Institutional investors show caution toward tech favorites in US 13F filings

The quarterly 13F filing season functions like a slow-motion replay of what the smartest money in the room was actually doing while everyone else was watching headlines. The Q2 2026 edition, covering 6,371 institutional investors, tells a story of measured retreat from the stocks that defined the last bull run.

Among the group Wall Street calls the Magnificent Seven, which includes Microsoft, Meta Platforms, and their megacap peers, 44% of institutions trimmed their holdings last quarter. Only 42% increased or initiated positions.

A sector-by-sector split

Software had a rough quarter in the filing data: 28.2% of institutions reported as net sellers, compared to 26.3% as net buyers.

Semiconductors told a different story entirely. Some 48% of filing institutions were net buyers of chip stocks, against 34.5% who were net sellers.

Tiger Global Management reduced its stakes in several key tech names and also cut its position in Taiwan Semiconductor.

AI enthusiasm, with caveats

AI-themed stocks sit in an interesting middle ground. Roughly 36% of filing institutions were net buyers of names like CoreWeave and Arista Networks, companies tied directly to the infrastructure buildout underpinning large language models and data center expansion.

Market observers have flagged July’s unwinding of crowded trades as context for some of this positioning. The oil sector fared even worse in institutional sentiment, with institutions broadly net sellers of energy holdings.

What the data means for the months ahead

The semiconductor signal is probably the cleanest takeaway. When nearly half of a large institutional sample is adding chip exposure while pulling back on software and platform names, it suggests conviction in the physical infrastructure layer over the application layer.

The CoreWeave and Arista Networks buying points to a specific thesis: institutions that remain committed to AI are increasingly expressing that view through companies closer to the data center hardware stack.

Software’s modest net-selling position is a quieter concern. The 28.2% versus 26.3% spread is not alarming yet, but enterprise software multiples remain elevated relative to historical norms.

Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our Editorial Policy.