Recommended Free Tools
Meta’s AI infrastructure budget pays for an entire computing system, not just GPUs. The company is buying and leasing servers and accelerators, building and colocating data-center capacity, expanding high-speed networks, contracting for third-party cloud capacity, and paying the depreciation, finance-lease principal, operations and technical compensation needed to run it. Meta’s infrastructure serves AI projects such as Meta Superintelligence Labs as well as Facebook, Instagram, WhatsApp and the rest of its core business.
The figures behind Meta’s infrastructure buildout
Meta’s reported numbers describe different layers of the same buildout. They should not be added together as if each were a separate AI bill: capital expenditure, property-and-equipment purchases, lease obligations and contractual commitments overlap in timing and scope.
As an Amazon Associate I earn from qualifying purchases.
| Figure | What it represents | Time or measurement date |
|---|---|---|
| $72.22 billion | Capital expenditures, including principal payments on finance leases | Meta’s 2025 full year |
| $69.69 billion | Purchases of property and equipment, primarily servers, data centers and network infrastructure | Meta’s 2025 full year |
| $115–135 billion | Planned capital expenditures, including finance-lease principal payments | Meta’s 2026 outlook; forward-looking guidance |
| $103.77 billion | Lease obligations that had not yet commenced, mostly for data centers, colocations and network infrastructure | Balance as of Dec. 31, 2025 |
| $131.05 billion | Contractual commitments, mostly for third-party cloud capacity, servers, network infrastructure, data centers and Reality Labs consumer hardware | Balance as of Dec. 31, 2025 |
| $30.63 billion | Portion of those contractual commitments due in 2026 | As of Dec. 31, 2025 |
The $115–135 billion range is a plan, not money already spent. It also covers investment in Meta’s core business, so it is incorrect to label the entire range “AI spending.” The $69.69 billion property-and-equipment figure is narrower than total capital expenditure because the latter includes finance-lease principal payments.
What Meta actually buys
Servers and AI accelerators
Servers are the physical computers that train models, serve inference requests and run ordinary application workloads. AI accelerators—specialized processors used for machine-learning calculations—fit inside those server systems alongside CPUs, memory, storage and power equipment. Meta’s filings identify servers as a major property-and-equipment category, but they do not publish a dollar split between GPUs, other accelerators, CPUs, storage or complete server systems.
#1 Best Overall
- NVIDIA Volta GV100 Architecture — 4,608 CUDA Cores, 640 1st-Gen Tensor Cores delivering 14 TFLOPS FP32 and 112 TFLOPS deep learning performance for AI training, inference, HPC, and scientific computing workloads
- 32GB HBM2 ECC Memory — 900 GB/s Bandwidth — High-bandwidth memory on a 4096-bit bus with ECC error correction provides the memory capacity and throughput required for the largest AI models, simulations, and datasets
- PCIe 3.0 x16 Interface — 250W TDP — Standard PCIe Gen3 connectivity with passive cooling designed for enterprise rack server deployment in HPE ProLiant, Dell PowerEdge, and Supermicro platforms with adequate chassis airflow
- NVLink — Scale to 96GB Unified Memory — Connect two V100 GPUs via NVLink at 300 GB/s bi-directional bandwidth to scale GPU memory from 32GB to 96GB for larger AI training and HPC workloads
- Multi-Precision Computing — Supports FP64 (7 TFLOPS), FP32 (14 TFLOPS), FP16 (112 TFLOPS) and INT8 precision modes for flexible deployment across training, inference, and scientific simulation workloads
Training requires large groups of machines to work together for long periods. Inference requires enough machines to answer user requests reliably and at acceptable latency. The same fleet can be reassigned between those jobs as models, products and demand change, which is one reason a company-wide infrastructure number cannot be translated directly into a GPU-only number.
Data centers, colocations and long-term capacity
Compute hardware is useful only when it has a suitable home. Meta’s spending therefore includes new and expanded data-center facilities, equipment installed in those facilities and colocations—space leased inside facilities operated by another company. Long-term arrangements can appear as lease obligations before a site is ready to use.
The $103.77 billion in lease obligations that had not commenced at the end of 2025 was mostly tied to data centers, colocations and network infrastructure. It signals future capacity commitments, not a cash payment made entirely in 2025 and not a count of buildings. The obligation can cover a multiyear term and may begin when a facility or portion of it becomes available.
Network infrastructure
Large AI clusters depend on fast connections between machines. Network infrastructure links servers within a cluster, connects clusters to storage and ties data centers to Meta’s consumer services. It also carries traffic for Facebook, Instagram, WhatsApp, Reality Labs and other products.
Network equipment is included among the primary types of property and equipment in Meta’s 2025 disclosure. The filings do not provide a separate dollar total for switches, optical systems, cables, routers or related installation work, so those components cannot be assigned an evidence-based percentage of the budget.
Rank #2
- Powered by Radeon AI PRO R9700 - Supercharge you workflow with the cutting-edge RDNA 4 Architecture and 2nd-gen AI Accelerators.
