Peter Thiel dumped Nvidia and invested $45 million into Microsoft and Apple—sending a strong signal about who will win the AI race

Thiel Macro sold all 537,742 Nvidia shares (well over $100M) in Q3 2025 and cut U.S. equity exposure from $212M to $74M, a reduction of more than 50%. The fund redeployed into roughly $45M of Apple and Microsoft positions, signaling a preference for platform/consumer-integrated AI plays over pure-play AI chip exposure. The trade highlights concerns about an AI-driven valuation bubble around Nvidia (noted $5T market cap) and favors diversified businesses with large distribution moats.

Analysis

Platform owners that control distribution, payment rails and end-user interfaces have a structurally easier path to monetize AI than pure-play compute vendors because monetization shifts from one-time capex to recurring ARPU expansion. Over a 1–3 year horizon that favors subscription, ads and services (Office/365, App Store), expect incremental revenue per active user to rise while datacenter GPU demand becomes more volatile: training cycles are lumpy and OEM/cloud procurement can swing +/-30% quarter-to-quarter as models finish major training epochs.

Second-order supply effects: an inventory or demand hiccup in datacenter GPUs would not only pressure the chipmaker’s gross margins but also flood a secondary market for used accelerators, compressing new-chip ASPs and accelerating design wins for lower-cost inference accelerators (Apple silicon, Google TPUs, bespoke ASICs). Catalysts to watch across timeframes include quarterly datacenter spend guides (days–weeks), major model releases or open-source milestones (weeks–months) that change cloud usage patterns, and multi-year shifts from training to edge inference where smartphone SoCs and on-device models capture more spend.

Consensus is underweighting two points: first, the stickiness of CUDA and ecosystem lock-in gives the incumbent more runway than simple capex cycle analysis implies; second, platform leverage (bundled services + hardware) can reprice TAM capture without proportional datacenter spend. That makes a blended approach — owning platform optionality while hedging compute cyclicality — the most robust portfolio response over the next 6–18 months.

Key Decisions for Investors

• Long MSFT 9–12 month call spread (buy 12-month ATM call / sell 30% OTM call) sized 3–5% notional. Rationale: monetization optionality from enterprise AI; aim for asymmetric 2–3x upside if Office/Cloud ARPU prints accelerate. Stop if Azure commercial commitment growth decelerates two consecutive quarters. Max loss = premium paid.
• Buy AAPL on pullbacks or initiate 12–18 month LEAPS (buy Jan-2027 10–20% OTM calls) sized 3% notional. Rationale: edge-inference capture and services ARPU make Apple a durable AI monetizer; target 25–40% upside if product integration increases attach rates within 12 months. Trim 50% on a 20% gain or if Services growth falls below company guide.
• Relative pair: Long 60% MSFT + 40% AAPL vs Short NVDA 1:1 dollar-weighted, horizon 6–12 months. Use this to express platform-over-infrastructure view while neutralizing broad market beta. Risk: if Nvidia extends leadership and prices hold, expect ~10–15% mark-to-market pain; scale out if pair outperforms by 15%.
• Defined-risk NVDA short via calendar/verticals: buy 3–6 month 20% OTM put spread financed by selling 3–6 month 40% OTM call (net-debit small). This captures downside from a compute demand reset or inventory destock while capping capital at risk. Exit on a sell-side inventory guide cut or if implied vol normalizes below historical ranges.