Pieter Slegers — Portfolio Update March 2026

LVMH owns Louis Vuitton, Dior, Moët & Chandon and dozens of other luxury names. What it actually sells is not a better handbag — as the post says, it "doesn't sell purses that hold more, or have better organizational features. It sells exclusivity." That distinction is what makes the pricing work.

Economists call these Veblen goods: things people want more as the price rises, because the price is part of the product. A business that can raise prices without losing customers is close to the definition of a good business, and it is not something software can imitate. The second half of the case is ownership — Bernard Arnault keeps buying more of his own company, and the family now holds roughly half of it, so the people running it are exposed to the same outcome as the shareholders.

When a drug company wants to test a new medicine, it hires a contract research organisation like Medpace to run the trial: recruiting the patients, coordinating the doctors, collecting the data and steering the whole thing past the FDA. Medpace is paid for running the trial, not for whether the drug works — so it is not a bet on any one medicine.

The reason this is the post's answer to AI is that the work is physical and legally exposed. "You can't digitize a blood draw"; regulators want evidence from real bodies, not model predictions; and a badly run trial can destroy "$100 million and 10 years of work overnight." When the downside is that large, buyers do not shop on price or novelty — they pick the most trusted operator, which is exactly the position a reputation-based business wants to be in.

Brookfield owns real, physical things — hydro and wind power stations, pipelines, ports, office and industrial property — and it also manages money for pension funds and other institutions, taking fees for doing so. Both halves produce cash that arrives regardless of what happens in software.

The interesting move in this issue is to turn the AI worry upside down. AI does not run on ideas; it runs on electricity and physical logistics, and those take "billions of dollars, decades of permits, and years of construction" to create. So the more AI is built, the more valuable Brookfield's existing assets become — Microsoft and Amazon are already signing contracts with it just to secure power for their data centres. Behind it is Bruce Flatt, nicknamed "Canada's Warren Buffett," who has compounded the business at about 18% a year for twenty years with a stated goal of doubling it every five.

Topicus is the European offshoot of Constellation Software, and it does the same thing: buy small software companies whose product runs one specific type of organisation — a library, a public transit system, a specialist clinic — and keep them forever.

Investors have marked it down on the theory that AI will make such software easy to replace. The post gives three reasons that is wrong, and they are practical rather than technical. Twenty years of accumulated data and process means switching "feels like moving a mountain." The software encodes local rules a general model does not know — "AI might know the law… but it doesn't know how a small-town court in the Netherlands actually runs day to day." And the annual cost is trivial next to the damage of it failing, so nobody is shopping for a cheaper option. Meanwhile the numbers went the other way from the share price: 2025 revenue rose 20% and the free cash flow available to shareholders rose 23%.

Nothing here is a criticism of Walmart as a business. The objection is what people are paying for it. Frightened of anything AI might disrupt, investors have crowded into shops and warehouses — things that obviously still need to exist — and pushed Walmart to about 46 times next year's expected earnings.

The comparison that makes the point is that NVIDIA, the company at the centre of the AI boom, trades at about 40 times. So the "safe" choice is now the more expensive one. The underlying rule is the one the archive repeats constantly: "Even the best company in the world can be a bad investment if you pay too much." Buying safety at a high enough multiple stops being safe.

Costco is the more extreme version of the same problem. At roughly 54 times expected earnings it is the most expensive of the three companies compared in this issue — a third dearer than NVIDIA — for a membership warehouse retailer whose profits grow at a fraction of the pace.

The post's diagnosis is behavioural, not analytical: "fear is driving investors into 'AI-proof' businesses," and fear makes people follow the crowd. The consequence is that the multiple, not the business, becomes the risk. It is a useful reminder that a derating can come from a starting price rather than from anything going wrong.