AI Investment Risk Analysis: Depreciation Accounting vs Growth Capex Returns

This analysis is based on a Reddit post [10] published on November 14, 2025, which challenges Michael Burry’s recent criticism of AI companies’ depreciation practices, arguing that the focus on accounting timing differences misses the more fundamental risk: uncertain returns on massive growth capital expenditures in AI infrastructure [10].

The debate centers on two distinct but related risk perspectives:

: Michael Burry, famed investor from “The Big Short,” has accused AI hyperscalers of artificially boosting earnings through aggressive depreciation schedule extensions [1][2][3]. His allegations include:

• $176 billion earnings inflation
  : Estimated understatement of depreciation from 2026-2028 across major AI companies [1][2]
• Specific company impacts
  : Oracle potentially overstating earnings by 26.9% and Meta by 20.8% through these accounting practices by 2028 [2][3]
• Accounting methodology
  : Companies extending useful life estimates for AI hardware from 2-3 years to 5+ years, reducing annual depreciation expenses [1][4]

: The counter-argument posits that depreciation is merely a timing difference in accounting, while the fundamental risk is whether massive AI infrastructure investments can generate adequate returns before hardware becomes obsolete [10]. This perspective emphasizes that the uncertainty isn’t captured by GAAP earnings and is already priced into lofty AI valuations [10].

The analysis reveals that both perspectives highlight legitimate but different aspects of AI investment risk:

: Research indicates AI hardware faces significant obsolescence risk, with technology purchased today potentially becoming outdated within 18-24 months [8]. This creates a mismatch between accounting depreciation schedules (5+ years) and actual economic useful life.

: The scale of AI infrastructure investment is unprecedented. Morgan Stanley estimates a $1.5 trillion funding shortfall for the $2.9 trillion global data center capex needed from 2025-2028 [9]. With $250 billion already spent and another $1 trillion planned, these assets must generate $187.5 billion in annual net income (15% returns) to justify the investment [8].

: Reddit (RDDT) presents an interesting case study with its unique market position [0]:

• Stock Performance
  : $195.88 (+3.41% on November 14, 2025), +48.53% over 52 weeks
• Financial Health
  : Q3 2025 revenue of $584.91 million (+68% YoY), net income of $163 million (28% net margin)
• Low Capex Model
  : Only $2 million in quarterly capex (less than 0.5% of revenue)
• AI Exposure
  : Data licensing partnerships with Google and OpenAI generating $36 million in Q3 2025 (+7% YoY)

The Technology sector is currently outperforming with a +2.20% gain on November 14, 2025, while Communication Services (Reddit’s sector) is down -1.86% [0], reflecting some concerns about AI-related valuations. Reddit’s high P/E ratio of 89.85x suggests extremely elevated growth expectations that may be difficult to sustain [0].

:

• Massive Capex Commitments
  : Companies like Microsoft, Meta, and Amazon are spending tens of billions annually on AI infrastructure [9]
• Technology Obsolescence Risk
  : Rapid advancement in AI hardware could render current investments obsolete faster than depreciation schedules assume [8]
• Monetization Uncertainty
  : Clear paths to generating returns on AI infrastructure investments remain unclear [8]

:

• Valuation Risk
  : P/E ratio of 89.85x suggests extremely high growth expectations
• Data Licensing Concentration
  : Heavy reliance on a few major AI companies for data licensing revenue
• Competition Risk
  : Other platforms may develop competitive data offerings

:
Companies like Reddit that provide AI training data without bearing infrastructure costs may offer more attractive risk-adjusted exposure to AI growth trends [6][7].

:
The market may be overvaluing infrastructure-heavy AI companies while undervaluing data providers, creating potential arbitrage opportunities.

The analysis reveals critical timing considerations:

• Hardware Replacement Cycles
  : Actual vs. projected hardware replacement cycles will become apparent within 18-24 months [8]
• Monetization Timelines
  : AI infrastructure investments must begin generating returns within 2-3 years to justify current spending levels [8]
• Accounting Scrutiny
  : Regulatory attention to AI accounting practices is likely to increase as more investors question depreciation assumptions [1][2][3]

:

• Reddit’s strong financial performance: 28% net margins, $2.2 billion cash position [0]
• Minimal capex exposure: $2 million quarterly capex vs. billions spent by hyperscalers [0][9]
• AI data licensing growth: $36 million in Q3 2025 (+7% YoY) [6][7]
• Sector performance divergence: Technology +2.20% vs. Communication Services -1.86% [0]

:
The analysis reveals that while Burry’s depreciation critique highlights important accounting transparency issues, the Reddit author’s focus on economic returns may identify the more fundamental risk. The massive scale of AI infrastructure investment ($2.9 trillion needed through 2028) [9] combined with rapid technology obsolescence (18-24 month cycles) [8] creates significant uncertainty about whether these investments can generate adequate returns.

:
Reddit appears strategically positioned as a “picks and shovels” play in AI, providing essential data while avoiding massive infrastructure risks. However, the company’s high valuation suggests these advantages may already be fully priced in by the market [0].

the AI investment cycle faces significant uncertainty regarding returns on infrastructure spending, and technology obsolescence risk could create substantial value destruction if hardware replacement cycles accelerate beyond accounting assumptions. This risk is particularly acute for companies making massive capex commitments, while data providers like Reddit may offer more attractive risk-adjusted exposure to AI growth trends.