This week’s reading list is about crowding, and how seldom anyone actually decides to join a crowd. Wall Street’s largest firms have ended up at the same few tables funding the same buildout, not because they agree but because the career cost of being the one who stayed out is so much higher than the cost of being wrong alongside everyone else. J.P. Morgan has now put a number on where that leaves the rest of us, and it is not comfortable: AI related industries are roughly half of the S&P 500 by market capitalisation, which means a passive equity allocation is a concentrated position whether or not anyone chose one. Bond index investors have been accumulating the same exposure by the same accident. Insurers are retreating from the same postcodes at the same time, and the political reflex is to cap the price rather than read it. A household holding three years of expenses in cash has joined a crowd too, and is paying for the privilege. The counterpoint sits three thousand years back, in the moment a scarce metal was replaced by an abundant one and every power that had controlled the scarce version discovered what its advantage really rested on. The useful lesson for investors is that a crowded position rarely announces itself, because each step into it looked reasonable on its own. The best investors are not trying to be contrarian for its own sake. They are trying to notice which of their positions they actually chose.

📚 AI turns Wall Street into a herd: Liz Hoffman on the strangest feature of the AI financing boom, which is how little disagreement there is inside it. Nvidia’s US$500 billion consortium includes the four largest alternative asset managers and six of the top eight. Blackstone and Apollo are jointly funding Anthropic’s compute purchases, and nearly every large private equity firm outside that group is backing OpenAI’s competing version, with Goldman Sachs somehow in both. Brookfield’s Connor Teskey explains why rivals happily share a table: there is enough to go around. Hoffman reaches back to a 2008 postmortem for the phrase that describes the real incentive, “I’ll be gone, you’ll be gone”, and is candid that her own early 2024 call on contrarianism starting to pay has not aged well. A PwC estimate of more than US$30 trillion of global data centre spending through 2050 sits underneath all of it. (Semafor)

📚 AI-conscious asset allocation: Portfolio construction in a capital investment boom: Michael Cembalest and J.P. Morgan’s Strategic Investment Advisory Group with the most useful institutional paper we have read this year, and the one that turns the AI question from stock picking into asset allocation. Their central claim is that AI has become a cross asset risk factor inside portfolios that were never built with it in mind. AI related industries are about 50.6 per cent of S&P 500 market capitalisation, so passive US equity is a concentrated AI position by default, and the same factor then reappears in venture and growth equity, in private credit lending to data centres, in value add real estate and in investment grade bond indices. Apparent diversification is therefore partly an illusion. They map an AI intensity spectrum from most concentrated to most defensive, note that core real estate holds under 0.5 per cent data centre exposure in the NFI-ODCE index, and argue high quality fixed income remains the real protection against a shock to the AI narrative, while honestly flagging that inflation uncertainty could undermine that too. The line to keep is that allocators must consider both diversification within AI and diversification from AI. Note this links to the full PDF; the landing page shows only a summary. (J.P. Morgan Asset Management)

📚 AI Debt Issuance Taking Up More Room in Bond Indexes: Debbie Carlson on the same problem at the level a retiree can actually see on a statement. Alphabet, Amazon, Meta, Microsoft and Oracle together issued around US$35 billion of debt a year between 2020 and 2024, on Vanguard’s numbers. Industry estimates for AI related issuance in 2026, once chipmakers and utilities are included, run between US$300 billion and US$570 billion. Amazon is now the fifth largest issuer in the Bloomberg US Corporate Index at 1.4 per cent, after a US$53.8 billion dollar and euro sale, with Oracle eighth at 1.3 per cent following a US$30 billion bond and preferred deal. Nobody holding a global credit index fund decided to increase their exposure to the AI capital cycle. It happened because the issuers kept issuing and the index kept weighting. (Chief Investment Officer)

📚 How the insurance industry can help society adapt to climate change: David Gunn and Rob Bailey at Chatham House on why uninsurability is a policy failure rather than an insurer failure, and why that distinction is about to matter commercially. Their framing is the useful part: insurers are the messengers, and a rising premium is among the first honest signals that a risk has become intolerable. In California, political caps on premiums contributed to insurers withdrawing alongside the physical risk itself. They cite a UK Government Office for Science estimate of £3.5 trillion of cumulative global adaptation spending over the next decade, and warn that if blame lands on insurers the result will be tighter price controls and pressure to write cover at a loss. For Australian readers the application barely needs spelling out, across both residential property and the banks holding the mortgages against it. (Chatham House)

