Meta's $145B CapEx Bet: What Platform Leads Should Learn

Meta just posted $59.4 billion in Q2 advertising revenue, up 27% year over year, while quietly raising the floor of its full-year capital expenditure range to $130 billion. That single juxtaposition, a monster ad business funding an even more monstrous infrastructure build, is the number every platform lead spending real money on paid acquisition should be staring at this quarter. The question isn't whether Advantage+ works. It's who pays for the compute, and when the price shows up in your CPMs.

The Numbers

Total Q2 revenue came in at $60.8 billion, up 28% year over year for the period ending June 30, as Marketing Dive reported. Revenue beat Wall Street expectations. Earnings missed. Shares slid. That's the shape of the print: top-line strength, margin compression, and a market that has stopped giving Meta the benefit of the doubt on AI CapEx.

The forward guide is the more interesting data point for anyone modeling media budgets. Meta guided Q3 revenue to a range of $61 billion to $64 billion. CFO Susan Li flagged two headwinds explicitly: the company is lapping a period of strong ad impressions growth, and it could see additional pressure from European policy changes allowing less personalized ads. Translation for performance marketers: the inventory tailwind that padded 2025 comps is running out, and the addressability regime in the EU is degrading further.

Inside the ad stack, Advantage+ reached a $75 billion annual revenue run rate in Q2. That's not a rounding error. That's the majority of Meta's ad business now flowing through an AI-mediated buying surface where the advertiser cedes granular targeting control in exchange for algorithmic optimization. Meta also rolled out the Meta Generative Recommender in Q2, which Li described as a shift in how the platform runs ads, using LLMs to reason about ad content and user preferences together rather than scoring every ad individually.

Outside of ads, the "Other" line for the family of apps was up 73% year over year and crossed $1 billion for the first time, driven by WhatsApp paid messaging and subscriptions. Real growth rate, small absolute number relative to the base. On the cost side, the full-year CapEx range was narrowed to $130 billion to $145 billion, up from a prior floor of $125 billion. The ceiling didn't move. The floor did. That tells you management has conviction that the low end of spending is no longer plausible.

Forrester's Mike Proulx put it bluntly: Meta's ad business is still a monster, but everything else got more expensive, and what the quarter generated in cash almost all got eaten by AI infrastructure spending.

What's Actually New

The Meta Generative Recommender is the piece of this print that engineering leaders should actually care about, and it's the piece that got the least airtime in the mainstream coverage. Meta is describing a shift from a scoring architecture, where every candidate ad is evaluated against a user, to a generative reasoning architecture where LLMs jointly consider ad content and user preferences to predict the best match. If that description is honest, and not marketing gloss on the same two-tower retrieval systems the industry has run for a decade, it's a genuine change in the auction mechanics.

Here's what it means downstream. When the buying surface becomes generative, the advertiser's creative asset library and product feed matter more than bid strategy or audience segmentation. The lever shifts from "how do I target" to "what do I feed the model." That's a very different skill set on the growth team. It also collapses the value of the middle layer of independent tooling that has spent the last five years arbitraging Meta's audience APIs. If the model reasons about ad content directly, third-party audience enrichment gets squeezed from both ends.

The second genuinely new element is the CapEx floor moving. Meta didn't raise the ceiling of its spending guidance, it raised the floor. In corporate finance terms, that's management removing optionality on the downside. They are telling investors: we cannot spend less than $130 billion this year even if we wanted to, because the infrastructure commitments are already locked in. For anyone negotiating vendor contracts with Meta, or any hyperscaler for that matter, that's a signal about pricing power. Meta needs the ad revenue to keep pace with a spending schedule that no longer bends.

The Marketing API surface is where these architectural shifts land in your codebase. Teams that built pipelines assuming stable campaign structures and manual optimization loops are going to spend the next 18 months rewriting them to feed Advantage+ style objectives.

What's Priced In for Performance Marketing

The market already understood that Meta is dominant in social advertising. Some researchers expect Meta could surpass Google in ads revenue for the first time in 2026. That's priced in. Zuckerberg's line that Meta is reporting faster year-over-year revenue growth on a dollar basis than any other ad business is a victory lap on a race that finished last quarter.

What's not fully priced in, at least not in how most performance marketing teams are staffed, is the operational consequence of Advantage+ eating the buying surface. If your paid social team is still organized around campaign structure specialists, audience segment analysts, and manual bid managers, you are staffing for a 2022 workflow. The $75 billion run rate on Advantage+ tells you the platform has already made the decision for you. Your headcount composition should reflect that: fewer campaign operators, more creative production capacity, more measurement engineers who can validate incrementality against a black box.

Also underpriced: the European personalization headwind Li flagged. This isn't a one-quarter blip. It's the beginning of a durable structural change in what data Meta can use in one of the highest-CPM regions in the world. Teams with heavy EU exposure need to be modeling a permanent efficiency haircut on that geo and reallocating either budget or measurement expectations accordingly. The Privacy Sandbox conversation on the open web is a preview of what platform-side addressability starts to look like when regulators keep pressing.

Contrarian View

The consensus reading of this print is that Meta is a monster ad business burdened by an AI spending problem, and that investors are right to be nervous. I'd push back on half of that.

The CapEx isn't the risk. The risk is what happens if the Generative Recommender actually works as described. If Meta genuinely compresses the auction from a scoring problem into a reasoning problem, the platform's ability to extract advertiser surplus goes up, not down. Every efficiency gain Meta reports as "compounds performance gains for advertisers" is also, mechanically, a reduction in the arbitrage available to sophisticated buyers. The gap between a mediocre advertiser and a great one on Meta narrows every time the model gets better at reasoning. That's bullish for Meta's take rate and bearish for the agencies and in-house teams whose value proposition rests on knowing the platform better than the platform knows itself.

The CFO who should be nervous isn't Meta's. It's the CFO of every performance-dependent business that has assumed Meta CPMs would stay flat because competition among advertisers keeps prices in check. If the model closes the skill gap, more advertisers bid effectively, and prices go up structurally. That's the scenario nobody in the earnings coverage is pricing in.

Who Should Be Asking What This Week

The VP of Engineering at any mid-market fintech or iGaming operator running material paid social spend should be asking their Head of Growth this week: what percentage of our Meta budget is already flowing through Advantage+ objectives, and what does our creative production pipeline look like relative to that share? If the answer is "most of the budget, and our creative pipeline is a two-person team plus a freelance editor," you have a staffing mismatch that will show up in blended CAC within two quarters. The build-vs-buy question on creative automation tooling just got more urgent, and the hiring market for people who can operate at the intersection of production and measurement is going to tighten fast.

Key Takeaways

• Meta's $130B to $145B CapEx range, with a raised floor, signals management has locked in infrastructure commitments that require ad revenue to keep growing at current pace. Vendor pricing power sits with Meta, not with buyers.
• Advantage+ at a $75B run rate means the majority of Meta ad spend now flows through AI-mediated buying. Growth teams organized around manual campaign management are staffing for a workflow that no longer exists.
• The Meta Generative Recommender, if it delivers on the LLM-based reasoning architecture Li described, shifts the advertiser's competitive edge from targeting expertise to creative asset quality and product feed hygiene.
• European personalization headwinds flagged in Q3 guidance are structural, not cyclical. Teams with EU exposure should model a durable efficiency haircut.
• The contrarian risk isn't that Meta's AI investments fail. It's that they succeed, close the skill gap between advertisers, and push CPMs structurally higher for everyone.