Belvedere Trading 13F Portfolio

The biggest adds are not subtle; they are a coordinated grab for the guts of AI infrastructure.

• NVDA: Doubling the stake in NVIDIA (up 111.0% in shares, now 33.03% of the book) is an explicit call that GPU scarcity and pricing power still have legs. They’re adding roughly $208.7M at an average cost that’s already 27.9% in the money, suggesting they’re willing to press a winning position rather than “trade around” it.
• MU and SNDK: A new 12.5% in Micron plus a 755.5% increase in SanDisk to 10.26% says they see high-bandwidth and NAND/storage as the next choke points. Both are effectively leveraged calls on AI server configs and data proliferation rather than consumer PCs.
• INTC, AVGO, SMCI, KLAC, ASML: New stakes in Intel (4.22%), Broadcom (1.95%), Super Micro (0.54%), and KLA (0.09%), alongside a 34.5% add to ASML, round out a full-stack foundry thesis: from lithography (ASML, KLA) to networking/custom silicon (AVGO) to x86 and accelerators (INTC) to AI server assembly (SMCI).
• AAPL and AMZN: New Apple at 2.27% and a 568.3% increase in Amazon to 2.66% are less about betting on iPhones or retail and more about owning the hyperscalers and devices that will absorb and deploy this compute wave.

Even the ETF usage echoes this view: IYW, IGV, IGM, and VLUE all saw meaningful increases, but in tiny sizes relative to the single-name chip complex. The message: broad tech exposure is now a sidecar; the real bet is the semiconductor capex chain.

If the buys are about deepening into bottlenecks, the sells are about cutting generic or over-extended expressions of the same themes.

• META: A 70.8% reduction in Meta, freeing up an estimated -$59.0M, looks like a straightforward funding trade. With the position still slightly above cost (gain vs. average buy at 1.7%), they’re clearly prioritizing semis over ad-driven social platforms within the AI narrative.
• AMD and SOXL: The book saw a -56.8% cut to AMD and a -67.7% reduction in the leveraged semiconductor ETF SOXL, crystallizing massive gains (AMD at +211.4% vs cost, SOXL at +203.5%). That capital rotated squarely into NVIDIA, Micron, and SanDisk — a move from broad, leveraged beta to more idiosyncratic, single-name risk.
• Crypto beta vs crypto convexity: IBIT was slashed by -72.0%, while BITI (short bitcoin) was boosted by 251.2%. They’re bailing on long-beta spot exposure and replacing it with more tactical, convex instruments that can benefit from volatility.
• Precious metals reshuffle: GLD and SLVP were both cut hard (GLD -30.0%, SLVP -76.6%), yet they introduced GDX as a new gold miners position at 0.12%. That move upgrades from a pure metal ETF to higher-beta, operationally geared miners while freeing capital for chips and commodities.
• Single-stock leverage de-risking: AAPU and MUU were trimmed -41.2% and -24.5% respectively, while the underlying AAPL and MU exposures sit in outright common. They’re clearly comfortable with the fundamental bets, less so with embedded leverage.

The sector picture is deceptively simple: Technology rises from an estimated 70.96% to 78.35%. But beneath that are two important rotations: out of unclassified, often leveraged wrappers, and into clean, single-name exposure and more deliberate macro hedges.

The “Unclassified” bucket — a soup of ETFs and structured products — falls from 28.08% to 14.37%. Within that, they’re shrinking high-beta and tactical sleeves (SOXL, IBIT, MUU, AAPU, SLVP) while adding more stable or targeted exposures like value (VLUE), broad tech (IYW, IGV, IGM), and factor or index funds (IVV, MTUM, DIA, SPMO).

At the same time, they introduce a new 4.23% classified as Industrials via Tesla and expand Consumer Discretionary from 0.96% to 3.05% with AMZN and FDX. That hints at a second-order thesis: EVs, logistics, and consumer platforms will be key demand channels for AI and compute, not just abstract “tech.”

Elsewhere, real assets creep higher but in a more curated way. They add or increase copper (COPX, DBB), uranium and nuclear (URA, NUKZ), and multiple oil sleeves (UCO, USO, BNO, plus CRAK), while dialing back blunt precious metal and silver exposure. Taken together, the book is rotating from opportunistic thematic trades into a tighter core of AI infrastructure plus resource and inflation hedges.

Belvedere is positioning as if the AI build-out is not a trade but a multi-year industrial cycle — and as if the real constraint will be inputs, not end-demand. Heavy, incremental capital into NVIDIA, Micron, SanDisk, Broadcom, Intel, and the semi equipment ecosystem says they expect a sustained, capital-intensive race to add capacity and bandwidth.

The supporting bets line up with that macro: copper, uranium, oil, and rare-earth exposure all nod to the physical requirements of data centers, EVs, and grid upgrades. Tesla and FedEx add a layer of real-economy adoption — EV fleets, logistics automation, and time-sensitive shipping as beneficiaries of AI optimization.

On the risk side, the fund isn’t pretending this will be smooth. Additions to VXX and short-bitcoin BITI while shrinking spot IBIT show a preference for owning volatility and asymmetry rather than raw beta. The paired cuts in leveraged single-stock and sector ETFs also suggest they believe they can get all the upside they need from concentrated, unlevered stock selection.

Going forward, expect the center of gravity to remain in foundry and memory names unless there is a structural break in the AI capex story. The most likely tweaks from here are around the edges — rotating among resource exposures, volatility tools, and broad-factor ETFs — while the core AI infrastructure bet either compounds or forces a wholesale rethink of the book.