The "divisive quarter" within 13F filings: Chinese concept stocks were bought as a group, while NVIDIA (NVDA) became the single most contested holding.
As the third-quarter 2026 13F disclosures drew to a close, a seemingly contradictory positioning picture emerged. Berkshire Hathaway, Bridgewater, Renaissance Technologies, Hillhouse, Greenwoods, Gaoyi, and Duan Yongping's H&H, among others, revealed their cards one after another, and Chinese concept stocks became a rare area of consensus, with PDD Holdings Inc (PDD), Alibaba (BABA), Trip.com (TCOM), and BOSS Zhipin (BZ) receiving varying degrees of increases; but in NVIDIA (NVDA), institutions essentially fought a "war of direction"—some liquidated 160 million shares, while others added more than six times their position. Consensus and division existing at the same time is precisely what makes this quarter's 13F worth reading closely.
Chinese concept stocks: from "avoidance" to "group buying"
The most uniform action in the third quarter appeared in Chinese concept stocks. Hillhouse's HHLR increased PDD Holdings Inc from 500,000 shares back to 4.447 million shares, a 789% increase, and PDD returned to its largest holding, with a period-end market value of $347 million, accounting for 22.46% of the portfolio; in the same quarter, HHLR increased its holding of Alibaba by about 2.3059 million shares, a rise of more than 1,250%, making Alibaba its second-largest holding. Gaoyi Asset Management made moves in Alibaba and BOSS Zhipin: its Alibaba position rose from about 3,399 shares at the end of the second quarter to 493,400 shares, rebuilding the position from a very low base; it increased BOSS Zhipin by about 2.2504 million shares, and after the adjustment, BOSS Zhipin rose to its largest holding with a 22.37% weight; Trip.com and NetEase were also increased. Duan Yongping's pace was more continuous. H&H increased PDD Holdings Inc by 2.9688 million shares in the third quarter, lifting PDD's share of its portfolio to 12.69%, making it the third-largest holding, with a period-end market value of about $2.182 billion. This was already Duan Yongping's fourth consecutive quarter of adding to PDD, with holdings rising from 7.23 million shares at the end of 2024 to 27.99 million shares, nearly four times the original amount. Li Lu's Himalaya Capital was also present, adding another 1 million shares of PDD in the third quarter, bringing its holding to 11.761 million shares, and PDD, calculated on a single-security basis, has risen to its largest position; in addition, Li Lu increased Tencent Music from 6.59 million shares to 14.824 million shares, an increase of about 125%. Foreign institutions were also moving toward Chinese concept stocks. Fidelity Investments added to JD.com, Vipshop, and KE Holdings, BlackRock added to JD.com, Trip.com, and JinkoSolar Holding, and JPMorgan Chase added to PDD Holdings Inc and BOSS Zhipin. After several previous quarters of valuation compression, Chinese concept stocks became a rare "consensus line" this quarter. Yet beneath the consensus, the choices were not the same: Gaoyi added to Alibaba, Hillhouse added to PDD and Alibaba, but both institutions also made increases and reductions in other Chinese concept stocks within their respective portfolios. The internal rebalancing within Chinese concept stocks is far more complicated than the phrase "collective bullishness."
NVIDIA: between liquidating 160 million shares and adding sixfold
According to 13F data from 497 institutions compiled by holdingsintel.com, NVIDIA was the single most divisive name this quarter: on one end was a decisive exit involving the liquidation of about 160 million shares and a reduction of nearly 99.7%, while on the other end was an increase of 648%, and neither side lacked landmark moves. The intensity of reductions varied notably: Sixth Street Partners liquidated about 160 million shares; Renaissance Technologies cut its holding from 7.0913 million shares at the end of the second quarter to 23,200 shares, a reduction of about 99.67%, from a prior position market value of about $1.42 billion, nearly a full liquidation, while at the same time raising Micron Technology to its largest holding and reducing most of its positions in Meta, Apple, and Amazon. Duan Yongping's H&H reduced NVIDIA by about 3.9612 million shares, with the position weight falling from 6.58% to 3.08%, but Duan Yongping himself later said that "the moat and demand remain very strong; I am not bearish but for now I will not add either," and noted that the reduction may have been partly related to options operations. Bridgewater reduced about 1.178 million shares, a decline of about 30%; Soros Fund Management reduced nearly 20%; and Gaoyi Asset Management simply liquidated its previously held 80,000 shares of NVIDIA. Wellington Management reduced about 2.0153 million shares, but still held about 120 million shares at period-end, with NVIDIA remaining its largest holding. The adding side was more about "rebuilding positions or multiple increases." ARK raised its stake by 85.43%, Point72 expanded its holding to about four times the original size, Third Point rebuilt a 1.5 million-share position after exiting in the second quarter, DekaBank increased by 648%, and Tepper was also among the buyers. Perhaps the most noteworthy detail lies in the timing window: this batch of 13Fs reflects holdings as of September 30, while the Jensen Huang departure controversy occurred on November 12. Institutions' NVIDIA position adjustments in the third quarter were more a rebalancing of valuation, position concentration, and AI compute expectations, rather than a "front-running" of later personnel changes. Reading NVIDIA reductions simply as "bearish on AI" would miss the real signal.
