Meta's $70 Billion Bet: The Market Is Mispricing the Capex Surge

Most investors still think of Meta as a social media company that sells ads. That framing is three years out of date, and the gap between the perception and the reality is where the valuation discount is hiding.

Meta operates platforms with three to four billion daily active users across Facebook, Instagram, WhatsApp, and Threads. The advertising system that monetises these users generated $201 billion in revenue in 2025, a 22 percent increase from $164.5 billion in 2024. This is not a business approaching saturation. It is a business that keeps finding new ways to extract value from an audience and data asset that competitors cannot purchase, rent, or recreate at any price.

Alongside the advertising machine, Meta is running one of the most strategically sophisticated AI buildouts in the technology sector. The Llama family of open-source large language models forms the foundation of this strategy. By making frontier AI capabilities freely available, Meta positions itself as the default infrastructure layer for developers building on open-source AI, seizing the ecosystem in ways a proprietary model strategy cannot achieve. This is not philanthropy. It is competitive strategy executed at the scale only a company generating $115.8 billion in annual operating cash flow can afford to run.

The business has three distinct components that the market tends to bundle: the advertising engine, the AI infrastructure buildout, and Reality Labs. Each deserves a separate analytical lens. The market is applying one discount rate to all three, and that averaging is where the mispricing originates.

The 2022 operating margin collapse from 39.6 percent to 24.8 percent remains the defining moment in Meta's recent financial history. The company had overspent on headcount and infrastructure during the pandemic-era growth surge. When revenue stalled at $116.6 billion, roughly flat against 2021's $117.9 billion, operating income fell from $46.8 billion to $28.9 billion. That is a $17.9 billion decline in a single year.

What followed is what defines the current investment thesis. Meta's Year of Efficiency in 2023 eliminated more than 21,000 positions and reset the cost structure. Operating income recovered to $46.8 billion on $134.9 billion in revenue, restoring the 2021 dollar level on a meaningfully higher revenue base. By 2024, operating income had reached $69.4 billion at a 42.2 percent margin, the highest in the company's public history. The 2025 result of $83.3 billion in operating income on $201 billion in revenue reflects a 41.4 percent margin, confirming the expansion has stabilised at historically elevated levels rather than mean-reverting.

The earnings beat streak compounds this picture. In the six completed quarters through December 2025, Meta exceeded consensus EPS estimates by an average of 15.2 percent. The beat magnitude accelerated as the recovery deepened: from 8.4 percent in the June 2024 quarter, to 14.0 percent in September, to 20.1 percent in December 2024, to 23.2 percent in March 2025. The December 2025 quarter came in at $8.88 against an $8.18 consensus. Analysts have consistently underestimated this business through the recovery cycle, and the forward estimate of $6.61 for the March 2026 quarter sits in a pattern where beats have become the baseline expectation.

EBITDA grew from $37.6 billion in 2022 to $104.5 billion in 2025, a 178 percent increase in three years. That trajectory on a $200 billion revenue base has very few comparables in the history of large-cap technology.

The $69.7 billion in capital expenditures Meta deployed in 2025 is the number that has divided the investment community. It represents an $32.4 billion year-over-year increase, an 87 percent jump from $37.3 billion in 2024. This spending is concentrated in AI data center infrastructure: custom silicon, network buildouts, and the compute capacity required to run AI inference across four billion daily active users.

The bear case writes itself. Meta is spending at a rate that becomes reckless if AI monetisation timelines slip. The FCF yield has compressed. Management has limited forward visibility into the ultimate return profile of this infrastructure. And the company has a history of expensive bets that did not pay off on the original schedule. Reality Labs has consumed tens of billions across multiple years with limited consumer adoption to show for it.

The more useful question is what the counterfactual looks like. Meta's advertising business runs on AI systems for targeting, measurement, and creative generation. Advantage+ AI-powered ad tools have measurably improved advertiser return on ad spend, which is the reason advertisers are increasing budgets on Meta platforms rather than diversifying away from them. The infrastructure buildout is not speculative in the sense that it underpins revenue streams that already exist and are already growing at 22 percent annually. The incremental capex is buying more capacity for a system that is already generating returns on existing capacity.

The April 2026 confirmation that CoreWeave secured a $21 billion AI cloud computing contract with Meta underscores the scale and contractual reality of the commitment. This is not an internal planning assumption, it is externally contracted capacity with a defined counterparty and timeline. The investment is locked in. The question is whether the returns will justify it, and the advertising revenue trajectory suggests they are already beginning to.

At $115.8 billion in operating cash flow, Meta can absorb $69.7 billion in capex and still generate $46.1 billion in free cash flow. The FCF yield compression is a choice, not a financial constraint. A company that cannot cover its capital program from operating cash flow is in a different situation than Meta. Conflating the two has been the source of most of the market's mispricing.

