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

Michael Burry is betting against AI-focused stocks on the thesis that hyperscalers have materially understated depreciation on AI hardware — he estimates about $176 billion of understatement industry-wide from 2026–28 and argues that companies like Oracle and Meta could be overstating earnings by roughly 27% and 21%, respectively. SEC filings show many cloud providers have extended useful‑life assumptions for servers and network gear (Alphabet and Microsoft moving from ~4 to 6 years; Oracle and Meta to ~5–6 years), though Amazon recently shortened server lives to five years and Nvidia says A100 GPUs remain fully utilized after six years; critics counter that large hyperscalers are cash‑generative, focus on long‑term recurring revenue, and that EBITDA/cash flow metrics matter more than GAAP net income. If Burry is correct, hyperscalers would face higher ongoing capex and lower reported earnings, pressuring returns on AI investment and potentially trimming sector spending growth—investors should prefer cash‑flow‑based valuations and be cautious on less robust names such as Oracle and Amazon.

Analysis

Michael Burry argues hyperscalers have materially understated depreciation on AI hardware, estimating roughly $176 billion of understatement industry-wide between 2026 and 2028 and claiming Oracle and Meta could be overstating earnings by about 27% and 21%, respectively. Depreciation is significant because longer useful-life assumptions (lower depreciation expense) boost near-term GAAP earnings while deferring expense and potentially masking higher future capex requirements.

SEC filings show a general trend toward longer useful-life assumptions: Alphabet moved servers from four to six years (2023), Microsoft moved server and network equipment from four to six years (2022), Oracle moved servers from five to six years (2025), Meta shows 5.5 years for certain assets (2025), while Amazon shortened server life from six to five years (2025). Nvidia management’s comment that A100 GPUs shipped six years ago remain at full utilization supports the longer-life assumption for at least some hardware.

If Burry is correct, hyperscalers would face higher recurring capex, lower reported earnings, and potentially weaker measured returns on AI investment as companies rebase depreciation; this could slow sector investment growth. Counterarguments in the article note hyperscalers’ strong cash generation, the focus of credit markets on cash flow and EBITDA, and Google Cloud’s improvement from a $700 million loss in Q3 2022 to a $3.6 billion profit with 85% year-over-year growth this year.

For investors the key takeaway in the article is to prioritize cash-flow-based valuation metrics over GAAP net income and to be cautious on names the piece highlights as more exposed, notably Oracle and Amazon, while monitoring utilization and capex guidance closely.

Key Decisions for Investors

• Prioritize cash-flow and EBITDA-based valuations rather than GAAP net income when assessing hyperscalers given the risk of depreciation-policy revisions
• Monitor 10-K/earnings-call disclosures for changes in useful-life assumptions, capex guidance, and utilization metrics (especially Nvidia A100 utilization) as triggers for position review
• Consider underweight or hedge exposure to companies the article flags as most exposed to depreciation repricing (Oracle, Amazon, and to a lesser extent Meta) until cash-flow resilience and capex cadence are validated
• If depreciation assumptions are shortened across peers or capex materially accelerates, re-evaluate sector allocations for potential earnings downgrades and weaker implied returns on AI investments