Which Artificial Intelligence (AI) Stocks Are Billionaires Buying the Most?

13F filings for Q3 2025 show ten prominent billionaires and their firms largely directed capital into AI-related names, with all but Bill Ackman and Carl Icahn buying at least one AI stock. Alphabet and Nvidia were the most widely purchased — roughly half of the investors added one or the other — including Berkshire initiating a material new position in Alphabet and Paul Tudor Jones increasing his Nvidia stake more than sevenfold; Millennium boosted Nvidia by 126% and Citadel added 1.73 million NVDA shares (a ~21% increase to its holding). Honorable mentions include Broadcom, Meta and Microsoft (Soros more than tripled Microsoft), underscoring concentrated positioning in a handful of mega-cap AI beneficiaries. The filings highlight influential investor flows into key AI/tech incumbents that could inform relative positioning and risk exposures across quant and discretionary portfolios.

Analysis

Market structure: Winners are NVDA and GOOGL (GPU/AI infra and cloud/TPU adoption) along with MSFT and AVGO as picks-and-shovels suppliers; losers are legacy CPU incumbents and smaller AI chip hopefuls whose R&D/capacity can't match TSMC-backed roadmaps. Expect sustained pricing power for cutting‑edge accelerators through 2025–26 driven by constrained wafer capacity (TSMC/SMIC lead times 6–12 months) and enterprise cloud procurement cycles, supporting higher equipment/semicap order books.

Risk assessment: Tail risks include renewed US‑China export controls or a Taiwan disruption that could cut NVDA revenue >20% in 12 months, aggressive anti‑trust action vs large cloud/ad platforms hurting GOOGL/MSFT, or a sentiment unwind leading to >30% drawdowns in richly priced AI names. Near term (days–weeks) volatility is driven by 13F-driven flows and earnings; medium term (quarters) by product launches/capex; long term by TPU vs GPU architectural winners and fabs' capacity expansion.

Trade implications: Direct: establish a starter long NVDA (1–2% portfolio) and GOOGL (2–3%) sized to risk budget, scale into 10–15% pullbacks; use 3–6 month call spreads to express NVDA upside while capping premium. Pair: long NVDA vs short BRK.B (growth vs cyclical/value hedge) or long GOOGL vs short META for relative ad/cloud exposure; overweight semiconductors and cloud infra, underweight late‑cycle consumer names.

Contrarian angles: Consensus underestimates TPU economics and potential margin pressure on GPU ASPs over 2–4 years; NVDA's current multiples price near‑perfect execution—a 20% miss in datacenter growth could force 25–40% re‑rating. Historical parallel: 1999 hardware winners then shakeouts — diversify across cloud/infra exposures and size positions with strict stop-losses (10–15%).

Key Decisions for Investors

• Initiate a 1–2% portfolio long in NVDA using a 3–6 month 1:1 call spread (buy at‑the‑money call, sell 20%‑out call) to cap premium; add to full 3–4% position only on a 10–15% pullback from current levels or after a material earnings beat (+5% rev guide).
• Buy 2–3% position in GOOGL (prefer GOOG) for multi‑year cloud/TPU exposure; hedge 30–50% of position with 6–9 month put spreads that trigger below a 20% decline to protect against ad/cyclical shock.
• Establish a pair trade: long NVDA (0.75–1.5%) vs short BRK.B (0.75–1.5%) to express AI growth vs value/insurance; size stops at 12% adverse move and rebalance if NVDA outperforms by 25% absolute.
• Avoid concentrated long exposure to single AI names >5% of portfolio; instead overweight semicap suppliers (AVGO, broad semiconductors) by up to +2% on conviction but trim if NVDA/GOOGL move >30% from entry — take profits incrementally at +20% and +40%.
• Monitor catalysts over next 30–90 days: (1) NVDA and GOOGL earnings/capex guidance, (2) any US export‑control headlines (threshold: any new ban affecting datacenter GPUs/TPUs), and (3) TSMC capacity guidance; if any single catalyst materializes (e.g., new export control) reduce NVDA/GOOGL exposure by 50% within 48 hours.