New 13F Signals: AI Wave Isn't Fading, Wall Street Is Just Getting 'Picky'

When AI becomes the market's greatest consensus, it means that although it remains the main theme, the era of 'buying AI with eyes closed' is coming to an end.

By Jim, MSX Maitong

Edited by Frank, MSX Maitong

Among the quarterly 13F filings, the least valuable take might be:

'Which stock did the big shots buy this time?'

Because 13Fs are inherently lagging snapshots of positions.

Under SEC rules, institutional quarter-end holdings can be disclosed up to 45 days after the quarter closes. The latest round of Q2 2026 13Fs reflects positions as of June 30, and the centralized disclosure deadline has already passed on August 14. More importantly, 13Fs primarily cover long positions in qualifying U.S.-listed securities; short positions and other exposures are not fully reflected.

Thus, they are not suitable for real-time 'copycat trading.'

But from another angle, the value of 13Fs is actually high—over the past three months, what exactly have the largest capital flows been buying and selling?

After sorting through the data, we identified a key signal: AI hasn't faded, but Wall Street is becoming 'picky' about AI.

I. The AI Consensus Remains, But 'Herding' Is Starting to Crack

If you only look at a few star funds, it's easy to be misled by individual trades.

What truly matters is the shift across the entire institutional landscape. Reuters' analysis of Q2 13Fs filed by 6,371 pension funds, hedge funds, and wealth management firms revealed:

• Nearly 44% of institutions reduced their 'Magnificent Seven' holdings, while about 42% initiated or increased positions—a near split;
• In semiconductors, however, the direction remained clearly net-long: about 48% were net buyers vs. 34.5% net sellers;
• In software, the opposite was true: among a group of major software companies, net sellers accounted for 28.2%, slightly higher than the 26.3% net buyers;

This indicates the AI consensus still exists, but internal divergence is accelerating.

After all, if Wall Street were systematically rejecting AI, the first sign would be a coordinated retreat across semiconductors, compute power, and data center chains.

That hasn't happened.

Chips remain a clear favorite among institutions, and AI infrastructure hasn't faced systematic dumping. Big money is simply starting to ask questions that weren't as critical two years ago:

Has the stock price already priced in this company's growth over the next two to three years? As AI CapEx continues to grow, who will truly convert capital expenditures into profits? If the market corrects, which asset classes have the most crowded institutional positions and are most likely to be the first sold off?

This is the most significant change in the Q2 13Fs: Wall Street is increasingly discussing 'whose AI bet offers better odds.'

Berkshire, Tiger Global, Third Point, and Duquesne (under Stanley Druckenmiller) happen to provide four completely different answers.

II. Four Institutions, Four 'Odds-Making' Approaches

1. Berkshire: Deploying Cash, Making a Large Bet on Google

In this round of 13Fs, Berkshire's moves regarding $Alphabet - C(GOOG.US) are particularly noteworthy.

At the end of Q1, Berkshire disclosed combined holdings of Alphabet Class A and C shares totaling about 57.84 million; by the end of Q2, this figure rose to approximately 106 million shares, an increase of over 80%.

Based solely on quarter-end market cap, Alphabet has now become one of Berkshire's most important publicly traded U.S. assets, alongside increased exposure to Delta Air Lines, Lennar, and other aviation and residential construction names.

Notably, Alphabet may be the least 'pure AI play' among the Magnificent Seven.

For years, one of the market's biggest concerns was whether generative AI would disrupt search entry points and erode Google Search's core commercial moat.

On the other hand, Alphabet still dominates Search, YouTube, Google Cloud, advertising, and boasts massive cash flow.

Berkshire's heavy bet suggests there may be room for revaluation in a company with strong cash flows and unproven core businesses, yet long criticized due to AI headwinds.

This is a fundamentally different trade from chasing the hottest AI winners.

2. Tiger Global: Reducing Big Tech, But Still Focused on Tech

Tiger Global's portfolio offers another typical sample.

In Q2, it cut its Alphabet stake from ~10.63 million to ~5.81 million shares, a 45.4% reduction; Broadcom holdings were nearly halved, TSMC also saw declines, and Microsoft, Meta, and NVIDIA were trimmed to varying degrees.

Looking only at this, one might conclude 'Tiger is exiting AI.'

But looking at what it bought tells a nearly opposite story.

Tiger established new positions in AMD, Applied Digital, and Cerebras in Q2, while the portfolio also includes AI compute and data center plays like Cipher Digital and Core Scientific; Intel holdings grew from ~1.64 million to ~4.25 million shares.

This looks more like an internal rebalancing of AI positions—reducing extremely crowded top-tier assets and shifting chips to the next layer of opportunities where market expectations are less uniform.

NVIDIA is the classic example. A fund can be long-term bullish on AI compute without needing to constantly increase its NVIDIA weight.

If position weights are already high, or if the stock price rises faster than earnings revisions, trimming may simply be portfolio management, not a reversal of industrial logic.

This is a crucial point for future U.S. equities: continued earnings growth does not guarantee stock prices will continue rising at the same pace as the past two years.

Because stock prices depend not just on 'how good the results are,' but on how much the market had already priced in beforehand.

3. Third Point: Taking Profits, Seeking the Next Wave of AI

Daniel Loeb's Third Point made even sharper moves.

