Meta drops 9.8% as Zuckerberg says open-source AI lags frontier models

Meta Platforms Inc. shares plunged 9.8% on July 29 after Chief Executive Officer Mark Zuckerberg told analysts on the earnings call that open-source AI models are not as powerful as frontier models, a candid assessment that undercut the company's massive artificial intelligence spending narrative.

"The gap between open-source and proprietary frontier models remains significant, and we're being transparent about where we stand," Zuckerberg said on the call, according to a transcript reviewed by Edgen.

The selloff erased about $60 billion in market value, pushing Meta shares to around $535 from a January peak near $712. The company reported Q2 revenue of $56.31 billion, up 33% from a year earlier, with ad impressions rising 19% and average price per ad climbing 12%. But the headline EPS beat of $10.44 against a $6.66 consensus estimate included an $8.03 billion tax benefit tied to CAMT and R&D treatment, putting normalized earnings closer to $7.31 per share.

The admission comes as Meta plans to spend $125 billion to $145 billion in capital expenditures this year, much of it directed at Meta Superintelligence Labs. If the company's own models cannot match proprietary alternatives from OpenAI, Google, or Anthropic, investors are left questioning the return on what would be one of the largest corporate investment programs in history.

The Open-Source Reality Check

Zuckerberg has long positioned Meta as a champion of open-source AI, releasing models like Llama as a counterweight to the closed ecosystems of rivals. But his latest remarks mark a notable shift in tone. Rather than arguing that open-source models would eventually catch up or surpass proprietary ones, he acknowledged the current hierarchy: frontier models — the most capable systems developed by companies like OpenAI, backed by Microsoft Corp., and Alphabet Inc.'s Google DeepMind — hold a meaningful performance advantage.

The distinction matters for Meta's investment thesis. The company's AI infrastructure spending, which has climbed from $35 billion in 2024 to a projected $125 billion-plus this year, was partly justified by the argument that open-source models would democratize AI and reduce the industry's dependence on a few proprietary players. If Meta's own models are not competitive at the frontier, the spending looks less like a strategic moat and more like a cost of staying in the game.

The CapEx Conundrum

Meta's capital expenditure plans have been a central concern for investors all year. The $125 billion to $145 billion range for 2026 represents a more than threefold increase from 2024 levels, with much of the spending directed at data centers, graphics processing units, and custom silicon for AI workloads. For context, that exceeds the entire annual capital budget of Exxon Mobil Corp. and approaches what the US Department of Energy spends on nuclear weapons.

Reality Labs, Meta's augmented and virtual reality division, lost $4.03 billion in the most recent quarter and $19.2 billion in 2025, adding another layer of cost pressure. The division has yet to produce meaningful revenue, though Zuckerberg has said the goal is to deliver personal superintelligence to billions of people.

What This Means for Investors

Meta shares now trade at about 19 times forward earnings, a discount to their January multiple of 25 times but still above Alphabet Inc.'s 14 times. Of the analysts tracked by Bloomberg, 57 rate Meta a buy with a consensus price target of $826, implying roughly 54% upside from current levels. But the bull case depends on the advertising business continuing to fund the AI buildout — and on the AI investments eventually producing a return.

The Q2 earnings report, due in full after the close, will offer a clearer picture. Investors will be watching for any revision to 2026 operating income guidance and for signs that Meta's AI investments are translating into advertising revenue growth. Without those signals, the 9.8% drop may not be the end of the selloff.

This article is for informational purposes only and does not constitute investment advice.