China’s Hedge Funds Shift AI Bets From Nvidia to Memory and Semiconductor Hardware

The combination is more informative than the Nvidia sale alone. It points toward a shift from the headline AI-chip beneficiary to memory, an increasingly important part of the hardware stack supporting large-scale AI systems. The filing reflects holdings at the end of June 2026, rather than a live view of the portfolio.

Greenwoods exits platform names and adds upstream suppliers

Greenwoods Asset Management made an even more visible change in its reported U.S. equity holdings. It exited Nvidia and Meta and also reported exits from Amazon and Alibaba, while opening positions in ASML, Applied Materials, and Applied Optoelectronics.

Those additions span chip-production equipment and optical networking. The resulting portfolio is less centered on companies monetizing AI applications or cloud services and more focused on the suppliers that enable additional semiconductor and data-center capacity.

Greenwoods’ reported U.S. equity portfolio also declined substantially during the quarter, so the changes should not be interpreted as a simple one-for-one switch from technology giants into hardware. The filings show both a reduction in the size of the disclosed portfolio and a reallocation within its remaining positions.

Oriental Harbor broadens its AI hardware exposure

Oriental Harbor Investment took a less absolute approach. It reduced Nvidia by about 202,000 shares and cut Alphabet, but Nvidia and Alphabet remained significant holdings at quarter-end. The fund also initiated positions in Intel, SanDisk, AMD, Marvell Technology, Arm, Broadcom, and Lumentum, according to reporting on its Q2 filing.

SanDisk was reported at approximately $176 million, while Intel became one of the portfolio’s largest positions. Oriental Harbor also increased its Micron holding. The pattern is therefore better described as rebalancing and broadening AI exposure than as a wholesale rejection of Nvidia or hyperscalers.

What the reallocations may signal

1. A search for relative value below the mega-cap layer

The filings suggest that these managers may see more attractive risk-reward opportunities in less crowded parts of the AI supply chain. Memory, storage, connectivity, optical components, and semiconductor equipment offer ways to participate in data-center investment without holding only the largest AI platforms.

This is a portfolio interpretation, not a guaranteed market outcome. The disclosures show what was held at quarter-end; they do not establish which sector the managers expect to outperform next.

2. More scrutiny of AI capital-spending returns

Reducing exposure to hyperscalers can also be consistent with greater caution about the returns on massive AI infrastructure spending. Cloud and platform companies must commit substantial capital before the timing and durability of AI revenue are fully clear. Hardware suppliers may offer a different exposure: they can benefit when customers order components and equipment to expand capacity, even while investors remain uncertain about how quickly end users will monetize AI services.

The filings alone cannot prove that this was the motivation behind every trade. They reveal positioning, not the internal investment thesis or the conversations behind it.

3. A more mature and selective AI cycle

The clearest broad conclusion is that the AI investment narrative may be becoming more granular. Instead of treating AI as a single trade dominated by Nvidia and hyperscalers, managers are separating the chain into compute, memory, storage, networking, optical connectivity, and manufacturing equipment.

That approach reflects a bottleneck-focused thesis: as AI clusters expand, the next opportunity may sit in the components required to build and connect them. But the Q2 filings do not show that this thesis will permanently outperform the mega-cap leaders. They show a shift in disclosed long positions during the quarter ended June 30, 2026.

How much weight should investors give the 13F data?

13F filings are useful for identifying large U.S.-listed equity positions and quarter-to-quarter changes, but they are delayed disclosures. They do not provide a complete picture of a manager’s strategy: short positions, derivatives, non-U.S. holdings, and trades made after the reporting date are not captured in the same way.

The strongest reading of these filings is therefore not “China’s hedge funds are selling AI.” It is that several influential managers are expressing AI exposure differently—moving from a concentrated bet on the most visible platforms toward a wider set of suppliers tied to memory, manufacturing capacity, computing hardware, networking, and optical infrastructure.

That makes the next question less about whether AI investment continues and more about which layer of the supply chain captures the economics of the next expansion.