Swiss National Bank

The single loud decision in an otherwise incremental quarter is the move in KLA. Taking KLA up to 0.58% of the book, up +858.7% in shares with about $1.00B of incremental capital, is a deliberate call that process control and inspection are the choke point of the next semiconductor cycle.

That KLA sits slightly underwater versus their average cost (gain_vs_avg_buy_pct at about -9.2%) makes the add even more revealing. They are averaging down into a core piece of fab infrastructure rather than chasing what worked.

The second, quieter expression of the same view is SanDisk:

• SNDK: exposure increased +5.1% in shares with roughly $32.0M more capital and a gain_vs_avg_buy_pct near +966.7%, signalling they are adding to a huge winner in NAND and storage, not backing away.

Taken together, the biggest buys say the bank is betting that AI’s next leg is about capacity: more wafers, more bits, more yield — not just more GPUs.

On the sell side, the pattern is methodical: lighten the biggest, most profitable winners just enough to fund fresh risk further down the stack. NVIDIA alone freed roughly $1.24B of capital, with shares cut -8.7% while the position still sits at 6.81% of the book and about +735.8% versus their average buy.

The same playbook shows up in Apple, Microsoft and Alphabet (both GOOG and GOOGL), each trimmed in the -3.9% to -8.7% share range and all sitting on multi‑hundred‑percent gains. These are not thesis breaks; they look like board‑approved risk management in names that have run far ahead of their original underwriting.

Even Amazon and Broadcom were tapped as funding sources, with modest cuts and still-strong gains. When your biggest trims are all huge winners and your only real add is a cyclical equipment name, you’re clearly less worried about AI demand rolling over than about price‑to‑perfection at the top of the stack.

At the sector level, the portfolio looks almost eerily stable: Technology at 65.72% versus 65.63% last quarter, Consumer Discretionary at 10.83%, Health Care at 8.71%. The bar chart says “nothing to see here,” but the stock‑level moves tell a different story.

Within Tech, the internal rotation is from mega‑platforms and consumer‑facing growth toward semis, equipment, and storage. KLAC’s promotion and the incremental SNDK buy, combined with trims in NVIDIA, Broadcom, Intel, and the broader software complex, shift risk from demand proxies to the capacity enablers.

Outside Tech, the book is essentially ballast. Health care (Lilly, J&J, AbbVie, Merck, AstraZeneca, UnitedHealth) and Consumer Staples (Coca‑Cola) all see small trims around -3.7% to -8.9% in shares, keeping sector weights flat while releasing cash. Real-world Financials exposure comes via Visa, Mastercard, and Berkshire, each gently reduced, consistent with a manager using defensives and quasi‑bond proxies as funding for higher‑beta AI infrastructure bets.

Put together, the 2026-Q2 changes read like an institution that still believes in AI‑driven earnings power but is no longer comfortable paying peak multiples for every incremental dollar of exposure. The 10.71% latest‑quarter performance and strong multi‑year track record give them room to harvest, and they are taking it.

Recycling more than $1.24B from NVIDIA and hundreds of millions from other megacaps into KLA and storage says they expect the bottleneck to move from GPUs to fab throughput and memory. If that view is right, the current quarter’s under‑the‑surface rotation should show up as relative resilience in the next capex cycle, even if the AI narrative gets noisy.

At the same time, the near‑unchanged sector weights and modest trims in health care, energy and staples show a central bank that still has to look like a diversified, systemically cautious allocator. The forward message from this book is clear: the AI theme is intact, but the easy beta is gone — Swiss National Bank now wants to own the machinery and the bits that everyone else will have to buy, regardless of which front‑end platform wins.