Meta Q3 2025: The $70 Billion CapEx Pivot That Turns AI Into Infrastructure
Meta’s Q3 signals a bold shift as it turns ad revenue into one of the world’s largest AI infrastructure plays.
Welcome to Global Data Center Hub. Join investors, operators, and innovators reading to stay ahead of the latest trends in the data center sector in developed and emerging markets globally.
Meta’s third quarter marked a structural turning point. What was once a social and advertising platform is now transforming into a global compute operator. The company reported $51.2 billion in revenue, up 26% year-over-year, but beneath those strong results sits a more profound evolution a $70–72 billion CapEx program for 2025, and guidance that 2026 spending will be “notably larger.”
The magnitude of this investment positions Meta alongside hyperscale peers like Microsoft and Amazon, not as a platform, but as an AI infrastructure builder. The company is redeploying its immense advertising cash flow into the physical foundation of AI GPUs, custom silicon, and multi-region data centers built for liquid cooling, high-density compute, and sovereign-cloud compliance.
This is a decisive shift from optimizing algorithms to owning the energy, hardware, and real estate that determine AI capacity at planetary scale.
Earnings and Infrastructure Strategy
Meta’s core business remains extraordinarily profitable, providing the fuel for its infrastructure buildout. Advertising revenue reached $50.08 billion, a 26% year-over-year increase, supported by 14% higher ad impressions and a 10% rise in average pricing. These results demonstrate that Meta’s use of AI in ad targeting and conversion modeling is already translating into measurable returns.
Operating income climbed to $20.5 billion, maintaining a 40% margin despite surging expenses. GAAP net income fell to $2.7 billion, or $1.05 per share, primarily due to a non-cash, one-time $15.9 billion tax charge related to new U.S. tax legislation. Excluding this, adjusted EPS would have been approximately $7.25, with adjusted net income of $18.6 billion, underscoring the strength of the underlying business.
Reality Labs generated $470 million in revenue, up 74% year-over-year, though it posted an operating loss of $4.4 billion. Beyond headsets, this division now plays a strategic role in Meta’s AI roadmap: its AR glasses and spatial devices act as data-capture interfaces, feeding multimodal training pipelines essential for next-generation AI models.
CapEx for the quarter totaled $19.37 billion, bringing the year-to-date total above $48 billion. Management confirmed that full-year spending will close within the $70–72 billion range, primarily allocated toward GPU cluster expansion, new high-efficiency data centers, and the continued rollout of Meta’s in-house AI accelerator chip (MTIA v2). The pipeline now includes several gigawatt-scale campuses under construction in the United States, Europe, and Asia each designed to support AI workloads requiring advanced cooling and ultra-high power density.
Meta is also supplementing its owned infrastructure with leased cloud capacity, using third-party hyperscalers to bridge near-term compute gaps. This hybrid model indicates a strategic emphasis on time-to-compute the new determinant of competitive advantage in AI infrastructure.
Regional Expansion and Site Selection
Meta’s growth remains geographically broad, with strength across all major regions. However, the pace of revenue globalization is outstripping infrastructure localization. While advertising growth in Asia-Pacific (+31%) and Rest of World (+32%) now exceeds that of developed markets, much of Meta’s compute footprint remains concentrated in the U.S. and Northern Europe.
This imbalance is driving a quiet but deliberate reconfiguration of Meta’s global data center network. The company has accelerated site acquisitions and energy negotiations in Southeast Asia, India, LATAM, MENA, and Africa, marking the emergence of a new regional diversification strategy.
In Southeast Asia, Meta is expanding its presence beyond Singapore through projects in Johor, Malaysia, and Batam, Indonesia areas offering lower grid congestion and proximity to undersea cable routes. Thailand is under evaluation for AI training infrastructure co-located with renewable-energy zones.
In India, new development zones in Tamil Nadu and Maharashtra are being assessed for hyperscale builds linked to government digitalization initiatives. Discussions with renewable-power producers suggest Meta could secure multi-gigawatt PPAs to anchor its first AI campus in South Asia.
In Latin America, planning activity is centered around Querétaro, Mexico, and São Paulo, Brazil, where energy reforms and fiber interconnection corridors are attracting global hyperscalers. Local sources indicate Meta has been exploring joint ventures with infrastructure funds for phased development tied to sovereign-cloud requirements.
Across Africa and MENA, preliminary site evaluations are underway in Kenya, South Africa, and Saudi Arabia’s NEOM region, supported by renewable energy partnerships with utilities and developers in the Gulf and Southern Africa. These early-stage efforts suggest a clear intent to establish AI-ready compute zones aligned with data-sovereignty and clean-power availability.
This pattern reveals a consistent logic: Meta is prioritizing markets that combine population-scale data, renewable generation potential, and regulatory predictability. Each region adds a node to an emerging global compute grid designed to serve both AI training and low-latency inference across continents.
Power, Land, and Capacity Planning
Energy security now defines Meta’s infrastructure economics. The company’s 2025–2026 expansion program involves securing over 5 gigawatts of new capacity, supported by renewable PPAs, substation partnerships, and grid interconnection rights across multiple continents.
Meta is implementing an aggressive parallel development model acquiring land and grid rights before finalizing construction to shorten time-to-power and de-risk delivery schedules. Many of its new sites are being engineered for liquid immersion and direct-to-chip cooling, technologies that can double compute density while reducing water usage per megawatt by up to 50%.
The new builds will extend Meta’s total owned capacity beyond 10 gigawatts by late 2026, with active projects under development in at least nine countries. Management’s commentary also suggests a focus on energy diversification, integrating solar, wind, and battery systems to offset rising GPU demand and decarbonize its compute footprint.
Forecast and Strategic Outlook
Meta’s third quarter was less an earnings report and more a declaration of infrastructure intent. The company is no longer allocating capital primarily to product development but to the physical systems that power AI intelligence at scale.
With CapEx climbing toward $70 billion in 2025 and set to grow further in 2026, Meta is now one of the top three private-sector investors in digital and energy infrastructure globally. The near-term impact will be margin compression from depreciation and operating expenses, but the long-term benefit is strategic autonomy control over compute, energy, and latency.
Meta’s evolution from an advertising platform into a compute utility signals a new era in corporate infrastructure. The company is betting that ownership of the world’s AI capacity will yield returns far greater than optimizing ad yield or engagement.
The question now is not whether Meta can afford the buildout but whether any competitor can match its speed, scale, and energy access.