In Q2, it completely exited NVIDIA, Broadcom, KLA, Lam Research, and the VanEck Semiconductor ETF (SMH)—core beneficiaries of the AI CapEx cycle—while also exiting Meta.

Taken alone, this looks like a large-scale 'AI de-risking.'

But Third Point didn't leave tech.

It significantly increased Alphabet and TSMC, and built new positions in Keysight and Flex; meanwhile, capital flowed into media, financial, and industrial names like Warner Bros. Discovery, Capital One, and Norfolk Southern. Warner Bros. Discovery became its largest public U.S. holding by quarter-end.

So Third Point is essentially taking profits from the easiest-to-understand winners of Phase I, seeking opportunities in the next phase that haven't been fully priced in yet.

Why did NVIDIA, Broadcom, KLA, and Lam Research become Phase I winners? Because their logic was too direct:

Larger models need GPUs; advanced chip expansion needs semiconductor equipment; larger AI clusters need networking, ASICs, and increasingly complex infrastructure.

The logic isn't wrong.

The problem is, once all investors know this logic, the driver of next-stage returns becomes whether actual growth can exceed already high market expectations.

This implies top-tier institutions believe the most obvious Alpha in Phase I AI is getting expensive.

4. Druckenmiller: Trading Only Expectation Gaps

If there's one fund that best explains this round of institutional thinking, Stanley Druckenmiller's Duquesne is the most typical sample.

At the end of Q1, it held Broadcom and Micron; by Q2, both disappeared from its 13F.

Meanwhile, Duquesne initiated new tech positions in Alphabet, AMD, and Palo Alto Networks, and continued increasing TSMC and STMicroelectronics holdings: TSMC rose from ~495k to ~590k shares, and STMicroelectronics from ~2.61m to ~3.10m shares.

At first glance, this seems contradictory: same sector, why sell some and buy others?

The answer lies in the key keyword of this round of 13Fs: expectation gaps.

If a company rises too fast, the market has already priced in two to three years of growth. Even if the long-term industrial logic holds, it's wise to take profits first.

Conversely, if another company's earnings cycle is improving but the market hasn't formed a consensus, it may offer a better risk-reward ratio even if it's not the hottest AI leader.

Correct industry judgment is just the first step. Buying at the right valuation, position, and considering how much the market already believes determines final returns.

III. From 'Buying AI' to 'Calculating Odds': What is Wall Street Trading?

Putting these four institutions together reveals the truly valuable signals.

First, Alphabet is transitioning from a consensus leader to a 'divergent asset.'

Alphabet may be the most interesting large-cap tech stock this round. Berkshire heavily added, Third Point and Duquesne increased or rebuilt positions, while Tiger Global slashed its stake by nearly half.

Top capital gave completely different answers to the same company.

The market is uncertain whether AI will ultimately weaken Google Search's moat or allow Alphabet's massive traffic, data, cloud, and compute base to further unlock.

From this perspective, bulls see cash flow, valuation, and potential AI-driven incremental gains; bears see search entry disruption, expanding CapEx, and structural challenges to old business models.

Such assets are often more worth studying than those 'everyone knows are good,' because true alpha comes from areas of market disagreement.

Second, the semiconductor consensus remains, but the 'blindly buy chips' era is over.

Overall 13Fs show chips remain a clearly net-long sector, with net buyers significantly outpacing net sellers.

But looking at star funds reveals huge internal differences. Broadcom: some trimming; TSMC: some adding, some cutting; AMD: some rebuilding; NVIDIA: evolving from an almost undisputed core AI asset into one requiring recalculated position costs and crowding metrics.

This shows semiconductors can no longer be traded as a single Beta.

GPUs, ASICs, foundries, memory, equipment, networking, and data center infrastructure all look like AI hardware, but their earnings cycles, supply-demand dynamics, and valuations are now vastly different.

In other words, AI hardware is moving from 'buying industry Beta' to competing on 'stock-specific Alpha.'

Another overlooked change: non-AI assets are returning to portfolios.

This isn't a rejection of AI, but rather a correlation-hedging move. Residential construction, aviation, financials, healthcare, media, railroads, and industrials are reappearing in major adjustments by top institutions:

Berkshire increased aviation and residential exposure; Third Point poured capital into media, financials, and railroads; Duquesne's portfolio was never solely AI-centric.

In a sense, precisely because AI has become the most visible and easily understood market theme, big money increasingly needs to find return sources with lower AI correlation.

Over the past two years, simply betting on the right AI direction generated massive returns. Moving forward, portfolio management will become ever more critical.

This may be the most important takeaway for ordinary investors from the latest 13Fs: seeing how the smartest, best-resourced capital begins to diverge on the same industrial theme.

Final Thoughts

For the past two years, the easiest trade in U.S. equities was finding AI and buying in.

NVIDIA, Broadcom, Meta, Microsoft, TSMC, and the entire semiconductor chain enjoyed multiple dividends from industrial growth, earnings upgrades, and valuation expansion.

But the latest 13Fs are sending a clearer signal—AI isn't over, but the era of 'as long as the direction is right, everything goes up together' is ending.

This is the most significant change in Q2 2026 13Fs:

AI hasn't faded, but herding is loosening.

The next stage won't test who dares to chase hardest, but who calculates odds best.