Why Meta's bleak 'AI vision' is winning out

And why OpenAI's position looks more precarious by the day

Last month, Meta announced that it was going to begin firing thousands of “low-performing” workers as it shifted to a strategy that focused on AI. When the DeepSeek launch upended other American competitors and sent Nvidia shares downward, investors noted that Meta too has an “open source” approach to its AI model, Llama, even though its Meta AI is rarely in conversation with OpenAI’s ChatGPT or Anthropic’s Claude as a leading consumer product. Meanwhile, revelations surfaced that Mark Zuckerberg had likely personally OK’d the pirating of vast troves of data to train that model. On Monday, the layoffs began.

Here’s how the stock market has responded. Per Bloomberg, Meta shares saw

a rally of 16 straight sessions, the longest streak of any current Nasdaq 100 Index company going back to 1990. The stock added more than 17% over the surge, bringing its market capitalization above $1.8 trillion.

That story is headlined, “Meta’s 16-Day Winning Streak Is Victory for Zuckerberg’s AI Vision,” and that is indisputably true, if you consider “own a monopoly social media network and web infrastructure that is so large you can do anything you want on it” plus “do crude identity politics to placate the new president” as an “AI vision.”

But it is important to understand why it is that investors are cheering Meta, while OpenAI is facing more and more hurdles, its broader picture getting bleaker by the day.

Before we dive into that, however, a brief note that paid subscribers make this newsletter possible—it takes many hours to research, report, and write these columns, and I have some big projects coming down the pike. But I can only keep it up under this model if folks like you support the work; if you find any of this valuable, please chip in for the cost of a coffee a month to keep the lights on here. Many thanks, and onwards.

As the Great DeepSeek Event made clear, OpenAI’s (and Anthropic’s, and other AI model-making companies’) biggest issue now is that AI chatbots are on the brink of becoming a commodity, if they haven’t already. Building AI models, like OpenAI does, is suddenly a pretty bad business to be in. It’s expensive, and for all but the most invested users, the experience of using most chatbots on the market is becoming less distinguishable.

This is why OpenAI is focusing more and more on its ChatGPT app, which is popular, but still incredibly expensive to run, its enterprise businesses, and government contracts. It’s also why, despite being valued at whatever hundreds of billions OpenAI currently is, pending Softbank investment, the startup is in a somewhat precarious spot. If the ChatGPT app isn’t profitable, and the enterprise business is still unproven, and government contracts are now subject to first buddy and Sam Altman’s personal nemesis Elon Musk, the future of robust federal contracting for OpenAI may not exactly be robust.

In fact, Musk has just thrown up another roadblock to OpenAI; he formed a consortium to offer a $97.4 billion bid for control of the company. (Admittedly, this is a pretty expert bit of corporate trolling; OpenAI, as you may know, has a uniquely labyrinthine governance structure born out of the original promise to benefit humanity and remain profit-free, in which the now-for-profit-company’s assets are controlled by a nonprofit entity. Altman is trying to ditch the nonprofit structure to solicit further investment, which means purchasing its assets; previously, he was slated to do so at $40 billion; Musk’s ‘offer’, which Altman has already declined, essentially raises the market value of the restructuring—the nonprofit board has a fiduciary duty not to undersell its assets—and gives OpenAI a major headache at the least.)

Furthermore, OpenAI’s first enterprise contract with the federal government was with… USAID. Given that Musk and his lackeys are tearing the copper wiring out of the agency at the moment, and the nature of Musk’s feud with Altman, it seems unlikely that the contract will remain intact for long, if it hasn’t been cancelled already. It also raises the question of whether Musk’s adventuring at the foreign aid agency began in earnest in the first place in part because OpenAI—which Musk despises—was doing business with the department. These are the things we now have to consider! Given the scale and nature of Musk’s power over the government, his volatile temperament, and well-known vindictive tendencies, etc.

But I digress; back to Meta.

Cristoph Scholz, CC 2.0 license, Flickr.

Now it’s not exactly that the relations between Meta CEO Mark Zuckerberg and Musk are miles better—it was what two years ago that the two were threatening to do a public cage fight—that Meta’s in a better position. It’s that Musk can’t really touch Meta or its products. And that, to circle things back to Meta’s “AI Vision,” is why investors love it too.

Let’s return to the Bloomberg piece that listed all the things that have allegedly fueled Meta’s historic, 16-day rise to a $1.8 trillion market cap over the last couple of weeks:

First, CEO Mark Zuckerberg said the company plans to invest as much as $65 billion in AI projects this year, more than had been expected — reinforcing the perception it is spending from a place of strength.

Then, the stock bucked the tech selloff sparked by the emergence of DeepSeek, a Chinese AI startup that claims to be lower cost. Investors took DeepSeek’s success as a validation of open-sourced models, as Meta’s Llama employs.

Meta’s results have also underlined how AI is flowing through to its financials, improving how ads are targeted to its billions of users. Zuckerberg fanned the enthusiasm, saying 2025 would be a “really big year” for AI.

It also hasn’t hurt that Zuckerberg has moved closer to President Donald Trump, potentially giving the company a bit of breathing room from tighter regulation. The stock has risen every day since Trump’s inauguration on Jan. 20.

