Understanding markets beyond the obvious.

The period was defined by rotation. We closed four positions — two shorts and two longs — and moved the capital into a set of ideas where the setup had improved in our favour. The book is more concentrated in the memory and AI-infrastructure complex than it was a quarter ago, and carries two shorts against it.

The NVIDIA short was covered at a loss of 23%, inclusive of short-borrow and financing costs. The underlying view — that positioning and margin expectations had reached an extreme — has not changed, but the timing was wrong and the carry was expensive; we would rather step aside than finance a crowded trade indefinitely. The SpaceX short, by contrast, worked, and we covered it up 32%, before it reached our original target.

On the long side, Palantir and Palo Alto Networks were each sold into strength — up 34% and 25% respectively — ahead of their price targets, to free capital. In their place we re-established the memory complex, adding Micron and SanDisk, alongside United States Antimony after a period of price weakness. We also opened exposure to private-technology and AI infrastructure through CoreWeave, Destiny Tech100, and a re-established position in IREN.

Microsoft returns to the book — this time as a short, with a cover target of $460, reversing the long we stopped out of earlier in the year.