Meta's $145 Billion Question: What Exactly Is All That Spending For?

Omniscient

Meta's first-quarter results, reported Wednesday, were genuinely strong. Revenue of $56.31 billion grew 33% year over year - the fastest pace since 2021.^[1] Operating income rose 30% to $22.9 billion. Net income hit $26.8 billion, boosted by an $8 billion tax benefit tied to last year's federal tax legislation. By any conventional measure, the business is in excellent health.

Then came the guidance. Meta told investors its 2026 capital expenditures would land between $125 billion and $145 billion - up from a prior range of $115 billion to $135 billion - citing higher component prices and "additional data center costs to support future-year capacity."^[1] The stock fell more than 6% in after-hours trading.^[2] The reaction was not a panic; it was a pointed question.

Why is Meta's capex raise different from Google's or Amazon's?

All four hyperscalers reported Q1 earnings on the same day, and all four are spending at historic scale on AI infrastructure. Alphabet and Amazon also raised their capex outlooks - and their stocks moved up after hours.^[2] The divergence in investor reaction is the most instructive data point of the earnings cycle. The difference is not the size of the commitment; it is what the spending is visibly attached to.

Alphabet's Google Cloud grew 63% year over year in Q1, and backlog figures give analysts a concrete sense of committed future revenue.^[3] Amazon Web Services reached a $150 billion annualized run rate in Q1 2026, with AI-specific services - including a custom silicon business that itself surpassed a $20 billion annual run rate - accounting for an increasing share of growth. Both companies can point to infrastructure spending as a direct input to a cloud business that bills by the hour. The capex-to-revenue feedback loop is observable.

Meta cannot offer the same loop - at least not yet. Advertising accounted for 98% of the company's $200 billion in 2025 revenue.^[4] AI has improved ad targeting and recommendation quality, and those improvements have contributed to the revenue growth Meta reported Wednesday. But the company has not disaggregated how much of its growth is attributable to AI versus platform scale, and it has not articulated a new revenue line that the $145 billion will eventually produce. Melissa Otto, head of Visible Alpha Research at S&P Global, put it directly: "It raises this question about what is the real ROI on all this capex that they're spending. I think the investment community is getting a little frustrated at the amount of cash they're burning."^[2]

What did Zuckerberg say on the earnings call - and what did he not say?

Asked to identify specific signposts for measuring return on AI investment over the next 12 to 24 months, Zuckerberg replied: "That's a very technical question." He added that Meta does not have "a very precise plan for exactly how each product is going to scale month over month, or anything like that," but expressed confidence in "the shape of where these things need to be."^[2] That answer would be unremarkable from an early-stage startup. From a company preparing to nearly double its capital expenditure from the prior year, it is a different matter.

The CEO's stated theory of the case is familiar: build products that reach billions of people, then monetize them once they achieve scale. It is the formula that worked for Facebook, Instagram, and WhatsApp. But those platforms generated network effects that locked in users and created inventory for advertisers. AI assistants and agents are a more competitive and substitutable category - a user who finds Claude or Gemini more useful than Meta AI faces very low switching costs. Scale alone does not confer the same structural advantage here that it did in social networking.

Where is the $145 billion actually going?

Meta's capex increase breaks down along two axes. On the hardware side, the company is rolling out more than one gigawatt of custom silicon developed with Broadcom, alongside significant AMD deployments to complement new Nvidia systems.^[2] Zuckerberg cited "memory pricing" as a specific driver of rising costs - a reference to the HBM (high-bandwidth memory) market, where supply constraints tied in part to the Iran conflict have pushed prices higher across the industry.^[1] On the model side, the capex underpins Meta Superintelligence Labs, the research unit created after Zuckerberg recruited Alexandr Wang from Scale AI in a deal valued at $14.3 billion.

The labs produced their first public output earlier this month: Muse Spark, a proprietary model that marks a deliberate break from Meta's previous open-source Llama strategy.^[5] The pivot carries business implications that remain unresolved. Llama's open-weight approach cultivated a developer ecosystem even without a direct monetization path; Muse Spark's proprietary architecture requires Meta to compete for the same enterprise and developer dollars already dominated by OpenAI, Anthropic, and Google. "The only reason I would use Llama is that I could fine-tune it," said Joseph Ott, CEO of AI startup Samu Legal Technologies, when asked why he would adopt Muse Spark instead. "It's unclear what would make it stand out."^[4]

Morningstar analyst Malik Ahmed Khan framed the model launch charitably but honestly: "I think Meta had to show investors and operators they have been working on something of substance. That's the first step." The second step - monetization - remains uncharted.^[4]

What is the actual risk here?

Meta's financial position is strong enough that the spending is not existentially dangerous in the near term. The company generated $26.8 billion in net income in a single quarter. But context matters: that figure included an $8 billion one-time tax benefit; without it, diluted EPS would have been $3.13 lower.^[1] And Barclays analysts projected, following Meta's Q4 2025 earnings in January, that Meta's free cash flow could fall close to 90% in 2026 as capex escalates.^[6] The company has now raised its 2026 capital expenditure guidance for the first time since issuing it in January - each revision arriving with the same explanation (component costs, future capacity) and the same absence of new revenue specificity.

Compounding the pressure: Meta's daily active people count came in at 3.56 billion for Q1, a 4% year-over-year increase but a slight sequential decline from Q4 - missing the 3.62 billion Wall Street had projected.^[1] The company attributed the shortfall to internet disruptions tied to the Iran conflict and WhatsApp restrictions in Russia. Those are genuine external factors, but they underscore a ceiling on the platform growth that has historically underwritten Meta's ad business.

Meta spent $72.2 billion on capex in 2025. Its new 2026 guidance of up to $145 billion would exceed the combined total of what it spent in 2024 and 2025.^[2]

Meanwhile, the company announced last week that it is cutting approximately 8,000 employees - 10% of its global workforce - beginning in May, while simultaneously eliminating 6,000 open positions.^[6] CFO Susan Li noted on the earnings call that additional workforce reductions would be communicated internally in May. The pattern - escalating infrastructure spend alongside workforce contraction - is consistent across the hyperscalers this year, but Meta's version is starker because the AI revenue upside that would justify the combination remains largely theoretical.

What would change the investor calculus?

Two things, primarily. First, a concrete monetization path for Muse Spark beyond advertising improvement - whether paid API access, enterprise licensing, or an AI agent product with its own revenue line. Analysts at Citizens have argued that Meta's 3-billion-person user base is its genuine differentiator, but only if the company can convert that reach into something that bills separately from ad impressions.^[4] Second, a stable capex number. Investors can model a large, fixed commitment; they struggle to model a commitment that keeps growing with each quarterly call. The credibility cost of a fourth revision in Q2 would be substantial.

For now, Meta sits in an uncomfortable position: a company with an exceptionally profitable core business, a legitimate AI research operation, and a capex trajectory it cannot yet fully justify to the people funding it. That is not a fatal combination. But it is an increasingly expensive one to sustain on the strength of "we have a sense of the shape."

Industry