13F Filings Show Institutions Trimmed Mag7 and AI Favorites as Conviction Splits

What Q2 13F Filings Reveal About Institutional Tech Positioning

A review of second-quarter 13F filings covering 6,371 pension funds, hedge funds, wealth managers and other institutions shows a narrow divide between buyers and sellers in some of the market's most crowded areas. Nearly 44% of filers trimmed their holdings of Magnificent Seven companies such as Microsoft and Meta Platforms, while 42% initiated or expanded positions in the group. The small gap suggests institutions were not abandoning megacap technology outright, but they were no longer adding to those names with the same conviction.

The pattern was more clearly constructive in semiconductors. Of the institutions that had filed by early Friday afternoon, 48% were net buyers of semiconductor names, while only 34.5% were net sellers. AI-themed stocks also drew interest: 36% of filers said they were net buyers of companies including CoreWeave, Arista Networks and Broadcom.

The filings do not disclose investor reasoning, but strategists interviewed by Reuters said the data may reflect risk limits and already-large positions more than a fundamental turn against tech. Because the positions are reported only through June 30, they also predate the July unwind in technology trades that hurt many hedge funds.

Why Big Investors Cut Tech Even as AI Spending Grew

Mag7 flows show an absence of consensus, not a wholesale exit

The 44% of institutions trimming Magnificent Seven stakes and the 42% adding to them are so closely matched that they read as a lack of agreement rather than a coordinated sell-off. Shaia Hosseinzadeh of OnyxPoint Global Management said closely matched buys and sells signal "the absence of consensus." The market may still believe in AI spending, but it is no longer certain which large technology companies will turn that spending into durable profits.

Position limits help explain post-earnings sell-offs

Steve Sosnick of Interactive Brokers pointed to a practical constraint: many large institutions may already be as long as their risk parameters or investment policies allow. That would explain why companies that reported good earnings still sold off afterward. If the natural buyers already hold their maximum positions, strong results cannot attract the usual incremental demand.

Tiger Global's cuts reflect the crowded-trade problem

Tiger Global Management disclosed reductions in Microsoft, Nvidia and Meta, and cut its Alphabet stake by 45.4% to 5.8 million shares. The fund also reduced Taiwan Semiconductor, while SoftBank did the same. Tiger did boost its Intel position. JPMorgan noted that crowded technology bets made it hard for hedge funds to exit without giving back profits, and Bruno Schneller of Erlen Capital Management described the July selloff as "a classic crowded-trade unwind amplified by leverage."

Energy and data centers show selective skepticism

Despite higher crude prices, institutions were net sellers of a group of major energy companies: 40.3% reported reducing exposure, while only 28% were buyers. Data center positioning was almost exactly balanced, with net buyers and net sellers each at 24.3%. OnyxPoint moved the other way, establishing new positions in BP, Devon Energy, Fervo Energy and Keel Infrastructure, suggesting that some investors see value outside the most crowded technology names.

Portfolio Implications for Funds Holding Crowded AI and Semiconductor Trades

For fund managers and investment teams reviewing their second-quarter positioning:

• Use the 44% versus 42% Mag7 split as evidence of no consensus, not a broad exit. OnyxPoint's Hosseinzadeh said closely matched buys and sells signal "the absence of consensus," so treat single-stock conviction as uncertain even where AI spending is not disputed.
• Separate semiconductor exposure from broad software exposure. In the filing data, 48% of institutions were net buyers of semiconductor names, while major software companies showed 28.2% net sellers against 26.3% net buyers.
• Avoid reading Tiger Global's cuts in Microsoft, Nvidia, Meta and the 45.4% reduction in Alphabet to 5.8 million shares as a market-wide shift. The filings are a point-in-time snapshot through June 30 and do not state motives.
• Reassess whether a large Mag7 or AI-heavy allocation has reached internal risk limits, a specific explanation cited by Interactive Brokers' Steve Sosnick for why good earnings did not prevent stock sell-offs.
• Question whether AI-related positions have become a leveraged momentum trade before adding to names such as CoreWeave, Arista Networks or Broadcom. Erlen Capital's Bruno Schneller said the sector moved from a fundamental growth story into that kind of trade in the second quarter.