Interview

SemiAnalysis's Jordan Nanos on hyperscaler IT asset life extension, GPU depreciation, and Microsoft's Azure strategy

Nov 12, 2025 with Jordan Nanos

Key Points

  • Hyperscalers including Microsoft, Meta, Oracle, and Google have extended IT asset depreciation schedules from 3-5 years to 5-6 years since 2020, suppressing depreciation charges and inflating reported earnings.
  • Nvidia's stated 2-3 year GPU product cycle directly contradicts the longer useful-life assumptions hyperscalers are booking, suggesting AI infrastructure costs are being systematically understated.
  • The synchronized shift across asset classes and companies represents a structural accounting risk that distorts operating income figures relied upon by investors.
SemiAnalysis's Jordan Nanos on hyperscaler IT asset life extension, GPU depreciation, and Microsoft's Azure strategy

Summary

Michael Burry is arguing that major hyperscalers — Meta, Microsoft Azure, Oracle, and Google — are artificially inflating earnings by extending the useful life of their IT assets. Between 2020 and now, reported depreciation schedules for servers, switches, and storage have stretched from a 3-to-5-year cycle to 5, 5.5, or 6 years across the board. The practical effect is that depreciation charges are suppressed, making reported earnings look stronger than underlying economics justify.

The GPU depreciation question sits at the center of this debate. Nvidia's public position that its product cycle runs 2 to 3 years implicitly challenges the extended schedules hyperscalers are booking, since a shorter useful life would demand faster, heavier depreciation on GPU-heavy infrastructure. If Nvidia is right about the cadence of hardware obsolescence, the hyperscalers are understating the true cost of their AI buildouts.

Jordan Nanos of SemiAnalysis frames this as a structural accounting risk, not a one-quarter anomaly. The life extension decisions are applied broadly across asset classes and companies simultaneously, which magnifies the earnings distortion across the sector. Investors relying on reported operating income from these platforms may be working with figures that systematically understate capital consumption.