NVDA Chip Obsolescence: Implications for AI Industry Economics and Bubble Risks

On November 22, 2025, a Reddit discussion (ticker: NVDA) focused on chip obsolescence and its implications for the AI industry. Key arguments included:

1. Recurring Capex Burden
   : AI firms face annual chip replacement costs (vs. traditional 3–5 year cycles), likened to shovels lasting only a week for gold miners.
2. Depreciation Practices
   : Cloud companies extend server useful lives to inflate earnings, while Nvidia shortens chip cycles to 1 year.
3. Bubble Risks
   : Unsustainable free AI adoption and low ROI could trigger a market correction.
4. Counterpoint
   : Older chips can be repurposed for non-frontier tasks (e.g., inference), mitigating obsolescence concerns [1][2].

Leading cloud/AI firms (Amazon, Meta, Microsoft, Alphabet) are projected to spend $349 billion on AI data center capex in 2025 [4]. Meta lifted its 2025 capex guidance to $70–72 billion, with further increases planned for 2026 [3]. Nvidia estimates 40% annual capex growth for data centers until 2027 [2].

Michael Burry (Big Short investor) criticized cloud companies for extending server useful lives while Nvidia shortens chip cycles, calling this a “$4 trillion accounting puzzle” masking true costs [2].

Nvidia’s shift to annual chip updates increases recurring capex for frontier AI tasks, but older chips are repurposed for inference (creating a secondary market) [1]. This reduces waste but does not eliminate the need for frequent frontier chip investments.

Nvidia’s data center revenue accounts for 88.3% of FY2025 total revenue, positioning it as the primary beneficiary of frequent chip replacements [0]. Its $4.44 trillion market cap and 73.4% analyst Buy consensus reflect strong investor confidence [0].

The resale market for older chips lowers entry barriers for smaller AI firms, but Nvidia retains exclusive control over frontier chip supply [1].

While 73.4% of analysts rate NVDA as Buy, 3.8% have Sell ratings—one citing faster-than-expected chip obsolescence as a long-term risk [0][1].

1. Cycle Compression
   : Nvidia’s annual chip updates mark a shift from traditional 3–5 year hardware cycles, altering AI infrastructure economics [1].
2. Workload Optimization
   : AI firms are optimizing chip usage across frontier training vs. inference tasks to mitigate costs [1].
3. Regulatory Scrutiny
   : Burry’s comments may attract regulatory attention to cloud companies’ depreciation practices [2].

• AI Data Centers
  : Facing recurring capex burdens; need to balance frontier chip investments with repurposing older chips for inference [3][4].
• Nvidia
  : Short-term gains from frequent sales, but long-term risk if AI firms cut spending due to unsustainable costs [0][2].
• Investors
  : Monitor big tech capex trends and AI ROI metrics to assess bubble risks. Nvidia’s 44.77x P/E ratio reflects high growth expectations [0].

1. Chip Lifecycle
   : Speed of obsolescence for frontier AI tasks (current 1-year cycle vs. historical 3–5 years).
2. Depreciation Policies
   : Regulatory scrutiny on extended server useful life claims [2].
3. AI ROI
   : Ability of AI services to generate revenue covering recurring capex [3].
4. Secondary Market Efficiency
   : Repurposing older chips for non-frontier tasks to reduce costs [1].