The Minotaur Global Opportunities Fund returned -1.8% in October, 0.9ppts ahead of the MSCI ACWI (Net, AUD). It was a falling month for markets, and an uneven one. Communication Services fell 13% in the index and semiconductors fell hardest of all, as investors sold the companies building AI on the worry that the spending is running ahead of the returns. Artrya was also one of our largest detractors, after its quarterly update withdrew the target date for FDA clearance of its Flow product, though we are still confident this is just a delay. The Australian dollar also rose against the US dollar, the euro and the yen, and that currency move accounts for more than half of our fall.

Our AI-build holdings (Meta, SK hynix, NVIDIA, Kokusai, Marvell and Broadcom) cost us about 3ppts between them. Crocs, Eli Lilly and Freee won almost 2ppts of that back, and the data owners we wrote about last month recovered what September’s rate shock had taken.

The best evidence yet that AI can pay for itself arrived in the same week the market sold the builders hardest.

Anthropic: someone is making money from AI

For three years the bear case on AI spending has been that nobody downstream makes any money. In August we asked who is actually earning a return on the build-out, and concluded that the chipmakers clearly were, the data centres were earning roughly their cost of capital, and the missing piece was the labs paying for it all. On 26 October Anthropic filed its prospectus, the first audited look inside a frontier lab, and filled in that piece.

We wrote about the filing in detail in our Livewire column. Revenue was US$4.6 billion for all of 2025 and US$11.6 billion in the June 2026 quarter alone, which was also its first quarter of operating profit. It has raised US$141 billion and still held US$132 billion at the end of June. Each dollar of compute now earns about two dollars of revenue, up from about 40 cents six quarters ago. The revenue is outrunning the compute bill.

The filing puts Broadcom at the centre of Anthropic’s next wave of capacity: Anthropic has committed to about US$161 billion of non-cancellable rent on Broadcom-built TPU systems, the largest single compute commitment in the prospectus, and Broadcom has made available up to US$42 billion of loans that can only be used to pay that rent. Anthropic is also the end user of Hut 8’s River Bend campus in Louisiana, about a quarter of Hut 8’s contracted data-centre rent, and press reports link the filing’s US$35 billion agreement with Lambda to Hut 8’s larger Texas campus. We added to Broadcom and opened a small position in Hut 8. NVIDIA, SK hynix and Kokusai are the other holdings whose customers have just shown they can pay.

The filing answers the demand question and opens a financing one. Anthropic’s own suppliers lend it money and also sell it much of what it buys, and October’s sell-off was the market pricing that question across the whole sector: Meta announced a US$97.5 billion capital raise, Alphabet’s free cash flow was negative for a second quarter, and CoreWeave’s credit spreads widened. We think the demand is real. What worries us now is the financing, and how long the returns take to arrive, and we’re doing the work to make sure we’re ok with this and where the builders are going.

AI’s winners and losers in advertising

If Anthropic showed that AI can make money, October’s advertising results showed who else is making it, and who is paying for it.

Meta’s advertising revenue in the September quarter was US$63.1 billion. Google Search earned US$63.4 billion. Three years ago Meta’s advertising was 70% the size of Google Search and now it is effectively the same size and growing twice as fast (26% against 12%).

Meta's advertising has caught Google Search

Quarterly revenue, US$ billions

Source: Company earnings releases; Minotaur Capital analysis.

The reason is AI. Meta sells attention inside its own apps, so better AI means better-targeted ads on space it already owns. The number of ads Meta showed rose 10% in the quarter but the price advertisers paid for each one rose 14%, which is what happens when ads work better. On its call Meta pointed to specific gains: its new ranking model lifted app-install conversions by 9.6%, and advertisers who followed its AI assistant’s suggestions got almost 5% more conversions at the same cost per conversion.

Google is on the other side of the same shift. Search growth slowed to 12%, from 17% in June and 19% in March. The price of each click held up, but the number of paid clicks grew 8%, down from 13%, and Google’s chief business officer put it plainly: “the strong growth in AI Mode queries had an impact because of the lower monetization we were seeing in AI Mode.” Google’s response is fair: query growth is the fastest in five years, most AI Mode questions are new rather than replacements, and it has only just started putting ads in AI answers. But in our September quarterly we wrote that an agent doing the searching “skips the results page where Google is paid”, and that Google has the most to lose. This is the first quarter where Google’s own numbers show the cost.

The sites that lived on Google’s traffic are feeling it more. Reddit said search referrals had fallen by more than a million a week in October and that daily users could fall by a low-double-digit percentage this quarter. Its CEO was blunt: “AI Mode does not drive traffic… It used to be a traffic driver, now it’s not.” The shares fell 29% the next day.

Meta’s share price didn’t follow its advertising. The shares fell this month, making Meta our largest detractor, on the cost of the build-out and the size of the capital raise rather than anything in the ads business, which had its best stretch in years. Muse, the consumer agent we wrote about in September, passed 10 million set-ups in its first six weeks. We made the long case for Meta in the September quarterly, and October’s results added evidence to it and we are excited about our Meta position.

The data owners came back

The other AI winners were the companies that own the data AI needs. Thomson Reuters, RELX, S&P Global and MSCI all rose between 16% and 21% and added about 0.7ppts, after costing us 0.3ppts in September’s rate shock. The evidence came with the rally. RELX said adoption of its AI legal platform had passed 50% in the US, S&P Global raised its revenue and adjusted earnings guidance for the year, and Moody’s said use of its research inside customers’ AI tools rose more than tenfold in three months. An AI answer is only as good as its source, and customers are paying the people who own the good ones.

Crocs and Lilly: the portfolio is more than AI

Our two best contributors this month had nothing to do with AI, and between them added about 1.4ppts.

Crocs rose 28% this month, almost all of it on 29 October, the day of its result. Earnings per share came in at US$3.69, against guidance of US$3.20–3.30 in July. HEYDUDE, the brand that had been dragging on results, grew for the first time in seven quarters, and management raised its outlook for the year.

Eli Lilly reported the same day and rose 7% on the result, and 23% over the month. Revenue grew 39% and it raised its full-year guidance by US$5.5 billion at the midpoint. Foundayo, its new GLP-1 pill, sold US$184 million in its second quarter on the market, nearly double the first, and Lilly says about four in ten patients starting an oral obesity pill now choose it.

We didn’t buy either as a hedge against AI but part of the appeal of these companies is the diversification they bring. We own them because we thought the market was underestimating them, and October’s results said the same. In a month when the AI trade fell hard, it helped that neither depends on it.

What is profitable is wealth

Most of October came back to who pays, and for what. Anthropic’s customers are paying for intelligence, enough for its revenue to outrun its compute bill. Meta’s advertisers pay 14% more for each ad because the ads work better, while a search on Google is worth less once an AI answer sits at the top, and Reddit’s traffic dries up when Google stops sending it.

More than 2,300 years ago, Xenophon recorded Socrates arguing in Oeconomicus that something is wealth only if it benefits the person who owns it. His example was a flute, which “may be wealth to him who is sufficiently skilled to play upon it, but the same instrument is no better than the stones we tread under our feet to him who is not so skilled.”

The market spent October questioning the economics of AI. Anthropic’s customers and Meta’s advertisers have shown that the demand is real, but demand alone doesn’t guarantee a return. The question now is who knows how to play the flute: who captures the value, and whether those spending hundreds of billions to build it will earn enough to justify the investment. AI’s value is becoming clearer but who gets to keep it, not so much.