Notice that absent in any of the above reasons for Meta’s soaring valuation is “it made a great or popular AI product that people are flocking to or even just using a lot.” No, it boils down to the fact that Meta owns the largest slabs of digital real estate on the web—Facebook, Instagram, WhatsApp—and it has paid lip service to investing heavily in the trend that is still Silicon Valley’s north star, which is AI. Remember, Meta made a big show of investing a similarly enormous figure in the metaverse, and that has decidedly not paid off, so it’s not like Wall Street is even expecting a great product to emerge from that firehose of cash.

No, what really matters to the money is simple, and same as it ever was: Meta is an all but intractable monopoly, and the fact that it’s outwardly keeping pace with new entrants is more than good enough for them, at least in terms of dollars invested. It has positioned itself to minimize any potential regulatory confrontations with the new administration. Better still is that the company is seriously trimming its workforce, firing costly engineers as it pivots to AI.

And that’s really about it. If investors honestly think generative AI is improving ad delivery on Meta’s networks, I have no idea what they’re looking at. Meta has long used automated systems to track user preferences and data and to then deliver personalized ads accordingly, but that predates generative AI or their product-side use of large language models. These tools, and Facebook’s presentation of them to investors, has perhaps benefitted from the AI effect of the last few years, lifting in stature features that can loosely be called ‘AI’ via the halo glow of buzz, but that’s about it as far as I can tell.

No, if anything, most users seem to actively dislike the AI that’s been introduced on Meta’s platforms—people find the stuff weird and alienating. I’m not sure who, besides the errant content farm operator, is actively enjoying the reams of spammy AI shrimp Jesuses or automatically generated images of suffering children. Or the bizarre AI-generated images of users served to themselves on Instagram. Or the unwelcome AI characters and bots showing up in Facebook groups uninvited, leaving odd comments. On the product side, I think most people who are familiar at all with what the company’s been up to would say that Meta’s AI plays have been a muddled disaster. (And on the model architecture side, there’s the piracy scandal, so that’s not great either.)

As the tech PR CEO and commentator Ed Zitron points out in his latest newsletter, Meta, Google, and Microsoft all took in tens of billions of dollars in profits last quarter, despite becoming increasingly hostile to their own users. He writes, “software has, for the tech industry, become far more about extracting economic value than it has in providing it.” After all, Google’s forays into AI have been as dispiriting as Meta’s, despite the huge research, product, labor, and ad spends, and yet its stock continues to rise, too. Microsoft’s as well—does anyone out there like AI Bing or stan for Copilot? Maybe! Though I haven’t met many avid fans.

So it’s important to stress just how little any of the above matters, to say nothing of piracy scandals and copyright lawsuits, and how relatively simple it is to articulate what does: That you are one of the handful of monopoly platforms that is large enough to reliably extract rents from your users, and to run experiments that those users may find unpleasant but will tolerate because the opportunity cost of leaving is too high. That you are large enough to invest in locating ways AI might extract more rents from users on your platforms in the future. And that you are willing to cut jobs and labor costs to maximize profits, and bonus points if you say you’re doing it because of AI.

That’s why it’s so bleak that Meta’s “AI vision” is prospering right now. There *is* little vision, as far as I can tell, apart from loading its platforms with AI bots and content, which users don’t seem to much want. It is largely interchangeable with Zuckerberg’s metaverse vision, and the crypto vision (remember Libra?) before that. But if Nvidia is the AI gold rush’s big winner so far because it’s selling the shovels, then Google and Meta stand to profit by owning the real estate on which AI might someday generate profits. OpenAI is, at core, a model, an app, and some enterprise services; even if it does have 300 million monthly users, that’s a fraction of the giants’ base, and it’s burning cash fast.

OpenAI at least had a novel product people were and are enthusiastic about—that it soon got catapulted into Big Tech-era logic necessitating rapid and immediate expansion at all costs just to compete says as much about the economics of our technological moment as Sam Altman’s titanic ambitions, and well, here we are. With our largest social media platform, whose existence was once predicated on connecting people, shoveling AI into every pore, hoping to automate the essence of those connections away, while Wall Street thrusts money into its maw.

And this is what I’ve long said about the AI boom: Ultimately, it boils down to automation. In the AI features stuffed into Google and Meta products, in, of course, the enterprise contracts Anthropic and OpenAI are inking, with image generation, even in the rise of government by Grok in DC under Musk’s direction, we have entered a new era governed overwhelmingly by big tech and its refreshed automation logic. Mark Zuckerberg and Meta are “winning” the way the leaders of so many of our institutions, online and off, seem to be “winning” these days, by leveraging institutional power, trimming headcount, and promising to dully automate what likely does not need to be automated at all.

---

There is, however, a ray of good news, too, today: Reuters won its lawsuit against the tech company Ross Intelligence, in the first major copyright victory in the AI era.

WIRED reports:

In 2020, the media and technology conglomerate filed an unprecedented AI copyright lawsuit against the legal AI startup Ross Intelligence. In the complaint, Thomson Reuters claimed the AI firm reproduced materials from its legal research firm Westlaw. Today, a judge ruled in Thomson Reuters’ favor, finding that the company’s copyright was indeed infringed by Ross Intelligence’s actions.

The big deal here is that the judge found that Ross violated fair use, and pointedly because it generated materials that could be used to compete with Reuters’ own original material. This is a contention of many of the claimants in the ongoing copyright lawsuits against bigger AI companies, and a promising sign that creators’ works may be protected by copyright law after all.

Okay okay, that’s it for now. I’ve got the stirrings of a cold, and I’m working on, let’s see, three separate longer form pieces right now and need to rest up a bit. So until next time—hammers up.