- 32GB GDDR6 with 256-bit memory bus - Tackle larger, more complex projects without limits.
- PCIe Gen 5 - Unlock lightning-fast data transfers with PCIe Gen 5 support.
- GIGABYTE TURBO Fan Cooling System - Indented metal cover and blower fan increase airflow intake, while the vapor chamber, all copper heat sink, and metal frame offer efficient heat dissipation. Optimized airflow design allows for easy multi-GPU scalability.
- Double Ball Bearing Fan - Delivers superior heat resistance and rotational efficiency for better performance and a longer lifespan compared to conventional sleeve fans.
Third-party cloud capacity
Meta is using a hybrid model: it builds and operates owned infrastructure while contracting for capacity from third-party cloud providers. Cloud contracts can provide additional compute or storage without waiting for every owned facility to be completed. They can also let Meta adjust capacity as AI workloads change.
Those contracts are reflected in the $131.05 billion of contractual commitments reported at Dec. 31, 2025. Meta says the commitments are mostly for third-party cloud capacity, servers, network infrastructure, data centers and Reality Labs consumer hardware. The disclosure does not identify the amount attributable to cloud alone, the providers involved or the price per accelerator-hour.
Operations, depreciation and finance-lease payments
Infrastructure costs continue after hardware arrives. Meta’s operating expenses include running deployed infrastructure and depreciating assets over their useful lives. Depreciation is an accounting expense that spreads the cost of owned equipment and facilities across the periods in which they are used; it is not a new purchase each year, but it affects reported profit.
Finance-lease principal payments are included in the capital-expenditure guidance and in the $72.22 billion 2025 capex figure. They represent repayment of financed infrastructure rather than a separate class of AI hardware. Meta’s chief financial officer said that most expected 2026 expense growth would come from “infrastructure costs, which includes third-party cloud spend, higher depreciation, and higher infrastructure operating expenses.”
Technical employees
People are not property and equipment, but they are part of the cost of turning infrastructure into products. Meta says employee compensation is the second-largest contributor to expected 2026 expense growth, including AI and other technical hires. That category covers the engineers, researchers, data-center specialists and other technical staff needed to design systems, train models, operate services and integrate AI features.
Rank #3
- [Local AI Inference & 70B Model Ready] Equipped with the AMD Ryzen 7 PRO 8845HS processor, NEXUS is engineered for heavy local AI workloads. With a full-size GPU bay, it runs 70B LLMs natively without an internet connection. Ideal for AI developers and tech enthusiasts who need private environment for coding and model testing.
- [132TB Mass Storage with ZFS Integrity] Features a hybrid storage architecture (3×NVMe + 4×3.5" HDD) supporting up to 132TB. Utilizing the enterprise-grade ZFS file system and ECC memory, it prevents data corruption and bit rot—a must-have for professional photographers and video editors safeguarding 4K/8K RAW footage.
- [OpenClaw-Driven Automation Workflow] The built-in OpenClaw execution layer allows complex automated tasks to be processed locally. Even when offline, your backup schedules and AI file organization continue seamlessly. Say goodbye to monthly cloud subscriptions and high latency.
- [Dual 10GbE & USB4 Ultra-Connectivity] Experience server-class speeds with dual 10GbE ports and a 40Gbps USB4 interface. It enables multi-user real-time collaboration on large project files directly from the NAS, ensuring zero-lag editing for creative studios and production teams.
- [Open-Source ZimaOS for Total Privacy] Running on the fully open-source ZimaOS, NEXUS ensures your data stays physically on-premise with no backdoors. It acts as a "Digital Fortress" for privacy-conscious families and small businesses who demand absolute data sovereignty.
How to read capex, leases and commitments without double-counting
| Term | What it tells you | What it does not tell you |
|---|---|---|
| Capital expenditure | Investment recorded for long-lived assets and, in Meta’s reported figure, finance-lease principal payments | The exact amount spent on AI alone or on GPUs alone |
| Property-and-equipment purchases | Cash purchases of long-lived items such as servers, data centers and network infrastructure | A complete list of every AI-related cost or a processor-by-processor breakdown |
| Lease obligations not yet commenced | Future contractual payments for capacity that was not yet available at the reporting date | Cash already paid, active capacity already online or an annual expense amount |
| Contractual commitments | Future purchases and services Meta has committed to, including cloud capacity and infrastructure | A guarantee that every commitment is AI-related or that all amounts are due immediately |
| Depreciation and infrastructure operating expense | The recurring cost of using and running deployed assets | New capital investment in the period |
The categories can overlap. A server purchased in 2025 can be part of property-and-equipment purchases, contribute to capex, and later generate depreciation. A data-center arrangement can appear in a future lease obligation before it produces depreciation. A cloud contract can be a contractual commitment and later become an operating expense when capacity is consumed.
Free tools Windows power users keep installed
One-click scans. No signup required.