Switching positions rather than leaving: reallocating exposure within the AI supply chain
If NVIDIA reductions are viewed within larger portfolio changes, the direction of institutions becomes much clearer: funds did not systematically withdraw from technology, but reallocated positions within the AI supply chain. Bridgewater is a typical example. While reducing NVIDIA, Amazon, Alphabet, and Microsoft, it increased its Qualcomm holding from 162,000 shares to 878,000 shares, a rise of about 440%; it added 120,500 shares of Micron Technology, added 272,000 shares of Applied Materials, and opened a new position in Cisco. What it reduced were "compute leaders that had already risen a lot," and what it added were names in more upstream hardware segments such as memory, equipment, and radio frequency. Soros's path was similar but more extreme in direction: it reduced NVIDIA by about 20%, while increasing its Intel holding from 100,000 shares to 2.83 million shares, a rise of about 2,730%, and opened a new position in SK Hynix, with the logic pointing to "reducing expensive compute and adding cheaper logic chips and memory." Greenwoods' choice was another kind of "internal divergence." In the third quarter, it increased its TSMC holding by about 892,200 shares, bringing the stake to about 11.74 times the level at the end of the second quarter, and TSMC jumped to become its third-largest holding; at the same time, it added about 1.3241 million shares of Intel, an increase of about 48.1%, and opened a new position of about 304,200 shares in Meta. For NVIDIA, it only bought back 118,000 shares, with a market value of about $27.04 million, far smaller than its positions in TSMC and Intel, looking more like retaining exposure than making a heavy bet. Gaoyi's direction was almost the opposite of Greenwoods: it liquidated NVIDIA, Micron, and SanDisk, reduced TSMC, but more than doubled its holding of Alphabet Class C shares. On the same AI supply chain map, some were betting on wafer manufacturing, some on memory, and some on software platforms, and the disagreements among them were no smaller than those surrounding NVIDIA itself. Hillhouse's other filing entity, Evergreen, increased its Google Class A holding by about 288 times, making it the largest holding, while also increasing NVIDIA by 477%. Within the same institutional system, HHLR made a heavy bet on Chinese internet stocks while Evergreen made a heavy bet on Google and NVIDIA, and the two lines ran in parallel without conflict. This further shows that third-quarter position adjustments were not a binary choice of "tech versus non-tech," but refined division of labor by the same institution across different products and different strategy frameworks.
Berkshire's "no action" is also a signal
Amid all the position adjustments, Berkshire Hathaway's choice was especially quiet. At the end of the third quarter, its U.S. equity holdings were worth about $304 billion, and the only significant move among its top ten holdings was to continue reducing Bank of America, cutting another 27.533 million shares, with the position weight falling from 9.20% to 8.16%. Its 78.79 million Class A shares of Google and 27.19 million Class C shares were untouched, with the two classes together worth about $36.375 billion and still the third-largest holding; Apple was also unchanged, and with the stock up about 15% in the quarter, its market value rose from $65.950 billion to $75.901 billion, with the weight returning to 24.95%. The more striking figure is not in the 13F: Berkshire repurchased about $9.6 billion of its own stock in the quarter, remained a slight net seller of external equities overall, and still held as much as $377.4 billion in cash and short-term Treasuries. While others debated the direction of AI compute and redrew technology positions, Berkshire expressed another judgment through buybacks and cash: when it cannot find sufficiently attractive external targets, its own stock is the best allocation.
The real signal is not in the direction itself
There are three places in this quarter's 13F that are easy to misread: equating "reducing NVIDIA" with "bearish on technology"; reading "adding Chinese concept stocks" as "a trend reversal in Chinese concept stocks"; and reading Berkshire's "no action" as "no information." But looking at the three sets of data together, the conclusion is actually clearer than "bull-bear disagreement": institutions did not leave AI, but switched positions within the AI supply chain. In the NVIDIA divide, what was reduced was valuation and position concentration, not the AI direction itself. Chinese concept stocks were the intersection of valuation repair and position replenishment, not a full reversal. Berkshire's standing still was not a lack of judgment either, but another choice expressed through buybacks and cash.
Comments