Mid-2026 institutional flows report: where is smart money moving in AI?
Five directions to read mid-2026 institutional money: Berkshire’s great pivot, hedge funds rotating into the memory chain, power as the new compute, sovereign funds’ $120B entry, and the starkest bull-bear split on record. All from public 13Fs and reporting, fully sourced.
Direction 1: the great pivot — from e-commerce cloud to full-stack AI
The heaviest 13F signal of Q1 2026 came from Berkshire: the book was cut from 40 to 26 names with 15 full exits — Amazon sold entirely — while Alphabet was boosted 225% and Apple left untouched (~22%). Abel-era Berkshire made its clearest AI statement yet: own the full stack of search cash flow + Gemini + in-house TPUs rather than pay up for narrative.
Direction 2: the memory chain is the new picks-and-shovels
Druckenmiller dumped all his Alphabet in Q1 and rotated into SanDisk, Seagate, Micron, Broadcom and Intel, doubling tech exposure from 9.4% to 18.4%. He isn’t alone: Goldman’s Hedge Fund VIP rebalance in late May showed about half of the biggest popularity gainers were AI-linked — led by SanDisk, Lam Research and Applied Materials — with semis at a record ~10% of hedge-fund books. Institutions are widening “picks and shovels” from GPUs to HBM, NAND and semicap equipment, where supply is tighter and crowding was (until now) lower.
Direction 3: power is the new compute
Coatue’s top five says it all: TSMC, GE Vernova, Lam Research, Applied Materials, Broadcom — a gas-turbine/grid company sitting beside four semiconductor names. For institutions, AI and energy have merged into one infrastructure story: the data-center power gap, like the HBM gap, is a hard constraint on compute growth — and both are best expressed through upstream suppliers.
Direction 4: sovereign funds raise the capex floor
Across 2025–26, sovereign wealth funds have committed roughly $120B to data centers, fabs and HPC networks; Abu Dhabi’s MGX closed a $49B AI fund on July 1, with ~30% of allocations aimed at national champions. For public-market investors this puts nation-state money under the AI capex cycle — extending demand for power equipment, semicap and storage — though state-driven overcapacity is the risk to watch.
Direction 5: the starkest bull-bear split on record
On one side, Duan Yongping added 91% to Nvidia (now his #3) and opened Tesla, while Tepper pushed Amazon to his #1. On the other, Michael Burry — after winding down his fund — put ~80% of his personal book into Nvidia and Palantir puts (~$1.1B notional). Institutions are also genuinely split on hyperscaler capex ROI. When style-diverse masters disagree this violently, diversification, position sizing and margin of safety matter more than ever — which is exactly what the Consensus Score quantifies: not who’s right, but where consensus and disagreement actually sit.
How to use this as an individual investor
Three steps: (1) watch how these moves reshape each stock’s Consensus Score on the leaderboard — Berkshire’s exit pulls Amazon’s score down, and that is the signal; (2) run your own holdings through the Portfolio Check to see your overlap with the new institutional direction; (3) remember 13Fs lag ~45 days and omit shorts and option detail — a research starting point, not a copy-trade list. Educational content, not investment advice; investing carries risk.
Sources: Seeking Alpha — Tracking Berkshire portfolio Q1 2026 · HeyGoTrade — Druckenmiller dumps GOOGL, Tepper doubles AMZN · Goldman Sachs — Hedge Fund Trend Monitor: All in on AI · 13F.info — Coatue Management Q1 2026 · Titan Investors — sovereign funds commit $120B to AI buildout · Sherwood — Burry’s $1.1B options bet against Nvidia & Palantir · 新浪财经 — 段永平最新持仓:清仓阿里、新进特斯拉、加仓英伟达