. As my partner Matt Belloni wrote shortly after the app was released, Sora 2 was “trained on hundreds of thousands of videos and actor performances it doesn’t own,” and its opt-out permission structure—wherein rights-holders have to proactively forbid OpenAI from using their work—became the source of handwringing across the entertainment industry. Altman has since walked things back slightly, saying that rights-holders should have more control over how their work is used. He also noted that OpenAI is “going to have to somehow make money for video generation,” since users—the vast majority of whom are not paying customers—are creating “much more than we expected,” and that he hopes to share “some of this revenue with rights-holders who want their characters generated by users.” At this point, it’s hard to say whether Sora 2 represents the birth of a new type of social media or if the excitement will wane. That said, the videos are impressively hyper-realistic—which has, naturally, already led to plenty of concerns about misinformation—and users seem genuinely floored by the app’s capabilities. To figure out what might come next, I called up my friend and Puck partner Julia Alexander to get her thoughts on the app’s place inside the broader social media ecosystem, and then rang Simon Pulman, a copyright attorney at Pryor Cashman, for his thoughts on the ever-present rights issue. The following conversation, as always, has been lightly edited for clarity.

Ian Krietzberg: I want to start with the straightforward social media element of this. I don’t think it’s a coincidence that Vibes from Meta, and then Sora 2 from OpenAI, which are essentially the same thing, were released so close together. Do these companies know something that we don’t, or is this a wild miscalculation about what social media will become? Julia Alexander: It’s a great question. For now, I think Vibes is the more interesting case study, given Meta’s existing social media ecosystem. If you’re OpenAI, you have to build some kind of network to incentivize people to produce content in order to then better train your data off that content and better understand what type of content performs well—especially if you’re going to introduce advertising within ChatGPT. That would put OpenAI and Altman in a really strong position to convince advertisers to spend money there.

But Meta already has three very popular social media networks: Instagram just crossed 3 billion monthly active users; Facebook’s sitting around 2 billion; and Threads now has like 450 million users. Those are substantial bases from which Meta can encourage people to participate in their A.I. content if they want to further integrate Vibes into its interoperable ecosystem. That means we’ll get a better sense, in short order, about the appetite for these platforms. But I want to get your opinion. I think that Meta, like OpenAI, created this separate app to collect data, better train its own models, and better understand how people might respond to this specific content. I also think that not fusing it with Instagram signals a belief that people don’t want A.I. slop in their feed, which might push more people toward TikTok. I also think that they’re worried about how potentially unhinged the content might be in a way that invites negative press and attention. I agree that the decision to create a separate app seems like a very calculated move. I also think it’s obvious that, for both companies, this is about the data. Meta, of course, recently said they’re going to start serving ads across their social media platforms based on how users interact with their A.I. assistants. And just last week, OpenAI introduced checkout with ChatGPT, which makes it seem like we’re one step away from ads. In short, these platforms feel like a shortcut to collecting more free user data and unleashing even more targeted advertising. Totally. I think it’s inevitable that OpenAI gets into advertising—and inevitable that they find ways to use all the data they’re collecting to sell to advertisers. But I’m curious for your thoughts, Ian, on whether or not the company aspires to make the Sora 2 social media network a pure ad play or if it has other ambitions. I think there are a few goals here. First, it’s an opportunity to simply showcase model capability, which is something they’re very interested in. Perhaps it’s because of their first-mover advantage, but OpenAI and ChatGPT are essentially synonymous, in peoples’ minds, with A.I., and any opportunity to reinforce this thinking is a valuable proposition. For example, Google’s Veo 3 is also interesting, but most people outside of the space don’t know about it, largely because of OpenAI’s mindshare. But I think, ultimately, this is an engine to serve advertising. I’ve written about this before, but the company is losing astronomical amounts of money as it makes deals and builds out data centers. Meanwhile, most of Google’s revenue comes from advertising, and the same goes for Meta. They’d like to get to a place where users are comfortable seeing advertising sit alongside their content. They’re just not there right now. That’s a great point—especially in light of the fact that, for the last year, the company has been edging closer to becoming for-profit. But from what I’ve seen so far, the Sora feed is mostly funny edits of Altman, because he’s one of the only people who has authorized his likeness to be used. This gets at a paradox that has existed within social media for the past half-decade or so: As the social internet has become more intimate, many people have calculated that the consequences of posting have started to outweigh the benefits. These days, a relatively small percentage of people are creating the posts that appear in front of the vast majority of users. It’s a fundamental psychological hurdle that OpenAI and others are going to have to face, especially as they try to edge closer to their goals for these platforms. People just don’t trust the internet anymore. The other thing I keep thinking about, regarding an endless feed of A.I. videos, is that other companies, including Netflix and Disney, have tried to put shortform video inside of their apps and there’s never any engagement. I think that’s because, when you have an array of content woven together—where you can scroll from a clip of stand-up comedy to a sports highlight to a political rant—this creates a stickiness that those platforms lack. Unless people become incentivized to post more, I don’t think people actually want to consume A.I.-only feeds. As impressive as some of the clips are, they can become monotonous. I’m curious where that leaves a company like OpenAI; Google and Meta have the edge there because there’s more human interaction happening.

