Meta Platforms plunged more than 9% premarket after third-quarter results mixed a clean top-line beat with a messy EPS miss and a bigger promise: even more AI spending in 2026. Revenue accelerated, ad pricing and impressions rose, and usage climbed. None of it mattered as the market fixated on what comes next — a capex curve that now tilts sharply higher to fund Mark Zuckerberg’s push into superintelligence, with a payoff timeline that Wall Street can’t model.
Meta delivered its fastest sales growth since early 2024, with Q3 revenue up 26% to $51.24 billion, above estimates. But management’s message on spending overshadowed the beat. CFO Susan Li said capital expenditures will grow again in 2026 at an even faster clip than 2025, when Meta plans to spend roughly $70 billion to $72 billion. That resets the bar higher for cash needs just as investors hoped the spending wave would crest. Evercore ISI called out the “significant increase” for 2026 as the driver of the after-hours selloff. Oppenheimer cut the stock to perform, arguing the superintelligence spend lacks a defined revenue path and is starting to rhyme with the metaverse cycle that sank the shares in 2022.
The quarter’s headline EPS of $1.05 badly trailed forecasts near $6.72, but the gap reflects a one-time, noncash income tax charge of $15.93 billion tied to implementing prior U.S. tax changes. On an adjusted basis excluding that item, EPS was $7.25. Under the hood, the ad engine looks healthy: Meta’s Family of Apps posted 14% more ad impressions and a 10% increase in average pricing, with users across Facebook, Instagram, Messenger and WhatsApp rising 8% to 3.54 billion. Guidance for Q4 revenue came in a touch light at a $57.5 billion midpoint versus $57.8 billion expected, not enough on its own to knock the stock this hard. The pressure is about durability of returns on AI investment more than current-quarter execution.
The market’s sticking point is ROI clarity. Microsoft (MSFT) and Alphabet (GOOGL) also ratcheted up their capex plans this week, but they have hyperscale cloud businesses that monetize AI infrastructure immediately through compute, storage, and platform services. Meta doesn’t run a commercial cloud; its AI stack must pay for itself through engagement lifts, ad targeting improvements, and new consumer products. Zuckerberg told analysts AI recommendations increased time spent on Facebook by 5% in Q3, and ad automation drove part of the sales acceleration. That helps, but it is not a cloud-like meter. Analysts also flagged a capabilities gap: Meta’s open-source frontier models have trailed offerings from OpenAI and Anthropic, making direct model monetization less certain. Needham’s Laura Martin warned that shareholders may be subsidizing “broad societal goals” rather than near-term, Meta-specific returns.
Zuckerberg wants Meta to be the leading frontier AI lab and said he will “aggressively frontload” compute. The vision is to be ideally positioned if superintelligence arrives sooner. If it comes later, he argued the extra compute will accelerate Meta’s core businesses. The cadence is bold — and revives awkward memories. Reality Labs lost $4.4 billion in Q3 on just $470 million in sales, and the stock has not fully forgotten that burn. Meta has reorganized under a Meta Superintelligence Lab umbrella, slowed hiring, and pushed to recruit talent from rivals. It has also leaned into external bets and infrastructure: a $14.3 billion commitment to Scale AI coupled with recruiting its CEO, Alexandr Wang, as Chief AI Officer, a $1.5 billion data center in El Paso, and a $27 billion financing package for the Hyperion campus in Louisiana. The scale signals conviction. It also locks in years of cash outlay.
Meta’s thesis hinges on compounding flywheels. More compute unlocks better recommendation engines and ad tools, lifting engagement and monetization; better models could spark new consumer AI products across Instagram, WhatsApp and Messenger; enterprise services may emerge through business messaging and agentic workflows; and open-source leadership could enable licensing or ecosystem monetization. The counter is that this requires breakthroughs and product-market fit that haven’t materialized at a scale commensurate with the spend. The technical risk is real: GPU availability, energy and cooling constraints, and supply chain timing could jeopardize schedules. The strategic risk is larger: if OpenAI, Anthropic or hyperscalers lock in model and platform distribution before Meta’s stack catches up, the marginal returns on each new dollar of capex diminish.
Technicals add fuel. After a 27.7% year-to-date run into the print, the stock looked extended. The premarket drop toward 680 suggests a decisive break of the 50-day moving average and an undercut of a developing base, opening the door to systematic and momentum selling. Options pricing implied a sizable move; vol sellers now need a floor that fundamentals alone did not provide overnight. Meta is still one of the cleanest ad-recovery plays, but the price action says the buy-the-beat muscle memory is gone when the spend curve turns vertical. Expect sell-side estimate revisions to push out AI monetization timelines and trim near-term margin trajectories, a combination that usually compresses multiples before stabilizing.
Two things. First, evidence that incremental compute directly drives incremental ad dollars at high flow-through — not just higher time spent but higher yield per impression sustained across regions. Second, a credible product roadmap for AI-native revenue: paid assistants, enterprise agents integrated with WhatsApp Business, and model licensing that is material enough to offset capex growth. Clarity on unit economics for new AI products would help. So would proof that Meta’s open models close the capability gap at a cadence that justifies the capital. Conversely, each step-up in 2026 spend without accompanying monetization milestones risks more downgrades and a reset of buy-side positioning.
Meta remains a dominant digital advertising platform with a rapidly improving ad stack. It is also becoming one of the largest private buyers of compute on the planet. Those dual identities can coexist if management turns compute into cash flow faster than the market expects. On the current trajectory — 2025 capex of roughly $70 to $72 billion and larger in 2026 — investors will demand a more granular bridge from teraflops to dollars. For now, the burden of proof sits squarely with Meta. The stock is trading the capex guide, not the revenue beat, and the next few quarters will be a referendum on whether the superintelligence bet is a moat builder or a margin drag. If Zuckerberg is right on timing, the payoff could be generational. If not, the market just put a price on patience.
