What Constitutes Depreciation and Its Income Statement Location
Meta Platforms records depreciation primarily on the physical hardware required to sustain its massive digital ecosystem. This includes hundreds of thousands of servers, specialized network equipment, and an ever-expanding global footprint of hyperscale data centers. These physical assets are fundamentally required to process the exabytes of data generated daily by its Family of Apps (Facebook, Instagram, WhatsApp) and to supply the immense computational power necessary for its deep-learning recommendation engines, AI-driven advertising targeting, and generative AI features. Additionally, Meta depreciates specialized equipment associated with its Reality Labs division, which focuses on augmented and virtual reality hardware.
On Meta’s condensed consolidated statements of income, depreciation is not itemized as a distinct expense. Instead, it is the absolute dominant driver of the “Cost of Revenue” line item, which houses the vast majority of the company’s technical infrastructure costs. A smaller, yet growing portion is also allocated to “Research and Development,” reflecting the hardware usage dedicated to training frontier AI models. During recent earnings calls, Meta’s CFO explicitly noted that depreciation is the single largest driver of its infrastructure cost growth.
Factors Affecting Depreciation
Meta is currently executing one of the most aggressive and highly concentrated infrastructure buildouts in corporate history, guided almost entirely by founder and CEO Mark Zuckerberg’s mandate to deliver “personal superintelligence” to billions of users. This unwavering commitment has led to capital expenditures soaring to nearly $70 billion in 2025, with stunning forward guidance estimating 2026 CapEx to land between $115 billion and $135 billion.
To blunt the devastating impact this volume of spending would naturally have on the income statement, Meta has repeatedly and aggressively manipulated its accounting estimates. After extending server useful lives in 2022, Meta completed yet another internal assessment in January 2025, deciding to further extend the estimated useful life of certain servers and network assets to 5.5 years. This highly specific maneuver reduced full-year 2025 depreciation expense by approximately $2.9 billion, artificially boosting operating margins.
Furthermore, Meta has engaged in highly complex, off-balance-sheet financing arrangements to manage its capital intensity. The company launched a massive $27 billion joint venture (dubbed the Hyperion data center project) with Blue Owl Capital to construct a 4-million-square-foot facility in Louisiana. Because Meta carefully structured the deal as a Variable Interest Entity (VIE) where it is not legally deemed the “primary beneficiary,” the physical asset - and its associated multi-billion dollar depreciation - is kept entirely off Meta’s consolidated balance sheet. This structure was highly controversial, prompting Meta’s auditor, Ernst & Young, to flag the decision as a formal “critical audit matter” in the annual report.
5-Year Historical Trends (2021 - 2025)
Meta’s historical cash flow data showcases a massive, accelerating wave of depreciation, perfectly tracking the company’s pivot from standard social media hosting to AI-driven algorithmic feeds and heavy compute infrastructure.
Fiscal Year (Ends Dec. 31) | Total D&A (Cash Flow Statement) | Capital Expenditures | Free Cash Flow |
2021 | $7.97 Billion | -$18.57 Billion | $39.12 Billion |
2022 | $8.69 Billion | -$31.43 Billion | $19.04 Billion |
2023 | $11.18 Billion | -$27.27 Billion | $43.85 Billion |
2024 | $15.50 Billion | -$37.26 Billion | $54.07 Billion |
2025 | $18.62 Billion | -$69.69 Billion | $46.11 Billion |
The jump from $15.50 billion in 2024 to $18.62 billion in 2025 represents a 20% year-over-year increase. However, it is crucial to note that this 2025 figure would have been significantly higher (likely exceeding $21.5 billion) if management had not intervened in January 2025 to stretch the useful life of the hardware to 5.5 years, shielding $2.9 billion from the income statement. Despite the nearly $70 billion in CapEx, Meta’s highly lucrative core advertising business allowed it to maintain a robust $46.11 billion in free cash flow.
