Step One: Trimming the Infra-AI Side

This week was housekeeping with a direction. The discretionary work across the books was almost entirely in one direction — taking size down in semiconductors and the infrastructure side of the AI trade. Not a market call so much as sizing discipline, and step one of a posture.

The bottom line on why we reduced: the infrastructure-AI trade looks exhausted, and the negative divergences under the surface keep persisting — the index holds while the internals weaken, and our own divergence and breadth tools have been flagging it for weeks. When leadership tires and breadth won't confirm, the most-stretched, most-crowded names are the first place we take size off. For the deeper read — the capex-to-monetization bridge and the Z.ai repricing — see this week's Daily Look #099.

What we reduced: ASX, TSM, COHR, DDOG and NBIS in AI Build-Out; STM, TOELY, ATEYY and TSM in Global Tech 50; ASX in Hypergrowth. Trimmed, not exited — the theses are intact, the size wasn't. The Global ETFs book pared back too — out of EEM, SMHX and XMMO, reducing emerging-market, semiconductor-heavy and momentum-concentrated exposure so the book leans less on any single theme.

The buy side was small and deliberate. We opened KLIC and MKSI in AI Build-Out — selective adds in the equipment and process-control layer, the part of the build-out we still want exposure to even as we trim the crowded large-caps. And the Volatility Harvesting engine wrote one short put, on ALAB — the structure we use when we're happy to own a name lower while collecting premium in the meantime. That was the whole of the buy side; the rest of the tape was routine dividend reinvestment.

So: this week was almost all reductions — step one. The lean toward the demand, agentic-winner side is the next move, not this one: if the divergences keep persisting, that's where we add, on the setups, in size we can hold — possibly as soon as next week. We're not calling a top and not trading a headline; the rate question is still the senior signal and the trend hasn't broken. Probability, not prophecy. Three engines, one rulebook. No heroes.