Safe Growth

Recent Performance — Last 5 Sessions

Today = vs our Oct 14 mark ($1,781.31). Wednesday’s update was emailed and texted, so this is also the last figure you received; there is no separate emitted basis today. Live marks (Thu, 7:31 MDT, one minute after the open); the daily move is vs Wednesday's (Oct 14) published mark, which was also emailed and texted, so it is the same figure you received. TSMC's third-quarter results and Anthropic's third-quarter revenue disclosure both landed BEFORE this mark was taken, so the market's reaction to both is already inside these prices. Wednesday's tape fell after our mark, so the move against Wednesday's official close is SMALLER than the move against our mark; both are printed, for the book and for every position. October MTD starts from Wednesday Sep 30's official month-end close. October MTD vs Sep 30 close ($1,802.53). Wednesday’s official close was $1,762.42 — 1.06% below our mark — so on the close basis today is -1.3%.

Baseline note: today’s move is measured against our Oct 14 mark ($1,781.31), which is both the figure published on the website and the figure emailed and texted to you yesterday morning — the same number, so there are two bases today rather than three. Wednesday’s official close was $1,762.42, 1.06% below our mark, because the tape kept falling after we marked. Measured from that close the book is -1.3% rather than -2.4% — so today the headline figure is the uglier of the two readings, and the close basis is the one your own screen will agree with. We lead with the harsher number rather than the flattering one. The Today column in the position table is on the our-mark basis so it ties to the headline; the vs-close reading is printed beside it, and for three positions the two readings differ in sign. TSMC’s results and Anthropic’s revenue disclosure both landed before this mark was taken, so the reaction to both is already in these prices rather than still ahead of them. October MTD is measured against the Sep 30 official month-end close of $1,803. September is locked closed in the monthly table at -3.47%.

Position Scoreboard — Sorted by Portfolio Weight

Today = vs our Oct 14 mark, with the vs-Wednesday’s-close reading beneath it wherever the two differ. Since Entry = vs Jan 22, 2026. Weight = % of unit. Live marks (Thu, 7:31 MDT, one minute after the open); the daily move is vs Wednesday's (Oct 14) published mark, which was also emailed and texted, so it is the same figure you received. TSMC's third-quarter results and Anthropic's third-quarter revenue disclosure both landed BEFORE this mark was taken, so the market's reaction to both is already inside these prices. Wednesday's tape fell after our mark, so the move against Wednesday's official close is SMALLER than the move against our mark; both are printed, for the book and for every position. October MTD starts from Wednesday Sep 30's official month-end close.

Portfolio Allocation by Category

Live Track Record — Since January 22, 2026 (Actual)

▲ Current month through today; July, August and September 2026 are locked closed; return measured from the Jan 22, 2026 entry. * Jan–Mar 2026 are estimated partial-period returns carried from our May 2026 updates; they predate our daily archive (begins 15 April 2026) and are shown as estimates, not as-published daily marks. Apr–Sep cells are locked closed from official month-end closes; close-basis restatement pending the anchor review. No backtests, no hypotheticals. Units are price-return; dividends are not accrued.

Calculate Your Position Size

Safe Growth is marked at $1,739 one minute into the session, -2.4% against our Wednesday mark and -1.3% against Wednesday’s official close of $1,762.42. Note the direction of that gap, because it is the reverse of the one we have been explaining all week: the tape kept falling 1.06% after we marked yesterday, so today the headline number is the harsher of the two readings and your own screen will show the milder one. We lead with the harsher figure anyway, because it is measured from the number we actually sent you. October is -3.5% month to date against -1.2% yesterday, and the year is negative at -9.4%, its worst reading since early August. Yesterday’s update was emailed and texted, so there are two bases today and not three.

A correction before anything else, because it was in the email you received yesterday. Wednesday’s letter said the year’s figure was “the weakest this book has been in 2026.” That was wrong, and not by a little. This book traded below yesterday’s level on 31 separate sessions this year, and its actual 2026 low is $1,659.87, set on July 17 — about 4.6% below where it is marked this morning. The year-to-date number was right; the superlative attached to it was not, and it made a bad day sound like a record. We have added a check to the build that measures any claim like that against the book’s full daily history before it can print. Today’s -9.4% is a poor number and the worst since early August. It is not a 2026 low.

What happened this morning is a global unwind of the AI trade, and the important thing is which end of it broke. Not chip demand. The opposite: TSMC — the foundry that actually manufactures the chips this entire boom runs on — reported before the open and beat. Third-quarter revenue of $46.72 billion against guidance of $44.6–45.8 billion, up 50.9% on the year, gross margin 66.4%, net margin 51.1%. It reiterated capital spending of $60–64 billion for this year and said spending across 2026 to 2028 will be significantly higher than the past three years. The company with the clearest view of real AI demand told the market demand is fine and it is spending more. Semiconductors fell anyway. When the supply side beats and the stocks drop, the market has stopped arguing about whether the demand exists and started arguing about who is going to pay for the buildout.

