Eight legendary investors’ AI bets — one score for consensus vs. divergence
The Compass Consensus Score turns the disclosed moves of Buffett, Duan Yongping, Cathie Wood and five more legends into a 0–100 score per AI stock — fully published methodology, refreshed every 13F season. With theme maps, stock profiles, and a long-term framework. Sourced and traceable, never a black box.
AI market snapshot
From picks-and-shovels to the energy base layer, the AI market splits into four clear layers. Know which one you are buying.
Compute & AI Chips
4 namesThe “pick-and-shovel” layer of the AI boom. Training and inference demand drives GPU/accelerator sales — the clearest cash flows today, but also the most crowded and richly valued.
Physical AI & Robotics
2 namesAI steps out of the screen into the physical world: embodied intelligence and humanoid robots. Jensen Huang calls humanoids a ~$40T market and robotics Nvidia’s second growth curve after AI; SoftBank’s Son calls physical AI the birthplace of the next trillion-dollar company. Listed pure-plays are scarce and the leaders are mostly private — narrative and valuation run ahead of deployment.
Cloud & AI Infrastructure
3 namesThe layer that turns compute into rentable services: hyperscalers and emerging GPU clouds. Capex is enormous, but it locks in long-term AI workload demand.
AI Applications & Platforms
2 namesThe layer that turns models into products and revenue: search, ads, productivity, vertical SaaS. Winners are decided by distribution and data, not raw compute.
AI Memory & Storage
5 namesThe new institutional consensus of 2026: AI data centers have pushed HBM memory and NAND/HDD storage into a shortage cycle. Hedge-fund semiconductor weight hit a record ~10%, and the newest Goldman VIP names — SanDisk, Lam Research, Applied Materials — all sit on this chain.
AI Energy & Power
2 namesThe overlooked bottleneck: surging data-center power demand puts nuclear, grid, and cooling on the AI map — a “second-order” beneficiary.
China AI
3 namesThe AI story at China’s internet giants: in-house models + cloud + e-commerce/ad monetization. Usually cheaper than U.S. peers, but carrying policy and geopolitical risk.
Latest updates
Updates →Memory stocks enter a bear market: Micron, Samsung and SK Hynix all down 20%+ from highs
In the week of July 13–17 the memory chain went from hottest trade to bear market: SanDisk fell 12.6% on Monday and another 8% Thursday to close at $1,354.82 on 7/17 (roughly -35% from its $2,000+ June peak); Micron closed at $848.95; Micron, Samsung, SK Hynix and the Roundhill Memory ETF are all 20%+ below recent highs. SKHY — SK Hynix’s ADR, public for just a week — faded from a +19% spike on 7/14 (the day its leveraged ETFs launched) to $154.03, only 3.4% above the $149 IPO price. Profit-taking is the main driver: Micron was up as much as 244% YTD and SanDisk 640%, while TrendForce sees 3Q26 contract-price gains slowing to +13–18% for DRAM and +10–15% for NAND (vs +90–95% and +58–63% the prior two quarters).
Cathie Wood buys the dip: adding CoreWeave in the sell-off, with 30% of ARK in 5 AI stocks
As memory and AI-compute names corrected, ARK leaned in: Wood kept buying CoreWeave through the mid-July sell-off. Roughly 30.4% of Ark Invest’s portfolio now sits in five AI stocks — Tesla, SpaceX, Alphabet, AMD and Amazon. Among the eight tracked investors Wood remains the most aggressive buy-the-dip bull, a sharp contrast with the drawdown in Druckenmiller’s new memory positions the same week (he rotated into SNDK/STX/MU in Q1 — see the 7/14 entry).
Berkshire’s great pivot: Amazon exited, Alphabet boosted 225%
The Q1 2026 13F shows Berkshire cutting its book from 40 to 26 names with 15 full exits: Amazon sold entirely, Alphabet boosted 225% as its clearest AI statement, and Apple untouched (still 22%) after three quarters of trimming. Under Abel, Berkshire is concentrating AI exposure into search cash flow + full-stack AI.
Featured legendary investors
Rather than chase the theme, see how genuine long-term investors allocate capital across AI.
Warren Buffett
CautiousQ1 2026 brought the “great pivot”: Berkshire exited Amazon entirely and boosted Alphabet by 225%, while leaving Apple untouched — shifting its AI exposure from e-commerce/cloud toward search + full-stack AI.
Cathie Wood
BullishOne of the most aggressive AI bulls. Bets on AI infrastructure and next-gen compute — CoreWeave, Cerebras — and on nuclear (X-Energy) as AI’s energy base layer.
Stanley Druckenmiller
CautiousA stunning Q1 reversal: exited Alphabet entirely and rotated into SanDisk, Seagate, Micron, Broadcom and Intel — betting on AI’s memory/storage bottleneck, with tech exposure doubling from 9.4% to 18.4%.
Bill Ackman
BullishHolds only a handful of high-conviction names. Gains AI exposure through quality compounders like Alphabet rather than speculation.
Duan Yongping
CautiousA value investor anchored in Apple and Berkshire who leaned hard into the AI chain in 2026 — adding 91% to Nvidia (now #3), opening Tesla and adding PDD in Q1 while exiting Alibaba. Especially relevant for Chinese readers.
David Tepper
BullishIn Q1 2026 he nearly doubled Amazon into his #1 position (AI angle = accelerating AWS), while trimming Nvidia and AMD and leaning into Micron — “betting on AI monetization and the memory cycle, not pure compute.”
Philippe Laffont
CautiousQ1 2026: $29.1B across 62 names, with the top five now TSMC, GE Vernova, Lam Research, Applied Materials and Broadcom — full-stack AI infrastructure (foundry + equipment + power) — while adding 18% to Microsoft.
Michael Burry
BearishThe “Big Short” investor and the AI trade’s starkest bear: after winding down Scion in November 2025, he disclosed in May 2026 that ~80% of his personal book sits in Nvidia and Palantir put options (~$1.1B notional).
Long-Term
AI will swing, but the principles of long-term investing do not. Put discipline into a checklist.
Circle of competence: invest in what you understand
Only bet where you can explain the business model and competitive landscape. Depth of understanding is what lets you hold through volatility.
AI · With AI, first separate compute vs. infrastructure vs. applications vs. energy — each layer has a very different moat and risk profile.
Moats: look for durable competitive advantage
Durable high returns come from advantages that are hard to copy: network effects, switching costs, scale, proprietary data, or ecosystem lock-in.
AI · Beware moat-less “wrapper” apps; favor companies that own distribution, proprietary data, or a full-stack position.
Valuation discipline: a great company still needs a fair price
The price you pay for growth determines your return. Even a great story can pre-spend years of future gains if bought too dear.
AI · Rich valuations make AI leaders acutely sensitive to any slowdown; keep a margin of safety and avoid going all-in at peak euphoria.
Get quarterly 13F updates
Each quarter: how the legends moved + Consensus Score changes, in one email. Subscribers get first access when Pro tools launch. Free, unsubscribe anytime.