Owned infrastructure versus rented capacity
Why Meta builds and leases its own systems
Owned or dedicated facilities give Meta control over hardware configuration, network design, deployment schedules and long-term cost structure. Buying servers and building capacity can make sense for workloads expected to run continuously at very large scale. It also lets Meta design infrastructure around its own models and applications instead of accepting a provider’s standard configuration.
Leased data-center space and colocations extend that control without requiring Meta to own every building. They can provide power, cooling and physical capacity in locations where constructing a new campus would take longer.
Why Meta contracts with cloud providers
Third-party capacity can be available sooner than a new owned facility and can help cover peaks in demand, specialized hardware needs or temporary gaps during construction. It shifts some capital requirements toward contracted service spending, although the commitment can still be large and multiyear.
Meta’s disclosures establish that both approaches are being used. They do not establish the percentage of workloads running on owned servers versus third-party cloud, nor do they reveal whether one model is cheaper for a particular AI job.
The Tool Desk
Outbyte PC Repair FREERepair Windows errors before they cause bigger problemsFix Now →Outbyte Driver Updater FREEScan for outdated or missing drivers - takes under a minuteDriver Scan →Rank #4
- Professional AI & Creator Workstation: AMD Radeon AI PRO R9700 GPU with 32GB GDDR6 is engineered for AI development, professional content creation, and compute-intensive workloads.
- Massive 32GB Memory Capacity: 32GB of GDDR6 memory on a 256-bit bus provides ample bandwidth for large AI models, 8K video editing, and complex 3D rendering.
- Advanced RDNA 4 with AI Accelerators: 64 Compute Units with 3rd Gen Ray Tracing and dedicated 2nd Gen AI Accelerators for groundbreaking AI performance and visual computing.
- Professional Blower Cooling: Efficient single blower design exhausts heat directly out of the chassis, ideal for multi-GPU workstation and server configurations.
- Enterprise-Grade Thermal Solution: Vapor chamber heatsink with industrial Honeywell PTM7950 thermal interface material ensures reliable cooling under sustained professional loads.
AI capacity is shared with Meta’s core business
Meta describes the 2026 investment as supporting Meta Superintelligence Labs and the core business. That means an inference server, data-center hall or network link may serve an AI assistant one moment and a conventional recommendation, messaging, advertising or content-delivery workload at another time.
Meta’s consumer platforms also require enormous non-AI infrastructure. The same facilities and networks that support model training and inference can carry photos, videos, messages, feeds and advertising systems. Reality Labs hardware is included in the contractual-commitment disclosure as well, so that total is broader than AI infrastructure.
Consequently, no published figure in the cited disclosures answers “what percentage of capex is AI?” A defensible answer is that AI is a major driver of the capacity expansion, while the assets and contracts are shared across multiple businesses.
Why Meta says it needs so much capacity
Meta frames infrastructure capacity as a prerequisite for its AI opportunities. Training more capable models requires large, tightly connected clusters for extended runs. Deploying those models to billions of users requires inference capacity, redundancy and geographic distribution. Product experimentation adds another demand: several model versions may need to be trained, evaluated and served before one becomes a production feature.
Meta’s Q3 2025 outlook called infrastructure capacity “a central requirement to realizing these opportunities.” Chief executive Mark Zuckerberg said on July 30, 2025, “I’m excited to build personal superintelligence for everyone in the world.” Those statements explain the strategic rationale for the spending, but they do not convert the total budget into a measured cost per model or per user.
What the public figures cannot tell you
- There is no disclosed GPU-versus-CPU dollar split.
- Meta does not publish a complete AI-only versus core-business allocation for the $72.22 billion 2025 capex figure.
- The filings do not identify individual GPU quantities, supplier contracts, power draw or cooling costs.
- The $115–135 billion 2026 range is guidance, not a completed spend, and actual results can differ.
- The $103.77 billion lease balance and $131.05 billion commitment balance are future obligations measured at Dec. 31, 2025, not single-year AI invoices.
Any article or estimate that assigns a precise share of Meta’s budget to GPUs, buildings, electricity or cloud without additional company disclosure is making an assumption rather than reporting a published figure.
How to interpret the $100 billion-plus headline
- Start with the period. The $72.22 billion amount is 2025 capex; $115–135 billion is Meta’s 2026 plan.
- Check the definition. Capex includes finance-lease principal payments, while the $69.69 billion property-and-equipment figure covers purchases of long-lived physical assets.
- Separate owned from contracted. Servers and facilities can be purchased or leased, and cloud capacity is purchased as a service.
- Ask whether the asset is shared. Meta explicitly links the investment to both AI initiatives and its core business.
- Keep future obligations in the future. Lease and contractual-commitment balances indicate planned payments and capacity, not money already spent or capacity already online.
The most accurate description is therefore a capacity bet: Meta is committing capital, leases, cloud contracts and operating resources to obtain enough computing and facility capacity for expanding AI workloads while maintaining its existing services.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
Do these 3 things before closing this tab:
1Fix the driver behind crashes, sound loss and screen glitches2Clear out junk files and repair common Windows errors3Scan for outdated or missing drivers - takes under a minute
