Pieter Slegers — Buy-Hold-Sell List: June 2026

SpaceX came to the stock market at $135 a share and was trading at $191 within a week, which valued the whole company at about $2.5 trillion — more than JP Morgan, more than Visa, more than Walmart. The question the article asks is simply what has to happen for that price to make sense.

Two arithmetic checks. Measured against sales, the company was valued at ninety times what it takes in — Nvidia trades at twenty, Tesla at sixteen. To get down to a still-generous twenty times sales it would need revenue of $88 billion, nearly five times today's, which at a very fast 30% growth rate takes about six years. Measured against profits, it would need roughly $50 billion of net profit; it currently makes none — in 2025 it lost $4.9 billion on $18.7 billion of revenue, and the artificial-intelligence arm alone burned through $7.7 billion in three months.

Morningstar's own estimate of what the business is worth is $780 billion, less than half the market price. The banks that arranged the listing earned over $500 million in fees. The general rule drawn from it — the archive's actual position — is that new listings are structurally priced in the seller's favour: "IPO… It's Probably Overpriced."

Games Workshop is the clearest illustration in this issue of separating a business from its share price. On quality it holds the firm's highest rating: a company they would happily own forever. On price, all three of their valuation methods say the same thing — it is expensive. The shares trade at about 33 times expected profits against a five-year average of 23, the fair-value model puts it roughly 50% above what it is worth, and working backwards from the price implies the company must grow profits at over 14% a year when the realistic expectation is 7%.

So the rating is Hold: keep what you own, buy nothing more. It is the mirror image of the portfolio's Strong Buys, where the business is unchanged and the price has fallen.

LVMH is the opposite anomaly to Games Workshop: every published model says it is cheap — roughly 38% below the fair-value estimate, a multiple a fifth below its own five-year average, and a reverse discounted cash flow that clears comfortably — and the rating is still only Hold.

Nothing in the issue explains why. It is worth noting because it shows the ratings are a judgement that the spreadsheets inform rather than determine, and because the judgement turned out to be the operative one: the position is sold out of the portfolio entirely in the September re-set, while the models still said it was cheap.

Berkshire is upgraded from Hold to Buy in this issue, and it is worth being clear that the published numbers do not support the upgrade. The earnings-based model puts fair value at roughly half the traded price; the multiple sits slightly above its own five-year average; and working backwards from the price implies the company needs to grow faster than anyone expects it to.

What is not in the models is the thing Berkshire actually is: a collection of wholly-owned businesses and a very large pile of cash and shares, whose value has little to do with a forward price-to-earnings ratio. The upgrade is a judgement about that, made in the same week the article argues the American index has become a concentrated bet on artificial intelligence. Berkshire is the alternative to that index — which is exactly the case the archive made in May.

CTS Eventim sells tickets and runs live events — concerts, festivals, sport — mostly in Germany and across Europe. It appears here for one reason: in a market where every valuation is being adjusted for what artificial intelligence might destroy, this is a business whose product is people physically turning up somewhere. As the article puts it, AI cannot replace a real-life experience.

It tops the firm's earnings-growth screen with an expected return of 21.2% a year, built from about 14% profit growth, a 2.6% dividend, and a shares price of 17 times profits against a "fair" 25. It is second on the reverse-DCF screen too: at today's price the market is implying the company shrinks slightly, while the expectation is near-14% growth. It is not on the Buy list — it is a watchlist name flagged for investors who want a return that does not depend on the AI question resolving one way or the other.

Mips is a small Swedish company that licenses a patented safety system built into helmets — a low-friction layer that lets the shell rotate slightly on impact, reducing the twisting force transmitted to the brain. It does not make helmets; it charges the manufacturers a fee per helmet, which is a very light way to make money.

The shares now trade at about 24 times expected profits against a five-year average of 60 — the cheapest the company has ever been — while the expectation is that profits roughly triple over the next few years. The single caution is in the third model: at this price the market is already assuming about 15% growth, so this is cheap relative to its own history rather than cheap in absolute terms. The tripling has to happen.

Medpace runs clinical trials for small and mid-sized biotechnology companies. It is paid to conduct the trial, not for the drug to succeed, which is why the archive holds it as a way to be exposed to drug development without betting on any individual drug.

The upgrade from Hold to Buy here is marginal on every measure — a 5% discount to fair value, and a multiple fractionally above its own five-year average. That is the point rather than a flaw: this is the company the archive has always described as a wonderful business at a fair price, so the upgrade reflects the price catching down to fair rather than becoming a bargain.

Fairfax is rated Buy here, but it is worth recording how thin the numerical case is: an 8% discount to fair value, a multiple slightly above its own history, and a reverse discounted cash flow where the growth the price requires and the growth expected are exactly equal — no margin at all.

What carries the rating is the record: 33.4% a year over five years, achieved by an insurer that invests the money it holds between collecting premiums and paying claims. Two months later the firm buys it with $50,000, on the argument that it is a smaller Berkshire with Indian growth attached — and that purchase, like this rating, rests on the track record rather than on the valuation.

Constellation is the largest discount in the whole book: the firm's fair-value estimate is more than two and a half times the traded price, and the shares change hands at about 16 times expected profits against a five-year average of 32. In other words, the business is not being valued differently because it earns less — it earns more — but because the market has decided software companies are threatened by artificial intelligence.

The reason the archive is comfortable is the shape of what Constellation owns: hundreds of small, dull, essential programs, each one cheap for the customer relative to the cost of replacing it. That is also what makes it the name a well-known outside investor was reported buying four days earlier, in the deepest share-price fall in the company's history.

Novo Nordisk carries the widest split in the portfolio: the highest possible valuation rating — Strong Buy, with the best expected return of any holding at 19% a year — and one of the lowest quality convictions, Medium. That combination is unusual and it is a warning, not a bargain signal.

The numbers behind it: the shares are at 12 times expected profits against a five-year average of 28, so more than half the multiple has gone. But the third model, which works backwards from the price, says the market is implying 10% growth while the firm only expects 8.6% — the one test that says the price is not cheap enough. Reading the two together: the market has stopped believing the growth story, and the archive is not fully sure the market is wrong.