And that is where the actual news landed. Anthropic disclosed third-quarter revenue of $16.8 billion before the open — a large number, and up roughly 45% on the quarter. The problem is the quarter before: that figure was up about 143%. Growth did not stop, it decelerated hard, and it did so at one of the two companies whose spending underwrites the demand forecasts for every GPU, every datacentre and every power contract in the AI complex. Reporting on both frontier labs now has OpenAI guiding to roughly 2.2x revenue growth for 2026 and Anthropic to 4x or less, where the market had been underwriting far steeper curves. Anthropic is also weeks from a November listing, and expectations for that valuation have slid below $1.5 trillion. Add Broadcom’s chief executive saying plainly that the binding constraint on AI expansion is access to capital rather than chip or memory supply, and you have the whole of this morning in one sentence: the cost of financing the buildout went up while the revenue expected to repay it came down.

The overseas tape shows the scale. South Korea’s Kospi fell 7.57% to 5,554.22, a seventh consecutive losing session and enough to put the index into a bear market after being the best-performing major market of the year; the Seoul press called it Black Thursday. Samsung and SK Hynix each fell more than 9%. The Philadelphia Semiconductor Index has now fallen six sessions running for roughly 10%, erasing on the order of $1.6 trillion of market value, and Wednesday alone took it down 4.3% and the Nasdaq 1.9% — the worst session for that index since July 23. This morning the Nasdaq 100 future was off about 1.3% while the S&P future was down only 0.3%, which tells you how narrowly the damage is aimed.

Here is the part of today worth your attention, because it is the first session in weeks where owning a diversified book clearly beat owning the exciting one. Five of the fourteen positions across both books are higher and nine are lower. The position that did the most work is the one this letter has spent the year apologising for: TMF is +9.7% against our Wednesday mark, +5.8% against Wednesday’s close — its best session of the year for us — as money ran out of equities and into Treasuries. The ten-year yield fell about nine basis points to roughly 5.10%, the thirty-year about eight to 5.43%, and the two-year about nine to 4.71%. That last one is worth a sentence on its own: yesterday the two-year rose as much as sixteen basis points on the hot inflation print, and today it was crushed back down as traders covered stretched bets against Treasuries into a falling stock market. So be clear about what TMF just did and did not do. This is a flight to safety, exactly as this letter described it on Tuesday and again on Wednesday, and flights to safety reverse. It is not a repaired thesis: TMF is still -24.5% since entry. We said that when the number was against us and we are going to say it when the number is finally for us.

Underneath that, the book behaved the way a diversified book is supposed to on a day the AI complex breaks. INDA is +1.8% against our Wednesday mark, +1.3% against Wednesday’s close, NUGT is +0.2% against our Wednesday mark, +2.1% against Wednesday’s close with gold finally acting like a haven rather than a rate proxy, and EWW is +0.5% against our Wednesday mark, -0.1% against Wednesday’s close. Three positions flip sign between our mark and yesterday’s close and each is worth checking against your own screen rather than our Today column: XLE is +0.2% against our Wednesday mark, -0.3% against Wednesday’s close, EWW as above, and PPA is -0.2% against our Wednesday mark, +0.3% against Wednesday’s close. The losses are concentrated exactly where this book touches the AI trade, which is the honest reading of why a “safe” portfolio was down at all: ETN is -6.6% against our Wednesday mark, -4.3% against Wednesday’s close on datacentre electrical demand, NVDA is -4.9% against our Wednesday mark, -3.4% against Wednesday’s close, URA is -4.1% against our Wednesday mark, -2.3% against Wednesday’s close on the same datacentre-power correlation we have flagged repeatedly and did not design, and the single MSTR share this book holds is -8.3% against our Wednesday mark, -4.3% against Wednesday’s close. Four positions carried the entire decline, and all four are the AI trade wearing different clothes.

Bottom line: Safe Growth is at $1,739 a unit — -2.4% against our Wednesday mark, -1.3% against Wednesday’s official close, -3.5% for October and -9.4% for the year. The AI trade repriced on financing cost and on slowing revenue at the frontier labs, the four positions here that touch it paid for that, and the long-bond and gold sleeves that have cost this book money all year are what cushioned the fall. Housekeeping, unchanged and stated once so the thread stays shut: our review of replacing TMF and NUGT is closed, we are holding both and every other position unchanged, and no trade was made at any point during that review. What we are watching: MSCI’s decision on Strategy tomorrow, Friday October 16, which touches the one MSTR share here and matters far more to the other book; whether the Treasury bid holds or reverses as these things usually do; the Federal Reserve on October 27–28; and the November 3 midterms.

Portfolio Builder | Market Update | Thursday, October 15, 2026

Trade Ledger — Entry & Exit History

Every position we open and close, newest first. Realized P/L on closes is marked against the original entry price.

Entry prices verified against the Jan 22, 2026 cost basis. “Open” = still held today. An entry marked “Closed <date>” was exited in that rotation — its realized result is shown on the CLOSE row for that date. This ledger builds our public track record going forward.