The 46% surge in total holdings is not the story. The story is where the capital went—and where it didn’t.
On August 12, Dongfang Hongyuan Overseas Fund, managed by Dan Bin, filed its 13F with the SEC. Total U.S. equity holdings hit $1.65 billion, up from $1.13 billion at the end of Q1. That is a 46% increase in a quarter where the Nasdaq Composite rose only 8%. The divergence is a signal.
Volatility is the tax on unverified trust. Trust in the narrative of AI had been priced into NVIDIA, TSMC, and Meta. The data now shows that trust is being transferred—from the frontrunners to the foundation builders.
Context: The Data Methodology
I analyzed the Q2 2026 13F filing of Dongfang Hongyuan Overseas Fund. The filing lists all U.S. equity positions above $200,000. I cross-referenced the Q1 baseline to calculate delta for each holding. The methodology is straightforward: change in share count multiplied by average price during the quarter. The goal is to isolate the capital flow direction, not the market price movement.
Pattern recognition precedes prediction. The fund’s behavior follows a clear pattern: rotate out of high-beta, high-valuation AI names and into the physical layer of the AI stack. This is not a panic sell. It is a structural reallocation.
Core: The On-Chain Evidence (In Traditional Finance Disguise)
The Reduction List
- Google C (GOOG): Reduced. Still the top holding at $370 million, but the reduction is notable.
- NVIDIA (NVDA): Reduced. The poster child of AI was trimmed.
- TSMC (TSM): Reduced. The sole manufacturer of advanced chips saw a stake decrease.
- Amazon (AMZN): Reduced. Cloud and AI infrastructure player.
- Meta (META): Reduced. Social media’s AI ambitions.
These five names represent the “AI narrative” portfolio. The reduction is not a vote against AI. It is a vote against the valuation of the narrative. The fund is monetizing the hype.
The Exit List
- Google A (GOOGL): Fully exited. A dual-class share structure complication.
- Apple (AAPL): Fully exited. Consumer hardware, not AI infrastructure.
- CRCL: Fully exited. Unclear ticker, likely a small position.
- Tesla (TSLA): Fully exited. Electric vehicles, not AI compute.
- Direxion 2x Daily GOOGL ETF: Fully exited. Leveraged ETF.
- ProShares 3x Daily NASDAQ ETF: Fully exited. Another leveraged ETF.
Liquidity evaporates when logic fails. The exit from leveraged ETFs signals a reduction in speculative exposure. The fund is not just rotating; it is cleaning house.
The Addition List
- Intel (INTC): New position. The underdog of semiconductors.
- SanDisk (WDC): New position. Storage technology.
- Advanced Micro Devices (AMD): New position. Direct competitor to NVIDIA.
- Marvell Technology (MRVL): New position. Data infrastructure chips.
- ARM Holdings (ARM): New position. Architecture licensing.
- Broadcom (AVGO): New position. Networking and custom chips.
- Lumentum (LITE): New position. Optical communication components.
The Increased Position
- Micron (MU): Increased stake. Memory and storage.
The Unified Thesis
The new positions and increases share a common thread: they are all upstream of the AI application layer.
- Semiconductor manufacturing: Intel, AMD, Marvell, Broadcom, ARM.
- Computing hardware: Intel, AMD.
- Storage: SanDisk, Micron.
- Optical communication: Lumentum.
This is not a random bet. It is a concentrated wager on the physical infrastructure that enables AI workloads. The fund is betting that the next phase of AI growth will not be driven by algorithms, but by the availability of chips, memory, and bandwidth.
History is written in blocks, not promises. The blocks here are silicon, not blockchain. But the principle holds: verify the flow of capital, not the press release.
Contrarian: The Blind Spot in the Narrative
Conventional wisdom: AI is a bubble, and the smart money is taking profits. The data supports that—the reduction in NVIDIA, TSMC, and Meta aligns with the bubble narrative.
But the contrarian angle is subtle. The fund is not rotating out of AI entirely. It is rotating into the hardware layer. The narrative of “AI is over” is incorrect. The correct narrative is “AI is migrating from the software layer to the physical layer.”
Wash trading is the ghost in the machine. In traditional finance, the equivalent of wash trading is the illusion of liquidity in high-flying stocks. The reduction in NVIDIA and TSMC may be interpreted as a lack of conviction, but the addition of Intel and AMD tells a different story. The fund is swapping one form of AI exposure for another, more tangible form.
Correlation ≠ causation. The 46% increase in total holdings could be misinterpreted as bullishness. In reality, it is a rebalancing. The fund added low-valuation hardware stocks while shedding high-valuation AI stocks. The net effect is a larger portfolio with lower average beta.
Takeaway: The Next Signal to Watch
In the next 90 days, monitor the following:
- Intel’s foundry progress: If Intel gains a major customer, the fund’s thesis is validated.
- ASML’s earnings: As a proxy for semiconductor equipment demand.
- Micron’s guidance: Storage demand is a leading indicator for AI compute buildout.
- Other fund filings: Look for similar rotations from other institutional investors.
In the noise, the signal remains silent. The signal from Dan Bin’s Q2 filing is not about the 46% increase. It is about the composition of that increase. The market is still digesting the idea that AI’s next leg is physical, not digital.
As a quantitative strategist, I have seen this pattern before. In 2021, the rotation from DeFi to Layer 1 blockchains. In 2023, the rotation from centralized exchanges to self-custody. Each time, the data preceded the narrative.
The truth is buried in the timestamp. The timestamp here is Q2 2026, and the truth is that the hardware cycle is beginning. The next wave of alpha will be found not in the stocks that everyone knows, but in the stocks that enable the known stocks to operate.
Follow the hardware. Ignore the hype.