I’m curious, did Altman’s decision to change the opt-out regime actually protect OpenAI from lawsuits focused on infringements before he backpedaled? Simon Pulman: The answer to that is no. I mean, in theory, it could mitigate their damages if they take remedial measures to prevent prospective infringement. But if infringement has occurred, and has potentially created damages, then no—they would be on the hook.

There are various ways you can read [Altman’s update]. One way would be that he was making a clarification—we forgot to say in the initial announcement. But I don’t really get that sense. There’s a second view, which is that they really were trying to make a land grab, and now they’ve received so much pushback that they’re retreating. Then there’s the third one, which is they really just haven’t thought about it. I’m fairly incredulous that this could be the case, but it speaks to the culture of the company and a lot of tech companies. When you get outside of Darth Vader and Homer Simpson and incredibly recognizable characters like that, we’re at the indie level, which is basically 98 percent of copyrighted works. The only way the system would know not to use something would be to show that thing to the system. But by doing that, I have now put that piece of art into the system, and I have no way to verify or audit that it hasn’t been used or infringed. So it’s immensely complicated, it’s very difficult to administer, and it’s even harder to police. To me, the initial policy felt like me going into a neighborhood and posting flyers around, saying, Unless you opt out, I’m going to break into your house this evening and steal everything. That’s just not how it works. So far, the lawsuits have been fairly nebulous about whether training an A.I. model is protected under fair use, but they seem to be pointing to the notion that infringing output is a bit of a different story. From a legal perspective, does Sora seem a more straightforward target? Well, if you look at the initial studios’ complaint, it was almost entirely just collections of images. Here is our character. Here is what was outputted. It was Darth Vader or the Minions. That feels fairly clear-cut to me. There’s a question, obviously, around Terms of Use, and whether it pushes liability onto the users for creating those images. But I would imagine the company is at least partially liable here. The whole thing is just an absolute mess. I think this is playing out on at least three levels simultaneously: legally, which is the slowest; on the technological level; and then also on sort of a cultural, societal level. That third one is possibly the most important at the moment, because they’re fighting in the court of public opinion about whether it’s reasonable, about what should be permissible, about what should be normal. But you know, there’s a long, long way to go to establish any kind of certainty around this. Do you expect media companies to start fighting this really aggressively, or do you think we’re going to see companies increasingly work to get in on the revenue-sharing arrangements? It was probably about 18 months ago that this whole thing really started to push into the popular culture, really around the guild negotiations. I wondered whether there would be a complete repudiation of A.I. by the studios. That hasn’t happened. And my feeling is that it’s on an individual basis within these companies, and also collectively, on behalf of the companies. You sort of have to follow the incentives: If somebody thinks their company can cut costs and increase profits, and therefore they individually can get a bigger bonus, they will generally pursue that course of action. And frankly, if you’re a big movie star, you’re thinking, Okay, how can I use my digital likeness and my digital replica to make more money for myself? So ultimately, it seems that we’re moving in the direction of making deals, having revenue splits, etcetera. It’s a little bit of a tragedy of the commons effect, because those huge conglomerates can make deals, and they can get revenue shares and the rest. If you’re an independent musician, however, you don’t have that same level of leverage. But you’re letting these big companies effectively dictate what happens in terms of both policy and the law. That’s the challenge around it, and for multiple reasons, I tend to think that A.I. is going to lead to the bigger brands and bigger creators just becoming bigger. This is a playground of legislators and billionaires.

That’s all for today. I’ll see you on Thursday.

Ian

Need help? Review our FAQ page or contact us for assistance. For brand partnerships, email ads@puck.news.

You received this email because you signed up to receive emails from Puck, or as part of your Puck account associated with {{customer.email}}. To stop receiving this newsletter and/or manage all your email preferences, click here.

Puck is published by Heat Media LLC. 107 Greenwich St, New York, NY 10006