Joseph Carlson — These Stocks Are Going Down

Meta runs Facebook, Instagram and WhatsApp, and is one of the four big "hyperscalers" he's invested in — the handful of giants pouring hundreds of billions into building AI data centres. He owns it and stays bullish on the group.

His one objection is the rumoured plan to raise cash by issuing new shares. Issuing new shares splits the company's profits across more owners ("dilution"), which only hurts existing shareholders a little if the stock is expensive, but a lot if it's cheap. Meta trades at a much lower price relative to its profits than Google (about 19× a year's earnings vs 30×), so to raise the same $80 billion Meta would have to hand over 5–6% of the company versus Google's ~1.8%. He thinks that's a bad deal at this price and that investors — who are already unsure how much profit Meta will earn back on its AI spending — would punish the stock. He hopes Meta finds another way to fund it.

Google is the search-and-ads giant that also rents out cloud computing — another of his four AI hyperscalers, and a stock he owns. The viral "AI is just a commodity with no edge" argument was aimed squarely at Google, and he spends most of the video rebutting it.

Unlike with Meta, he's fine with Google raising money by issuing new shares: because Google's stock is expensive (priced at about 30× a year's profits), it only has to give up about 1.8% of the company to raise $80 billion — a cheap way to fund its build-out. Investors also already trust that Google earns solid returns on the money it spends, so the move shouldn't spook them.

Microsoft is the Windows/Office and Azure-cloud giant and the third of his four hyperscalers spending enormous sums on AI infrastructure. He describes himself as "fully invested and fully exposed" to this group, so Microsoft is a core holding he's comfortable owning straight through the current AI jitters.

Amazon is the online-retail and cloud-computing giant, held in his Story Fund and the centrepiece of his argument that AI can still be very profitable. His proof: Amazon Web Services (AWS) sells plain online storage — about as basic a product as exists — yet grew it past $60 billion a year while keeping roughly 30 cents of profit on every dollar of sales, far above what a basic product should earn.

How? It didn't just sell raw storage; it wrapped it in security, reliability, global access, compliance and thousands of tools that make it easy and safe to use. The plain product is cheap and copyable, but the whole package around it isn't — and that package is what lets Amazon charge premium prices. He argues AI will work the same way: the model itself may be a commodity, but the company that packages and delivers it best will still make great money.

Spotify is the music-streaming app — and he uses it as a real-world example, not a stock pick here. The point: Spotify doesn't even own the music it streams (the labels do), and the exact same songs are on Apple Music, YouTube Premium and a dozen rivals. By that logic it should have no pricing power.

Yet it earns over $3 billion a year in profit and growing cash, because what it really sells is the experience around the music — the app, the recommendations, the playlists, the ease of use. That "wrapper" is the durable advantage, which is exactly his case for why commoditised AI models can still be very profitable for whoever packages them best.

Netflix is the video-streaming service, cited as another example rather than a pick. A lot of what it shows is licensed from others (i.e. not exclusive), but people still pay because the app simply works well — easy discovery, good recommendations, always something fresh to watch. They're paying to be entertained, not for content nobody else has. Same lesson: the product can be ordinary while the service around it is the moat.

Texas Roadhouse is the steakhouse chain, used as an analogy. Steak is the ultimate commodity — anyone can grill one, no special barrier to entry — so a steakhouse "should" earn only ordinary returns (roughly 8–12 cents of profit per dollar of money invested). Yet Texas Roadhouse earns about 17.7%, close to double, because it isn't really selling steak; it's selling consistent service, atmosphere and a reliably good night out that's hard to copy. His takeaway: an ordinary product can still support an extraordinary business — so "AI models are interchangeable" doesn't mean there's no money to be made.

Costco is the membership warehouse retailer, used to make a different point. It keeps barely 3 cents of profit per dollar of sales and spends heavily to build pricey new warehouses — by the textbook, a "low-quality," capital-hungry business. Yet investors happily pay one of the highest prices in the market for it (about 45–50× a year's earnings) because it's so consistent and reliable.

That's his counter to the worry that the hyperscalers are turning into heavy-spending, lower-margin businesses: heavy spending and thin margins don't automatically make a company a bad investment if it executes consistently. He's optimistic the AI giants' scale and infrastructure will likewise keep paying off.

Oracle is a big enterprise-software and cloud-infrastructure company that has become central to the AI story. It reports results this week, and he says strong headline numbers are basically guaranteed — so they won't be enough. The real test is its forecast ("guidance"): Oracle needs to raise its outlook. If it merely repeats the same forecast as last time, the market will treat it as one more AI name whose growth is slowing, which is exactly what spooked investors about Broadcom.

Adobe makes Photoshop and the rest of the creative-software suite. On paper it's one of the cheapest high-quality companies around — priced at only about 10× a year's earnings and throwing off cash equal to roughly 10% of its stock-market value — but it's cheap because investors fear AI image and video tools will eat its lunch.

It reports this week, and he says the numbers will look fine; the hard part is convincing investors that AI isn't hollowing out the business by showing real growth in paying customers and products. He calls it "a hard story to tell" with the market betting against them, so he stays on the sidelines.

ASML is the Dutch company that makes the ultra-advanced machines (EUV lithography) used to print the world's most cutting-edge computer chips — and it's essentially the only company on earth that can. He brings it up as the opposite of a commodity: there's no "top 10 list" of these machines, because if there were, all ten would be ASML's. He uses it to fairly concede the bears' point — AI models, unlike ASML's machines, are many and interchangeable — before arguing that being interchangeable still doesn't mean no profits.

Uber is the ride-hailing and delivery app, mentioned only as evidence that AI is suddenly getting expensive. Its CEO said the company burned through its entire annual AI budget in a single quarter and is now capping how much each engineer can spend on AI. Companies being cautious about AI bills was "unheard of a month ago," and that abrupt about-face is part of what rattled the market.