Much of the American economy is currently being propped up (some may argue artificially) by the massive amount of investment being made into AI projects, largely by hyperscalers. Few firms have made a bigger public to-do about their AI spending than Meta, whose CEO very publicly committed to pouring billions of dollars into AI infrastructure in front of the President of the United States.

So why, of all of the “Magnificent 7” Big Tech stocks that seem to have achieved “line only goes up” status, has Meta—arguably the hyperest of the hyperscalers—seen most of its stock gains for the year wiped out in recent days?

At least part of the answer may lie in the company’s Superintelligence Lab, which has undergone expensive change after expensive change since Meta’s pseudo-acquisition of Scale AI in June. The $14 billion investment that brought Scale’s CEO, Alexandr Wang, under Meta’s umbrella was followed by the company handing out multi-million dollar contracts to bring in top talent. The company’s deals for the world’s best AI researchers reportedly included giving out $100 million per year paydays and multi-year commitments that would get recipients over a quarter of the way to billionaire status.

The plan seemed to be to get all the talent in-house and figure out the rest later. But the figuring it out part has yet to happen. The company launched its Superintelligence Lab, separate from its previous AI efforts, in June. In August, it broke its lab into four separate teams. In October, it cut around 600 people who were working in those teams, sparing its primary “superintelligence” effort while downsizing everything else AI-related. As all that was happening, some of Meta’s big hires were reportedly already looking for the door and were handed promotions to keep them in the fold rather than letting them walk.

And all of that happened seemingly in response to the fact that the latest version of the company’s large language model, Llama 4 Behemoth, was underperforming expectations and had its release delayed. It’s not Meta’s only AI product to underwhelm.

The company released Vibes, a feed filled with AI slop, before OpenAI’s massive Sora 2 app flooded the internet with AI-generated videos, but it’s yet to garner even a fraction of the hype. Last month, TechCrunch reported that Meta’s AI app hit 2.7 million daily active users after the introduction of Vibes, which is available in the Meta AI app. That’s not nothing, but the standalone Sora 2 app got more than one million downloads in less than a week, and ChatGPT has more than 800 million active weekly users, per OpenAI.

That performance would underwhelm for just about any consumer-facing app, but it’s particularly bad for a company that has massive reach and an install base like Meta. Its attempts to capitalize on its user base of several billion people have not been well-received. Shortly after the launch of the Meta AI app, it was revealed that user prompts were publicly visible to others. The company tried to show off its capabilities on Instagram by creating wholly fabricated profiles in what felt like an effort to seemingly trick people into engaging with AI. Investigations into Meta’s chatbot applications found that it would have sexually explicit conversations with minors, and in an effort to ramp up engagement, the company reportedly lowered guardrails around what its chatbots could say.

Despite all that, Meta’s engagement traps have yet to catch anywhere close to as many users as it would prefer. According to the latest edition of venture capital firm Andreesen Horowitz’s “Top 100 Gen AI Consumer Apps,” which came out in August, Meta AI barely cracked the top 50 of most used web apps, ranking 46th, and failed to make the top 50 for mobile apps.

There’s an irony to the fact that Meta is probably the company best equipped to monetize its AI projects, and it simply can’t seem to do it. OpenAI has basically zero infrastructure in place to cash in on its huge user base outside of selling subscriptions, and all of its ideas are just things Meta already dominates, like targeted advertising.

Unfortunately for Meta, it seems investors simply may not trust the company’s capabilities to crack the code and turn its AI tools into anything other than money-sucking black holes. The company has committed more than $600 billion into building out its AI infrastructure over the next few years—a promise seemingly made to please Donald Trump—which has typically been viewed by the market as a positive indicator. But it doesn’t seem to have the benefit of the doubt that firms like OpenAI get when it comes to actually making those endeavors profitable.

That’s a bit of self-inflicted harm by Zuckerberg, who publicly said misspending billions isn’t a big deal. “If we end up misspending a couple of hundred billion dollars, I think that that is going to be very unfortunate, obviously,” Zuckerberg said on the Access podcast. “But what I’d say is I actually think the risk is higher on the other side.” There’s also his history of going all in on projects that don’t pay off. The company went so all in on the idea of the “metaverse” that it changed its name, and yet it has reportedly lost nearly $100 billion chasing a virtual reality dream that was basically dead on arrival.

Meanwhile, the company’s money-making businesses are under scrutiny, too. Earlier this month, Reuters got hold of internal documents from the company that showed it projected as much as 10% of its revenue was coming from advertisements for scams and banned goods.

Monetizing scams probably wouldn’t be the worst business model in the world when it comes to AI, given the amount of bullshit it can generate at scale. But it’s also not exactly the thing that investors want to hear is a major source of revenue on earnings calls.