In parallel with the AI infrastructure buildout, Meta has maintained a meaningful shareholder return program. Buybacks totalled $26.2 billion in 2025, and the share count has been reduced from a peak of approximately 2.89 billion in 2020 to approximately 2.19 billion outstanding today. That is a 24.2 percent reduction in the share base over five years, a level of buyback discipline that sits well above the large-cap technology median.

2024 marked Meta's first dividend payment: $5.1 billion distributed in 2024, and $5.3 billion in 2025. This is a signal rather than a commitment to income-oriented investors. The dividend represents roughly 11.5 percent of 2025 free cash flow, manageable at current FCF levels and eliminable without financial stress if the infrastructure program requires it. The initiation itself is meaningful, indicating that management views the capital generation as sustainable enough to promise regular distributions.

Stock-based compensation reached $20.4 billion in 2025, up from $16.7 billion in 2024. At approximately 10 percent of operating cash flow, SBC is not alarming in isolation, but it meaningfully offsets buyback progress in terms of net dilution. The share count reduction from 2.89 billion to 2.19 billion over five years looks more impressive before accounting for the SBC being issued in parallel. Investors who track dilution-adjusted returns should weight this accordingly.

The balance sheet carries cash of $35.9 billion against total debt of $58.7 billion, leaving Meta in a modest net debt position. This represents a structural change from the net cash fortress Meta operated from prior to 2022. It is not concerning given the cash generation profile, but it does limit financial flexibility at the margin compared to the prior capital structure.

Meta's advertising moat is better understood as a data flywheel than an audience scale advantage. The value is not that three to four billion people use the platforms. It is that their behaviour across those platforms generates a training and targeting dataset that no competitor can purchase, reconstruct, or replicate at any cost. Every incremental engagement compounds the advantage.

The open-source Llama strategy reinforces this position in a way that most competitive analyses miss. By releasing frontier models freely, Meta establishes itself as the default infrastructure layer for developers building on open-source AI foundations. This accelerates adoption of Meta's technical conventions, tooling, and ecosystem preferences in ways that proprietary model strategies cannot achieve. OpenAI's reported plans to introduce advertising into its products, which surfaced in April 2026 coverage, confirm that the ad model is the gravitational centre of AI monetisation. Meta is better positioned to execute this than any closed-model competitor precisely because the underlying infrastructure already exists at scale.

The competitive risk is more plausible from TikTok and YouTube than from AI newcomers. Both platforms compete directly for user attention and advertiser budgets in video and short-form content. Meta's response has been Reels, which has grown meaningfully as a revenue contributor. The platform mix is more diversified than the headline Facebook narrative suggests.

Reality Labs remains the moat question mark. Cumulative losses exceed $50 billion across multiple years, and consumer adoption of augmented and virtual reality hardware has not met the timelines originally articulated. The honest assessment is that Reality Labs is a long-duration option on a hardware category that may or may not materialise within the investment horizon relevant to current shareholders. The advertising business alone justifies the current valuation at current multiples. Reality Labs is either a bonus or a persistent drag, and the market has not resolved which it is yet.

At a trailing price-to-earnings ratio of 26.7 times, Meta trades in line with the broader S&P 500 index, despite growing revenue 22 percent in 2025 on a base of $164.5 billion. A business with 41 percent operating margins, seven consecutive earnings beats averaging 15 percent above consensus, $115.8 billion in annual operating cash flow, and institutional ownership of 78.7 percent does not typically trade at market multiples without a specific reason.

The EV/EBITDA multiple of 13.8 times is the more revealing figure. With $104.5 billion in EBITDA in 2025 and a trajectory that suggests further growth in 2026, this multiple implies that the market expects EBITDA growth to decelerate substantially and soon. The analyst community does not agree: 47 of the 68 analysts covering the stock hold strong buy or buy ratings, with only 2 at sell. The consensus price target of $860 implies approximately 21 percent upside from the April 2026 price level. This is not a divided or uncertain analyst view. It is a consensus that the stock is undervalued.

The most plausible explanation for the discount is the conglomerate penalty. The market is applying a single blended discount rate across three businesses with genuinely different return profiles: the advertising machine, the AI infrastructure buildout, and Reality Labs. Bundled together and averaged, the result is a discount that does not reflect the sum of the parts. This kind of pricing inefficiency is common in large conglomerates and typically persists until one of three things happens: the market re-rates the core business upward, the loss-making unit is separated or written down, or earnings compounding at a faster rate than the discount rate forces the multiple higher.

The 30-day sentiment trend for META through early April 2026 has remained broadly constructive, averaging 0.674 across the period, with only a brief dip to 0.365 during peak macro uncertainty around the March 25-27 tariff coverage cycle. The rapid sentiment recovery to above 0.65 in the first two weeks of April, despite ongoing macro uncertainty, reflects the market's recognition that Meta's revenue is driven by global digital advertising budgets rather than physical trade flows.