Bridgewater's Greg Jensen echoes Michael Burry on Nvidia's AI chips — and says they could help make themselves obsolete

Hedge fund manager Michael Burry has warned AI giants are overstating useful lives of Nvidia GPUs—arguing hyperscalers have stretched depreciation from roughly three years to six—to bolster near-term profits, and he disclosed bearish put exposures on Nvidia and Palantir. Bridgewater co-CIO Greg Jensen added that AI-driven innovation could rapidly obsolete current chips (accelerating depreciation) even as Nvidia attempts to lock customers into an ecosystem to fend off vertical integration by hyperscalers. Together the comments highlight heightened downside risk for GPU-driven revenue streams and signal potential negative reassessments of valuations across AI hardware suppliers and dependent cloud customers.

Analysis

Market structure: Shortening useful lives of GPUs benefits vertically integrated hyperscalers (GOOGL, MSFT, AMZN) and software/IP vendors while pressuring pure-play hardware incumbents (NVDA, PLTR exposure in AI tooling). Expect pricing power to bifurcate — sustained gross margins for software/services, sharper margin compression for commodityized silicon if cadence of new architectures accelerates to <36 months. In the near term (0–12 months) tight supply keeps NVDA pricing strong; over 12–36 months accelerated innovation implies unit demand growth but faster depreciation per asset.

Risk assessment: Tail risks include antitrust action (DOJ/EU) or a breakthrough in chip design/AI that reduces GPU reliance within 12–36 months; an abrupt hyperscaler pivot to in‑house ASICs could wipe 30–60% of addressable GPU revenue for NVDA over two years. Immediate catalysts (days–weeks) are earnings, product launches and capex guides; medium term (3–12 months) watch fab capacity, energy constraints and lead times. Hidden dependencies: data‑center power and real‑estate scarcity, and software lock‑in economics that can blunt hardware obsolescence.

Trade implications: Tactical trades: short-tail exposure to NVDA via 3–6 month puts (5–15% OTM) sized 1–2% portfolio; balance by 2–3% long in GOOGL/GOOG for 6–18 months to capture software/stack capture. Implement pair trade (long GOOG 2% / short NVDA 1%) and finance puts with call spreads; rotate 10–25% weight out of pure-play semiconductor hardware into hyperscalers, AI-software names and power/utility names servicing data centers.

Contrarian angles: Consensus underrates NVIDIA’s ability to monetize an ecosystem (software, SDKs, licensing) which caps downside — a >25% NVDA drawdown without fundamental shock is plausible but not certain. The market may be over-discounting obsolescence if transition costs and software lock‑in keep incumbents sticky; mirror historical hardware cycles (Intel vs custom ASICs) where incumbents earned rents for a decade despite competition. Monitor hyperscaler capex cadence and a 12–24 month shift to in‑house silicon as the real inflection metric.

Key Decisions for Investors

• Establish a 1–2% portfolio short via NVDA 3–6 month puts 5–15% OTM (size to cost ≤1% portfolio); increase to 2–3% if NVDA rallies >15% in 30 days or if hyperscaler capex guides disappoint.
• Deploy a 2–3% long position in GOOGL/GOOG over the next 4–8 weeks, target hold 6–18 months to capture AI software/stack monetization; hedge with 9–12 month 10% OTM puts sized to 0.5% portfolio if downside >20% would breach risk limits.
• Run a relative-value pair: long GOOG 2% / short NVDA 1% (dollar‑neutral), rebalance monthly and widen short if NVDA implied vol >100% or if NVDA releases a new architecture within 3 months.
• Reduce pure-play semiconductor hardware exposure by 20–40% vs benchmark over the next quarter; redeploy into data‑center utilities/power infrastructure and AI SaaS vendors (allocate 1–2% to copper/energy-linked names) to hedge power/real‑estate risk.