Meta's AI spending spree is Wall Street's focus in second-quarter earnings

Over the years, Meta has built a reputation for using rivals' innovations to bolster its technology. But its decision to copy a Chinese artificial intelligence lab in 2025 in an effort to compete with OpenAI backfired, forcing the company to overhaul its AI strategy.

In a rush to mimic the techniques developed by Chinese startup DeepSeek, Meta released a new version of its Llama family of AI models that disappointed third-party developers, according to people familiar with the matter. The reaction was so bad that CEO Mark Zuckerberg decided to spend billions of dollars to revamp the company's AI unit, and he's still considering more shake-ups to Meta's AI strategy, said the people, who asked not to be named due to confidentiality.

When Meta reports second-quarter earnings on Wednesday, Zuckerberg will make the case to investors for his AI hiring spree and the company's related strategy shift.

Meta's AI blitzkrieg kicked off in June, when it invested $14.3 billion into Scale AI, resulting in the data annotating startup's CEO, Alexandr Wang, joining Meta along with a handful of employees to oversee a cornerstone AI unit. This new Meta Superintelligence Labs will be led by Wang, now Meta's chief AI officer, and former GitHub CEO Nat Friedman, who also joined the company in June along with business partner Daniel Gross.

Gross was previously the CEO of AI startup Safe Superintelligence, which Meta tried to buy before being rebuffed by co-founder Ilya Sutskever. While Meta couldn't land the AI pioneer and former OpenAI co-founder, it did hire multiple top researchers from competitors like the ChatGPT maker, Apple and Google to help it regain its footing in the fiercely competitive artificial intelligence market. One such hire was ChatGPT co-creator Shengjia Zhao, who Zuckerberg last week named as his new AI lab's chief scientist.

Although Meta's AI talent grab may not result in the company raising its projection for 2025 total expenses, estimated to come in between $113 billion to $118 billion, Cantor analysts said in a note published earlier this month that the investment potentially "moves the target above the low end." Translation: all that hiring comes with a cost, albeit slight.

Meanwhile, revenue growth in the second quarter likely slowed to 15%, down from 22% a year earlier, according to LSEG. It would be the slowest rate of expansion for the company since early 2023, and analysts are expecting lower levels of growth in the coming quarters.

Zuckerberg believes that new AI talent as part of the Superintelligence unit is worth it if Meta can regain its momentum and potentially create more powerful AI technology that steamrolls the competition, CNBC reported in June.

For Meta, Llama 4 represented the company's answer to competing models from rivals like OpenAI, and executives have viewed it as helping the company dominate a potential computing platform of the future.

Similar to other Meta-incubated technologies like the PyTorch AI developer tools, the company released Llama to the open-source community, which can then access and use the software for free, subject to certain licensing terms.

While the predecessor, Llama 3, was a hit with developers, they haven't taken to Llama 4 because it's seen by some as more difficult to customize and integrate into their apps. That's resulted in many coders preferring Llama 3 over its successor, people familiar with the matter said. Additionally, Zuckerberg lost confidence in his generative AI team and its leadership in part due to a controversy over whether Meta may have gamed certain industry AI benchmark tests, the people said.

Llama 4's struggles can be traced back to January, when the sudden rise and ensuing popularity of the open-source R1 AI model by DeepSeek caught Meta off guard, leading to a reevaluation of Llama's underlying architecture, the people said.

DeepSeek's R1 is a so-called mixture-of-experts AI model, or MoE. R1 is similar to OpenAI's o1 family of models that can be trained to excel at multistep tasks like solving math equations or writing code.

By contrast, Llama's models — before their latest release - were dense AI models, which are generally simpler for most AI developers to fine-tune and incorporate into their own apps, the people said.

AI labs like OpenAI and Anthropic, researchers say, have been pushing MoE models to power AI agents that can perform a variety of step-by-step tasks. Those companies keep their designs closely guarded from competitors. OpenAI has been developing an AI model for the open-source community, but CEO Sam Altman said earlier this month that its debut is delayed indefinitely pending safety tests and other reviews.

Although Meta has previously published research on MoE models, DeepSeek's release to the open-source community wowed researchers because R1 appeared to be less expensive to train and run compared with other AI models, experts said.

Suddenly, Meta executives thought they had a clearer picture into how to create their own efficient and possibly cheaper MoE models, potentially leapfrogging rivals like OpenAI, people familiar with the matter said.

Still, some staff members in Meta's GenAI unit pushed for Llama 4 to remain a dense AI model, which though generally less efficient, is still powerful, and Meta originally planned on that architecture acting as the backbone supporting improved voice recognition capabilities, the people said.

Ultimately, Meta went with the MoE approach, due in part to DeepSeek's innovations and the promise of pulling ahead of OpenAI, the people said. Meta released two small versions in April and said a "Behemoth" version would come at a later date.

But the new MoE architecture disappointed some developers, who were simply hoping Llama 4 would be a souped-up version of Llama 3, people familiar with the matter said. Llama 4 also failed to deliver a significant leap over competing open-source models from China, the people said.

Executives at Meta as well as the Superintelligence Labs' high-profile hires are now questioning the company's current open-source AI strategy, and have considered skipping the release of Behemoth in favor of developing a more powerful proprietary AI model, the people said.

A Meta spokesperson said in a statement that the company's "position on open source AI is unchanged." "We plan to continue releasing leading open source models," the spokesperson said. "We haven't released everything we've developed historically and we expect to continue training a mix of open and closed models going forward."

The New York Times first reported that Meta was considering upending its open-source AI strategy.

Despite Meta's AI struggles, the company's core online ad business remains strong, and investors are hopeful that the recent AI investments and hiring will eventually pay off.

Zuckerberg said in July that the company would invest "hundreds of billions of dollars" into building out the computing infrastructure needed to power cutting-edge AI projects.

"Meta Superintelligence Labs will have industry-leading levels of compute and by far the greatest compute per researcher," Zuckerberg said in a Facebook post.

Analysts at Bank of America said in a note this month that Zuckerberg's comments indicate "a sign of confidence in Meta's revenue trajectory." The analysts said that his statement also "implies higher future Capex and Opex," which potentially equates to even more AI spending.

"We also see the post as reaching out to AI talent, signaling Meta as a place for AI innovation," the analysts wrote. "We expect AI investment to be a top focus area on the upcoming earnings call, and Meta likely needs to make a case for strong AI returns to drive multiple expansion."

Meta and its rivals' pursuit of AI researchers echoes the self-driving car frenzy of 2017, when companies like Google and Uber competed fiercely for talent, doling out "similarly crazy kind of pay packages across the board," said Megh Gautam, chief product officer at deal-tracking firm Crunchbase.

"The dynamics still feel very much like a winner-take-all-market, so you're trying your best to give yourself the best shot possible to go make that happen," Gautam said.

Investors appear more receptive to Meta's AI spending and strategy shifts, a contrast with a few years ago when the company was heavily pushing the metaverse, said Uday Cheruvu, an analyst and portfolio manager at Harding Loevner, which owns Meta shares.

OpenAI, Google and Anthropic are also trying to hire and maintain talent all while continuing to spend billions of dollars on developing their respective AI models, Cheruvu said.

"Now with AI, it's not just Meta – everyone else is doing it, so now the euphoria is much higher," Cheruvu said.

WATCH: Meta's ambitious AI plans.