Running the world’s largest advertising platform costs more every year — not because of inefficiency, but because the ambition keeps expanding. Meta’s decision to raise its 2025 expense budget above $114 billion, a 30%+ jump from 2024, was not a sign of bloat. It was a deliberate declaration that the company intends to compete on every frontier that matters: AI, hardware, content, infrastructure, and the long-horizon bets that won’t pay off for years.
Meta’s expense strategy reflects CEO Mark Zuckerberg’s conviction that AI will be transformative for Meta’s advertising business and that the window to invest at scale is now. The $60-65 billion capex commitment, if sustained, would be approximately twice Meta’s 2024 capex of $38 billion. The magnitude of this commitment has generated significant investor debate about returns, timing, and risk.
The Capex Breakdown
Meta’s $60-65 billion 2025 capex is allocated primarily to AI infrastructure: data center construction, AI training and inference hardware (primarily NVIDIA H100 and H200 GPUs, with growing investment in custom silicon), and network infrastructure. Meta has been building large-scale GPU clusters for AI model training, including a cluster with over 100,000 H100 GPUs announced in 2024.
Meta’s custom AI silicon development, through its MTIA (Meta Training and Inference Accelerator) chip program, aims to reduce dependence on NVIDIA GPUs for inference workloads. MTIA chips are designed specifically for Meta’s recommendation system and advertising inference tasks. Successfully deploying MTIA at scale would reduce Meta’s capex per inference by improving cost efficiency, similar to how Google’s TPUs provide cost advantages in Google’s AI infrastructure.
Data center construction has long lead times—typically 2-3 years from site selection to operational capacity. Meta’s 2025 capex investment in data center construction will primarily add capacity in 2027-2028. This lag between investment and capacity creates a cash flow timing challenge: Meta is spending now for returns that materialize in 2-3 years. This is the nature of long-cycle infrastructure investment, and Meta has sufficient balance sheet strength to sustain it.
Operating Expenses Beyond Capex
Meta’s 2025 operating expense guidance of $114-119 billion includes capex (roughly $60-65 billion) and operating expenses (roughly $54-59 billion). The operating expense component covers personnel costs, research and development, data center operations, cost of revenue, and general administrative expenses.
Personnel represents the largest operating expense after infrastructure. Meta employs approximately 70,000-75,000 people globally, with engineering and product talent the majority. Meta’s compensation packages are competitive with any US technology company, with senior engineers earning $400,000-$800,000+ in total compensation (base salary, bonus, and equity). The concentration of talent in AI research, AI infrastructure, and AI product development has grown significantly as Meta’s AI strategy has expanded.
Research and development spending, primarily on AI research through Meta’s FAIR (Fundamental AI Research) lab and applied AI teams, is substantial. FAIR has produced significant AI advances including the LLaMA open-source language model series. Open-sourcing AI models is a deliberate Meta strategy—creating an ecosystem that reduces the competitive advantage of closed model providers (OpenAI, Anthropic) by making comparable capabilities available freely, while Meta focuses on proprietary infrastructure advantages that cannot be open-sourced.
Revenue vs. Expense Growth Rate Comparison
The key question for Meta’s 2025 financial performance is whether revenue grows faster than expenses. Meta’s 2024 revenue was approximately $165 billion, with consensus 2025 estimates in the $180-190 billion range—roughly 10-15% growth. Expense growth of 30%+ creates an operating leverage headwind: expenses growing faster than revenue temporarily compresses operating margins.
Meta’s operating margin was approximately 41% in 2024, which was historically high. The 2025 expense guidance implies compression to approximately 35-38% operating margin, depending on revenue execution. This compression is intentional and reflects Zuckerberg’s view that the current AI investment cycle is the right time to invest aggressively, even at the cost of near-term margin.
The historical precedent is instructive. Meta underwent a similar “investment phase” in 2021-2022, when Reality Labs losses and increased infrastructure spending compressed margins. During that period, Meta’s stock fell significantly. Margins recovered substantially in 2023-2024 when the investment phase moderated and the Advantage+ advertising improvements drove revenue outperformance. Investors who correctly identified the investment phase as temporary were rewarded.
AI Expense ROI Framework
Meta has articulated a specific ROI framework for its AI investment: AI improvements to advertising targeting and delivery directly increase the revenue Meta can generate from each ad impression. If Advantage+ campaigns improve ROAS by 20-25% for advertisers, advertisers will increase budgets on Meta, increasing Meta’s revenue per user. This virtuous cycle—better AI, better advertiser ROI, higher advertiser spend, higher Meta revenue—is the core business case for $60-65 billion in AI capex.
Meta has provided some metrics supporting this framework. Advantage+ Shopping Campaigns have driven measurable revenue contribution. AI-powered Reels recommendations have increased time spent on Reels, creating more ad inventory. AI content discovery has increased engagement across Facebook and Instagram. Each of these improvements has a direct line to advertising revenue, validating the investment thesis in the current cycle.
The uncertainty is in the marginal return on the next increment of AI investment. As AI models improve, the most impactful gains come first. Later improvements require more compute for less performance gain. If Meta’s $60-65 billion 2025 capex produces only marginal improvements in advertising performance, the investment thesis would be challenged. If it enables step-function improvements in new capabilities (AI agents, AR advertising, commerce AI), the returns could exceed expectations.
2026 Expense Outlook
Meta has provided less specific guidance for 2026 expenses but has indicated that the elevated capex levels represent a sustained commitment rather than a one-year surge. If Meta’s 2025 capex of $60-65 billion becomes the new baseline, 2026 capex could be $70-80 billion as AI infrastructure requirements continue growing.
Operating expenses in 2026 will depend on headcount evolution, infrastructure cost management (where custom silicon MTIA chips could reduce costs), and the success of cost efficiency initiatives from Meta’s 2023 “Year of Efficiency” program. Meta reduced headcount by 20%+ in 2023, creating a leaner cost structure. Returning to pre-efficiency headcount growth would meaningfully increase operating expenses.
The 2026 expense trajectory is also influenced by regulatory costs. Privacy compliance, antitrust monitoring, and content moderation requirements impose costs that are growing as regulatory scrutiny of social media platforms intensifies globally. These regulatory operating costs are difficult to predict but represent a meaningful and growing component of Meta’s expense base beyond AI investment.
