Opinion · Positioning · August 27, 2026
The Macro Is Not My Problem. My Concentration Is.
Every conversation about markets right now starts with the Fed. Core CPI is 2.5% and still above target, the FOMC held at 3.50–3.75% on July 29 in a 9–3 vote where all three dissents wanted a hike, and the next decision lands September 15–16. That is a genuinely live macro question. It is also, for this portfolio, close to irrelevant — because the largest risk in a 32-position book sized like mine is not what the Fed does in September. It is that roughly 60% of the book is one bet expressed seventeen different ways.
This piece is the honest version of “how am I positioned,” which means most of it is about a risk I built on purpose and have chosen not to remove.
1. The backdrop, stated plainly
Six readings define the environment this book is sized against. Every one of them is permissive.
| Indicator | Reading | Read |
|---|---|---|
| Core CPI (July) | 2.5% | Down from 2.6% in June. Still above the Fed’s 2% target. |
| Unemployment | 4.1% | Not a labour market forcing a cut. |
| Fed funds target | 3.50–3.75% | Held July 29 on a 9–3 vote. All three dissents wanted a hike. |
| 2-year Treasury | 4.17% | |
| 10-year Treasury | 4.68% | +0.51 spread. Uninverted and slightly steeper. |
| High-yield spreads | 2.71% | Credit is not pricing distress. |
The one asymmetry worth naming: the dissents were hawkish. A 9–3 hold where the minority wanted to tighten is a different setup from a 9–3 hold where they wanted to ease, and it means the September meeting is not a one-directional “when do we get the cut” question. But an uninverted curve and 2.71% high-yield spreads are not the conditions under which a concentrated equity book blows up. If something breaks in this portfolio over the next quarter, the macro table above will almost certainly not be where it started.
2. Where the risk actually sits
My written policy sets three hard caps and I check them every month. All three currently pass, and two of them pass comfortably. That is precisely what makes the picture misleading if you stop reading there.
Bars are current exposure; the vertical rule on each row is that exposure’s cap. Aggregate Nebius counts the NBIL 2× fund at double weight plus the direct NBIS position, as the policy requires.
Note what the caps do not constrain. There is no cap on thematic exposure — deliberately, and that is the single most important sentence in this piece. The policy stops me from putting more than 10% in one company and more than 25% in speculative names. It says nothing about putting 60% of the book into companies that all sell into the same buildout.
Semiconductors, AI infrastructure, and the power that feeds them, as a share of invested capital. Seventeen of thirty-two positions.
3. Why HHI says everything is fine
The book’s Herfindahl-Hirschman Index is 492, which corresponds to roughly twenty equally weighted positions. By that measure this is a well-diversified portfolio. HHI is computed from position sizes, and it is structurally incapable of seeing that SanDisk, Micron, Intel, AMD, Broadcom, Arm, Nebius, CoreWeave, Vertiv, Talen and Vistra respond to overlapping news.
This is not a criticism of HHI; it is a description of what it measures. But it is why a reader should not take “HHI 492” as reassurance, and why I built The Concentration Premium to demonstrate the gap using this book’s own correlation matrix rather than arguing about it abstractly.
The version of this that would actually hurt
The scenario that damages this portfolio is not a hawkish September. It is a durable repricing of AI infrastructure spending — the moment the market decides the buildout is financed on worse terms than assumed, or demanded in smaller volume than assumed. I have written the mechanism up twice this month, once from the financing side and once from the pricing side, and both times the conclusion pointed at my own book. Seventeen positions would move together, the leveraged one would move twice, and every hard cap in Figure 1 would still pass on the way down.
4. What I am actually doing about it
Nothing, and saying so
No positions opened, closed, added to or trimmed in the last month. That includes the SanDisk trim, now carried across three consecutive letters without action. It is entry 07 in the error ledger rather than a footnote here.
Measuring it correctly
Aggregate Nebius exposure fell from 8.46% to 5.54% this month because Nebius pulled back, not because I acted. Both numbers are now computed from a single source rather than maintained separately on each page.
Defending the choice, or not
The concentration is deliberate: it is where I have done the most work and hold the most conviction. That is a real argument. It is also exactly what someone would say who had simply bought the same story eleven times.
WatchWhat would change the positioning
1. SNDK back through 10%. That is a live breach and the Compliance Ledger will record it as one. It is 0.77 points away. 2. High-yield spreads widening past roughly 4%. The macro table stops being background at the point credit starts pricing the buildout’s financing risk. 3. A second quarter of memory pricing rolling over. Micron’s last quarter was almost entirely price rather than volume; the same mechanism runs in reverse, and MU and SNDK together are 15.0% of the book.
The uncomfortable summary is that the thing most likely to hurt this portfolio is not in the macro table, is not capped by my own policy, and looks fine by the standard diversification statistic. It is disclosed on three pages and constrained by none of them. I would rather state that in the same piece where I report a +72% all-time return than let the two facts live on separate pages.