Predictions and Forecasts (2026 - 2028)
Despite the accounting adjustments and the off-balance-sheet machinations with Blue Owl Capital, the mathematical physics of a $115 to $135 billion 2026 capital expenditure budget cannot be entirely hidden from future statements. The sheer influx of new Nvidia GPUs required to train the LLaMA foundation models guarantees an explosion in future amortization. Independent modeling forecasts that Meta’s D&A will surge to approximately $26 billion in 2026, scaling rapidly toward the $39 billion threshold by 2028. Because Meta lacks a standalone enterprise cloud computing division to directly monetize this compute infrastructure, this incoming wave of depreciation must be entirely absorbed by digital advertising revenues.
The Superintelligence Mandate and Accounting Gymnastics
Meta Platforms presents the most fascinating - and arguably the most structurally precarious - case study in Big Tech financial engineering during the generative AI boom. Mark Zuckerberg’s unyielding, singular pursuit of embedding “personal superintelligence” across the Family of Apps has triggered a capital expenditure cycle that is breathtaking in its scale. With 2026 CapEx guided between $115 billion and $135 billion, Meta is preparing to deploy more capital in a single twelve-month window than it spent cumulatively from its inception through 2021. However, what demands intense, skeptical analyst scrutiny is not merely the astronomical volume of the spending, but the intricate accounting gymnastics utilized by management to insulate the income statement from the sheer gravity of these investments.
First, the continued manipulation of depreciation schedules must be addressed. By revising the estimated useful life of its servers and network equipment to 5.5 years in January 2025, Meta effectively erased $2.9 billion in depreciation expense from its 2025 Cost of Revenue. While perfectly legal and compliant with GAAP, this assumption is highly aggressive given the underlying technology. Meta is procuring hundreds of thousands of state-of-the-art GPUs to train frontier models. The fundamental nature of AI model training is that computational demands scale exponentially; hardware that is cutting-edge today will inevitably lack the interconnect bandwidth and memory capacity required to train future iterations long before 66 months have elapsed. By forcing a 5.5-year amortization schedule on assets with a realistic competitive lifespan of 24 to 36 months, Meta is artificially inflating its current operating margin (which printed a highly impressive 41% in 2025) while consciously pushing an inevitable wave of massive asset impairments into the future.
Second, the structural opacity of Meta’s operations has been severely compounded by its increasing use of Variable Interest Entities (VIEs) to mask capital intensity. The $27 billion Hyperion data center project - a sprawling, 4-million-square-foot, 2-gigawatt facility in rural Louisiana - has been kept completely off Meta’s balance sheet through a joint venture with Blue Owl Capital. Meta owns 20% of the venture, and because it successfully argued to auditors that it is not the “primary beneficiary,” the massive debt, the physical capital assets, and the ensuing depreciation are swept off its consolidated financial statements. The fact that Ernst & Young flagged this specific accounting treatment as a “critical audit matter” highlights the extreme judgment and borderline obfuscation involved. For analysts attempting to model Meta’s true capital intensity and authentic return on invested capital (ROIC), these off-balance-sheet vehicles act as a deliberate smokescreen, making the company appear far more capital-efficient than reality dictates.
Finally, the core monetization question remains uniquely acute for Meta. Alphabet, Amazon, and Microsoft possess the luxury of renting their rapidly depreciating assets to enterprise customers via their respective cloud divisions (Google Cloud, AWS, Azure). Meta cannot. Every single dollar of the impending $39 billion in annual depreciation hitting Meta’s income statement must be offset by an incremental dollar of digital advertising revenue or consumer hardware sales. Meta is betting the company on the thesis that AI will dramatically improve algorithmic ad targeting, user engagement, and automated content generation, thus allowing ad prices and impression volumes to rise faster than the underlying depreciation curve. In 2025, this thesis held together beautifully, with total revenue climbing 24% and ad impressions rising 18%. However, if global corporate advertising budgets contract, or if the “superintelligence” integrations fail to yield proportionate, sustained increases in user monetization, Meta’s income statement will be crushed under the weight of its own proprietary infrastructure.