Meta Bets $145 Billion on AI Scaling—Can Superintelligence Pay Off

The Bet Is Back on Scaling

Mark Zuckerberg did not mince words on Meta’s third-quarter earnings call Wednesday. Meta will build superintelligence through scaling, he told analysts, and the company is methodically climbing toward it. Larger models are already in training. Smaller, distilled versions will follow—cheaper, faster, and designed for personal agents like Muse, which already has more than 10 million users in the U.S. and Canada.

The broader implication is what matters here. Meta is not hedging its AI future on partnerships or acquisitions. It is placing its entire strategic weight behind one thesis: that raw computational scale, deployed at industrial level, is the path to general intelligence—and that the cost of that intelligence can eventually be driven down so dramatically it becomes the default interface for personal computing.

This is a high-stakes recommitment to a strategy that many in the industry questioned after the initial wave of generative AI hype. The market, for now, is buying it.

The Money Behind the Words

Meta reported third-quarter revenue of $64.6 billion, surpassing analyst estimates of $63.37 billion. Adjusted earnings came in at $7.40 per share against $5.74 expected. On the surface, these are strong numbers. But the forward guidance tells a more consequential story.

Meta now expects full-year 2026 expenses of $175 billion to $179 billion, a meaningful upgrade from its prior range of $165 billion to $169 billion. Included in that is a $9.9 billion legal charge. Capital expenditures are projected to reach $135 billion to $145 billion—a figure that would make Meta one of the largest corporate spenders on infrastructure in history, and nearly double the capex commitments reported by other tech giants for the same period.

CFO Susan Li stated the company will continue investing aggressively to support both its core business and emerging AI initiatives. To finance that spending without diluting voting control, Meta announced it will introduce a new class of non-voting shares in January. Zuckerberg keeps his super-voting power. The company gains a capital-raising mechanism that sidesteps the governance friction that has slowed similar moves at other companies.

It is a structural play as much as a financial one.

The Scaling Thesis, Tested

Zuckerberg’s core argument is straightforward enough in theory: build increasingly capable frontier models, then compress them into efficient versions that run cheaply on consumer devices. The frontier models prove the ceiling. The distilled models drive adoption and monetization.

Muse is the proof of concept so far. Ten million users is not a rounding error for a personal AI agent, but it is not yet a defensible moat. Google, Apple, and OpenAI are all building competing personal assistant experiences. Amazon has Alexa historically; Microsoft has Copilot integrated across Office. The window for Meta to establish dominance in this category is narrow and closing.

Zuckerberg acknowledged the tension. He expects Meta to deliver current-generation intelligence at a fraction of the cost within a year in smaller models. But he also admitted most people will not want to use that lower-tier intelligence when it arrives. The expectation is that the distilled models will be good enough for everyday tasks while the frontier models handle the hard problems—but that gap between the two tiers may widen before it narrows.

Who Wins, Who Loses

The immediate winner is the thesis itself. Every quarter that Meta delivers earnings beats while simultaneously announcing larger capex plans, the narrative that AI spending is unsustainable erodes. Investors are signaling they believe the money will eventually pay off—or at least, that Meta’s advertising engine is strong enough to fund the experiment until it does.

OpenAI and Google face the hardest competitive pressure. Both are already ahead in frontier model capability. But Meta’s strategy of open-sourcing its models—Llama, deeply integrated into the developer ecosystem—creates a different kind of lock-in. Developers who build on Llama-based architectures are harder to poach than users of a closed API. Google’s closed approach gives it tighter control but less organic ecosystem growth. OpenAI’s partnership model with Microsoft is powerful but creates dependency on another company’s infrastructure decisions.

On the labor and infrastructure side, the demand for AI-specific chips, data center construction, and energy supply will intensify. Companies in those supply chains benefit directly. The environmental cost of $145 billion in capex, much of it directed toward compute infrastructure, is a question the market has not adequately priced in.

The Regulatory Question Nobody Is Asking

There is a quieter dimension to this spending cycle. No major AI lab in the world is operating without regulatory scrutiny. The EU’s AI Act is already in effect. The U.S. is moving toward federal oversight. China has imposed its own guardrails. Meta’s decision to open-source frontier models is partly a strategic move to build goodwill with regulators, but it also invites scrutiny of exactly how much influence an open model can have when deployed at scale by bad actors.

If Meta achieves superintelligence through scaling, the regulatory response will be swift and likely severe. The company is racing toward a finish line that governments are still figuring out how to define. That mismatch between technological velocity and regulatory capacity is the single greatest unpriced risk in Meta’s AI strategy.

What Happens Next

Meta’s stock rose 2.17 percent in after-hours trading to $712.65. Momentum sits in the 59th percentile; growth in the 74th. The market is rewarding the earnings beat and the clarity of vision, not yet pricing the execution risk.

The next twelve months will be decisive. Meta needs to demonstrate that its frontier models are genuinely closing the gap with OpenAI and Google, not just matching them on paper. It needs to show that distilled models can deliver enough value to justify the massive capex outlay. And it needs to do all of that while navigating an increasingly hostile regulatory environment in multiple jurisdictions.

Zuckerberg’s framing is simple: methodical scaling, step by step. The reality is far more complex. Meta is attempting to spend its way toward a capability that may not respond cleanly to brute force. If the scaling hypothesis holds, Meta becomes the dominant force in personal AI. If it does not, the company has committed a staggering amount of capital to a strategy that may have hit diminishing returns.

The earnings were strong. The bet is enormous. The clock is ticking.