📚 Questions for the Chair: Claudia Sahm on the September meeting, where the Fed raised its policy rate a quarter point to 3.75 to 4 per cent on a unanimous vote, six weeks after only three of twelve voters had favoured a hike. The projections centre on one more rise in 2026 and do not reach 2 per cent inflation until 2029, with the median longer run neutral rate at 3.25 per cent, already below the funds rate before the increase. The dot plot implies a tightening cycle of 50 to 75 basis points against a median of 3 percentage points in past cycles. Her five open questions are the substance, and the sharpest is how Kevin Warsh will know when to stop, given he has filed no projections of his own and offers no forward guidance. She also notes the press conference ran about 30 minutes against roughly 50 for Powell’s last five, and that implied rates and the 10 year yield rose while it was happening. (Stay-At-Home Macro)

📚 Has AI impacted the labor market yet? Alex Imas and Jacob Schaal with the most even handed answer available to the question clients keep asking, and a model of how to weigh evidence that disagrees with itself. On one side, Brynjolfsson and colleagues found junior employment in high exposure roles down 13 per cent by July 2025, with their dashboard showing 19 per cent by June 2026, and a study of 6.7 million graduates found those in the most exposed majors 5 percentage points less likely to be employed on graduating. On the other, a Danish study finds null effects on earnings, hours and wages with confidence intervals ruling out anything above 2 per cent, Gallup finds only 1 per cent of laid off workers attribute their firing to AI, and around 90 per cent of executives report no significant effect. Google’s own data shows AI being used across 88 per cent of US employment but in only 21 per cent of tasks. Their conclusion is that aggregate stability and real stress in exposed subgroups can be true at the same time. (Ghosts of Electricity)

📚 The High Price of Feeling Safe: Vishal Khandelwal on a rule of thumb doing real damage, the advice to hold three years of living expenses in an emergency fund. He works the arithmetic, then makes the distinction that matters: preparation says something may go wrong and I should be able to handle it, while fear says I should arrange my whole life around that possibility. A young couple with two incomes and portable skills and a 55 year old sole earner with dependants and a large loan need genuinely different answers, and a single number serves neither. He borrows the Stoic practice of picturing the worst case deliberately in advance rather than carrying it around permanently. Written for an Indian audience, but the behaviour is universal, and it is one of the more common and least discussed drags on a self managed portfolio. (Safal Niveshak)

📚 Weapons of mass decentralization: Patrick Fitzsimmons on what happened when the cost of the critical input collapsed. Bronze required tin, and tin was scarce enough that Mari’s supply travelled more than 500 miles from Elam; the powers that could organise that trade fielded armies like the tens of thousands and 5,000 chariots that met at Kadesh in 1274 BC. Iron melts at 1,538°C, which Bronze Age furnaces could not reach, but bloomery furnaces at 1,200 to 1,300°C could work it without melting it, and iron is the fourth most abundant element in the crust. The measurable consequence is startling: across Assyria, Babylon and Athens, average reign lengths fell 29 per cent after iron was adopted, with Babylon getting almost twice as many kings. The closing line about drones accounting for up to 80 per cent of recent casualties in Ukraine is not subtle, and does not need to be. (Works in Progress)

📚 A good sceptic has more than doubt and less than certainty: Massimo Pigliucci on Cicero, who wrote the Academica in early 45 BCE at a villa near Naples, having just divorced after thirty years and lost his daughter, with Caesar newly dictator for life and two months from assassination. The idea worth taking away is Carneades’ pithanon, which Cicero rendered as probabile: not probability in the modern sense but graded confidence, running from an impression slightly more likely true than not, up through claims examined from several angles, with no step ever reaching certainty. It is a 2,000 year old description of exactly the habit of mind that investing requires and that crowds reliably destroy. Markets are never mentioned. We include it anyway. (Aeon)