Understanding Michael Burry's Bet Against AI: Here's What it Really Means for Investors

Michael Burry contends that hyperscalers are using overly long depreciation lives on AI-related servers, GPUs and network gear, estimating about $176 billion of understated depreciation industrywide from 2026–2028 and suggesting companies such as Oracle and Meta could be overstating earnings by roughly 27% and 21%, respectively—an adjustment he warns could force higher capex, hit reported earnings and weaken the AI spending narrative. Critics counter that lower depreciation increases taxable income and lowers cash flow (the metrics lenders and rating agencies focus on), that hyperscalers are investing for long‑term recurring revenue rather than near‑term EPS, and point to evidence such as Nvidia’s A100s still running after six years that supports multi‑year useful lives. The practical takeaway is that while some unproductive AI spend will likely be curtailed, the timing and magnitude are unclear, so investors should emphasize cash‑flow based valuations and be wary of financially weaker names that would be most exposed if depreciation assumptions reset.

Analysis

Michael Burry contends hyperscalers are understating depreciation on AI servers, GPUs and network gear, estimating roughly $176 billion of understated depreciation industrywide between 2026 and 2028 and arguing that Oracle and Meta could be overstating earnings by nearly 27% and 21%, respectively. He warns that a reset to shorter useful lives would reduce reported earnings and could force higher capex, undermining the returns case for some AI investments.

The article presents three counterarguments: lowering depreciation raises taxable income and reduces cash flow (the metrics lenders and rating agencies prioritize), hyperscalers are investing for long‑term recurring revenue rather than near‑term EPS, and company disclosures provide mixed evidence on useful lives—Amazon lowered server life to five years in 2025, Alphabet and Microsoft have moved certain assets to six years, and Nvidia reported A100 GPUs still running after six years. The Motley Fool highlights that hyperscalers remain highly cash‑generative, suggesting limited incentive to artificially boost short‑term EPS at the expense of cash metrics.

If Burry is correct, expect materially higher capex needs and pared‑back reported earnings at the most exposed firms and a potential reduction in marginal AI spending, but the article stresses timing and magnitude are uncertain. Practical investor guidance in the piece is to prioritize cash‑flow based valuations and be cautious on financially weaker names such as Oracle and Amazon; sentiment is mildly negative and the estimated market impact is modest, implying concentrated rather than systemic risk.

Key Decisions for Investors

• Prioritize cash‑flow and EBITDA metrics over GAAP EPS when valuing hyperscalers and AI‑exposed names and stress‑test models for shorter useful lives and higher capex
• Underweight or avoid the firms flagged as most exposed to a depreciation reset—notably Oracle and Amazon—until filings and capex guidance clarify useful‑life assumptions
• Monitor 10‑K/10‑Q disclosures for changes to server/GPU useful‑life assumptions, capex cadence and cash‑flow trends and consider hedges or trimming growth‑biased AI positions if multiple hyperscalers accelerate depreciation or guide higher near‑term capex