Meta is spending like a company that sees the future, but investors want proof that it can sell it. Following an earnings presentation that confirmed astronomical AI costs but offered few insights into how they might be commercialized, the share price nosedived, slashing more than $200 billion from Meta’s market cap. Operating costs rose by multiple billions of dollars compared to a year earlier, with capital costs nearing $20 billion, much of which was for compute, data centers, and snatching AI talent. The message from Wall Street was clear: the vision is not the product.
Why Wall Street remains skeptical about Meta’s AI strategy
The issue isn’t that Meta is looking at new AI investments; everyone is doing so. The problem is that nothing shown in the earnings presentation indicates that Meta has a flagship AI product with revenue potential. On the other hand, Microsoft can demonstrate expanding software ARPU as a result of Copilot seat licenses; Google can link subscription and cloud services growth to Gemini; and Nvidia boasts record data-center revenue due to demand. OpenAI is a private company—it claims to have thought through an AI transformation with a double-digit billion-dollar run rate and ten billion-dollar commercial subscriptions based on its SLOT mechanism. For Meta, as with Amazon, it’s “spend now, commercialize later.” These commitments are examples of the company making big bets, but none are likely to become flagship products for years to come.
It’s not that analysts don’t like scale—Meta is constructing modern infrastructure frontiers, more state-of-the-art data centers than I can track, and is building its silicon—they need to, as market expectations suggest AI/ML infrastructure in the U.S. alone could surpass the $0.1T mark in a couple of years. On a relative scale, $20B doesn’t feel overpriced, but it does if the firm isn’t anchoring products with measurable demand, margin, and pricing power.
Meta’s most visible AI product is Meta AI, a general assistant shipped across Facebook, Instagram, and WhatsApp. The company claims over a billion active users, though distribution across the feeds of social giants can inflate usage statistics. Completion of interactions, retention cohorts, and willingness to pay are vastly more substantial than raw reach, where Meta has stayed relatively silent. Many other launches seem extremely early. Vibes, the video generator, has driven a spike in creation and sharing but is not intimately tied to robust revenue mechanics apart from incremental ads. The Vanguard smart glasses, recently released, show on-device capabilities, but appear like an RL extension rather than a monetized LLM platform. These are exciting experiments, not genre-defining products. Internally, the Superintelligence lab architecture builder trains larger frontier models targeting features that rival or surpass the foremost systems. This could be a huge morale booster—provided it results in clear product differentiation. Baseline assistants could quickly become substitutes, not features.
The cost math is terrible. AI at Meta’s rate is a unit economics complication. Inference costs accumulate quickly when several billion people use assistants, render or remix media. Without huge performance gains like custom chips, model compression, and more on-device inference, prediction costs risk outpacing ad monetization gains. Meta fights this with in-house MTIA accelerators and a colossal GPU set plus smaller, faster models tuned to specific tasks. But silicon won’t repair everything. Energy and networking are nontrivial, also. Investors and board members need evidence that each AI feature elevates unambiguous revenue for Meta per user, not only extra compute burn.
Monetization paths that could work for Meta’s AI bets
Three avenues look credible if Meta executes:
- First, advertising performance: if AI improves content ranking and ad targeting, Meta should quantify it. A sustained lift in click-through rate or return on ad spend, even by low single digits across billions of impressions, compounds into real dollars. Publishing advertiser case studies and third-party verified lift metrics would help.
- Second, business messaging: WhatsApp and Messenger are well-positioned for AI agents that automate customer support, commerce, and lead qualification. Clear pricing—per conversation, per seat, or performance-based—and integrations for SMEs and enterprises can turn assistants into durable ARR, similar to how SaaS copilots monetize.
- Third, creator and consumer tools: AI-native creation for Reels, Stories, and ads can command premium features. If Meta bundles advanced generation, editing, and brand-safe asset libraries into subscriptions for creators and advertisers, it can capture willingness to pay without cannibalizing the free feed.
Product signals investors want to see from Meta’s AI
Clarity beats promises. The near-term signals that would reset sentiment are specific and measurable: a flagship AI product with pricing; disclosed unit economics and target gross margin; usage cohorts that show sustained retention; and a cadence of model upgrades that materially improve latency and quality.
From a platform perspective, Meta’s success varies based on how deeply AI is integrated into Instagram search and shopping, Reels, and business inbox creation on WhatsApp. If Meta’s subsequent models create distinct experiences—whether that is real-time multimodal creation, personal memory with strong privacy, or seamless, agentic workflows spanning across its apps—then that uplift will be marketable and defensible for the firm. Meta doesn’t have an AI issue since it didn’t get language processing right up to now. To date, Meta has had an AI product issue. It has world-class integrated research and delivery and is unable to convert it into a product that customers choose and that pays the electric power bills. Eventually, Meta’s infrastructure bet might turn out to be fine. However, until Meta puts its signature on the product and publishes the numbers, buyers will see “frontier” as a cost center rather than a stimulus.