2025 full year: slightly over 100%.
2026 YTD: +45.0%.
2026 month by month:
Jan −5.9% → Feb −26.5% → Mar −10.2% → Apr +75.1% → May +54.9% → Jun −9.7% → Jul −21.0% → Aug −2.3% → Sep +18.7%
Position weights
| Ticker | Jul 31 | Aug 31 | Sep 25 |
|---|---|---|---|
| IREN | 23.08% | 24.25% | 25.82% |
| ALAB | 18.13% | 16.56% | 18.21% |
| NBIS | 14.04% | 14.56% | 15.00% |
| APP | 15.60% | 15.08% | 13.45% |
| NVDA | 10.21% | 10.74% | 9.82% |
| CRDO | 7.47% | 7.81% | 6.52% |
| NET | 4.77% | 4.99% | 5.11% |
| CBRS | 5.51% | 4.89% | 4.92% |
| HIVE | 1.19% | 1.12% | 1.14% |
No sells in the entire quarter.
What I did
July — two adds.
- IREN: added roughly 20% to the position, in two lots, as the stock came down from its earlier highs.
- CBRS: opened a new position, which started at about 5% of the book.
August — two adds.
- APP: added roughly 28% to the position.
- IREN: added roughly 9% to the position.
September — nothing.
IREN
I’m very happy with the FY26 report. What I care about is the ARR build, not the GAAP quarter.
The quarter looks ugly: revenue of $137M was down 26.7% YoY, and the full year showed a $703M net loss. Two things are worth separating out. First, the loss is largely non-cash impairment from retiring miners as the sites convert to AI Cloud — mining is being deliberately wound down and substantially retired by this December. Second, the revenue decline is that same deliberate replacement, not a demand problem.
The number that matters is the contracted book: ~$4B of ARR is contracted against 2026 capacity, roughly $1B of it already operating as of late August, with management guiding to exit the December quarter above $4B as Horizons 2–4 come online. Note they raised that target — it was $3.4B as recently as the Q2 call.
So my expectation is specific and I’ll state it plainly: if they exit the year at $4B running ARR, then the March 2027 quarter should print roughly $1B of revenue. That is about eight times the $137M they reported for the June 2026 quarter. I’m not looking at trailing growth here — I’m looking at the next three quarters, which is a deliberate difference from how I usually read a company.
That is what makes IREN interesting to me right now. It is not a growth story on the income statement yet. It becomes one over the next two quarters, and when that revenue lands, a much larger group of growth investors will be able to own it on their own terms.
Why do we trust the $4B? I went back and checked this management’s record against its own promises: they targeted 15 EH/s, then 30, then 50 EH/s, and delivered each one, hitting 50 by June 2025 before pausing mining to fund AI. They put Horizon 1 in front of Microsoft and Microsoft accepted it. ARR went from $28M in mid-2025 to ~$500M at the June quarter and ~$1B once Horizon 1 was accepted. A team that keeps hitting its stated numbers earns the benefit of the doubt on the next one.
I don’t care if a data center slips a quarter. I care that the story is intact. A delay in commissioning changes when revenue is recognised, not the size of the contracted book. What would change my mind is a customer walking away, or pricing on new capacity falling — and pricing is going the other way, with recent three-year contracts above $20M of revenue per MW and active discussions nearer $25M/MW.
The risk I’m actually watching is funding, not demand. FY27 capex guidance is $25–30B against a ~$17B market cap. Customer prepayments are covering 45–55% of GPU capex and the data center portfolio is unencumbered. I’d rather not be diluted at these prices, but nothing about the financing has become difficult yet, and one supporting detail: the market for used GPUs has not softened the way a lot of people expected, which means the residual value behind the borrowing is holding up. A delay in the build-out is a timing cost. A funding problem at a bad price is a real cost. That’s the line I’m watching.
APP
APP is the one position where I think the market is simply wrong on the multiple, and Q3 gave me a better price on it. The stock fell 34.9% over the trailing 90 days and closed the quarter at $310.75, down about 54% year to date from a $733 peak at the end of last year.
The business did not fall apart over that period. Revenue grew 53% YoY to $1.92B, with a 77% operating margin and a 64.6% net margin. On today’s price that is roughly 16x forward earnings and about a 6% levered free cash flow yield, up from roughly 2% at the end of last year. For a business still compounding revenue above 50%, that is not a growth multiple.
Which is the point: at this multiple APP no longer needs the growth investor to keep believing. It needs a value investor to notice that the earnings are real. That is a different buyer, with a different test — and on the numbers, that buyer should be able to show up.
So: one more quarter to prove itself. Q3 earnings around November 4 is the test. If the deceleration shows up, I’m wrong and I’ll act on it. If it doesn’t, I own a business earning 65% net margins at roughly a market multiple.
I won’t spend much space on the bear case, because I don’t think it changes the arithmetic — the market has already marked the stock down 54% for it. Briefly: there is a lawsuit, and there are questions about how much more share its ad platform can take. If the second one is right, the growth is worth less than I think, and the multiple is a trap rather than a gift. That’s the bet.
On this board
It’s sad to watch the board thin out. The threads this past week — thanking the moderators, asking whether the board can be saved, asking after Saul — say it better than I can.
I started posting again last year precisely because I wanted to put something back into a place that taught me most of what I know about investing. I don’t post as often as I’d like. Work and my own research have me past midnight most nights, and 24-hour stretches happen more than I’d care to admit, so I don’t have the time I’d need to carefully tune my own tooling to post at the volume I’d want.
What actually differentiates us
Here’s the thing I keep turning over. In the AI era, what is actually our edge?
Not stock picking. Building an agent that follows something like this strategy is not hard anymore, and it will get easier. Anyone can get a decent screen, a decent summary and a decent bull/bear case in an afternoon.
I think the edge is that we’ve held this approach long enough to have conviction in it — and conviction is what lets you wait. We’re willing to wait years for growth-stock valuations to mean-revert. We’re willing to sit through a −70% drawdown without selling the thesis.
That is not an analytical advantage. It’s a temperamental one. And I’d argue it’s the harder one to replicate, because an AI can hand you a better analysis in seconds but it cannot make you hold when you’re down 70% and everyone is telling you that you were wrong.
I want to be careful about what this claim does and doesn’t say. It is not “conviction always wins.” It is that these rules are one set of rules, and the drawdowns — including the −70% ones — are part of the price of the roughly 30% annualised return the approach has produced over the long run. You cannot take the compounding without taking the drawdowns; they are the same decision. Conviction will sometimes lose money. It’s that over the full history it has made more than it lost, and the people who got shaken out at the bottom never collected the part that came after.
The uncomfortable corollary is that this only works with rules you can actually hold largely through the drawdown. Which is why the rules themselves have to be worth holding.
Disclosure
The numbers and the trades in this post are mine and are accurate. The prose is a joint product — I assembled it with my AI agent from my trade records and my own notes, and I haven’t had the time to rewrite it line by line. So I can’t put quotation marks around “the AI parts,” because at the paragraph level the two are mixed and I can’t cleanly separate them.
Please read it with that in mind and use your own discretion.
Because of that — and because I don’t have time to defend every paragraph — I’m keeping this post short, and I’ll be replying very selectively if needed at all. Hope we still see most people in the next quarter.