13Fs Show an AI Reallocation, Not an Exodus

A 13F is a lagging photograph, not a manifesto. The weekend still treated the August 14 cutoff as a verdict on AI. Reuters’ count of 6,371 institutional filers is the better caption: 44% trimmed Magnificent Seven exposure, 42% initiated or added, the rest stood still. That is a hung jury. Semiconductors kept a distinctly positive bias — 48% net buyers against 34.5% sellers. The named managers did not settle the case. They filed competing maps of the same stack.

Berkshire Hathaway’s map is Alphabet. Class A rose 45% to 78.8 million shares; Class C rose more than sevenfold to 27.2 million. Combined, about 106 million shares worth roughly $37.8 billion, vaulting Google’s parent to the third-largest disclosed equity after Apple and American Express. About $10 billion of the add arrived as a June private placement earmarked for AI infrastructure; the rest was open-market. That is not a rotation away from megacap tech. It is a bet that the cash-flow layer of the buildout still belongs to a search-and-cloud compounder — the same prove-it mood that shrugged at Nvidia’s cleanest quarter.

Same Stack, Different Floor

Tiger Global filed the opposite floor plan. Chase Coleman cut Nvidia, Microsoft, Meta, and Amazon, and slashed Alphabet 45.4% to 5.8 million shares. Broadcom was halved; TSMC was trimmed. Intel more than doubled, to 4.25 million shares. New lines appeared in AMD and SpaceX. Tiger did not abandon technology. It changed where in the technology stack it wanted exposure — foundry-and-satellite optionality instead of the crowded megacap tape that August’s cloud prints had already turned into a scarcity bid.

Renaissance Technologies went the other way inside the same neighborhood. Nvidia became its largest disclosed holding after an add of 4.56 million shares, to 7.09 million, worth about $1.42 billion. Meta and Intel sit just behind. For a book of thousands of names, a 2% sleeve as the top line is a shout. The quarter is dispersion within AI, not an AI-to-not-AI migration. Culled’s dot-com echo already said the capex curve is not the cash-register. These 13Fs are the cash-register argument, manager by manager.

Tollbooths, Miners, and the Power That Is Not “Energy”

Bill Ackman’s overhaul is the cleanest non-GPU map. Pershing Square added six names — Netflix, Visa, Mastercard, Alcon, Intercontinental Exchange, and S&P Global — the biggest reconstruction in years for a fund that usually lives inside a dozen names. Two of those (ICE, Alcon) were bought after June 30, so they will not fully photograph in this 13F. The thesis still holds: a concentrated investor is broadening beyond the Microsoft/Meta/Amazon/Uber complex into payments, financial infrastructure, media, and healthcare. Toll-takers on transactions and data. That is Ackman looking for a contract the AI narrative cannot reprice overnight.

Stanley Druckenmiller’s Duquesne filing looks, at first, like an AI exit: Broadcom, Intel, and Micron gone. Then the rest of the page. Amazon shares rose more than 1,000%. New AMD. New Alphabet A. More than 4 million Bitdeer shares and a 754,800-share Riot line — miners remodeling halls for compute, the channel Culled tracked when Bitcoin campuses became Nvidia landlords. That is not “he doesn’t like Nvidia anymore.” It is a rotation toward the infrastructure and financial assets being built around AI.

Energy, as a sector, was not the destination. Reuters found 40.3% of filers were net sellers of a dozen major energy names, against 28% buyers. AI-themed names — CoreWeave, Arista, Broadcom — still drew net buying from 36% of institutions. Data-center stocks split 24.3% each way. OnyxPoint’s specific book is the better sentence: new BP, Devon, Fervo geothermal, and Keel Infrastructure. Not AI versus energy. AI versus the physical plant required to make AI possible — custom silicon and power as one constraint.

The value sleeve is equally un-uniform. Li Lu’s Himalaya Capital did raise PDD 133.5%, to 10.76 million shares and about 22% of a $3.7 billion U.S. book. David Tepper’s Appaloosa exited PDD, JD.com, KWEB, Microsoft, and UnitedHealth in the same quarter — then kept Amazon, Micron, and TSMC as a near-40% AI trio. Healthcare produced concentrated bets and concentrated exits, not a rotation.

Watch three things. First, whether Berkshire’s Alphabet sleeve is a one-quarter Abel-era signature or a multi-year third pillar. Second, whether Tiger’s Intel/AMD/SpaceX map survives a year of packaging and power bottlenecks. Third, whether Ackman’s tollbooths and Druckenmiller’s miner-compute names outperform the GPU complex — the only test of “where the money is made.” Until then, treat the 13Fs as an argument over the stack, not a funeral for it.