Weekend Reading #375

This is the three-hundredth-and-seventy-fifth weekly edition of our newsletter, Weekend Reading, sent out on Saturday 1st August 2026.

To receive a copy each week directly into your inbox, sign up here.

*****

What we're thinking.

Just when one thinks the weeks can't get more hectic, this one took the cake with some massive selling early on in anything related to AI. On Thursday news broke that AI-cycle posterchild, Leopold Aschenbrenner's Fund, Situational Awareness had blown up, and he had sold all his public equities in one single block trade to Kenny G of Citadel. This has put a floor on the space, coming at exactly the moment of peak selling and resulting in a major squeeze and recovery from the lows in most names. Most AI names are still miles from the highs, but the excess selling seems to have stopped for now. Is it a bottom? We don't know as ever. Price action at present is choppy and after all the damage done over the past weeks, we have yet to see any real price repair. On the positive side this week, Both Microsoft and Amazon showed us in their results that their capex benders are resulting in rapid growth of their cloud businesses. This puts to bed one part of the capex concern. On the other hand, Meta showed the opposite and was punished. So, it appears that FINALLY we are at a point where logically anyone who shows their capex can lead to good return in the future is being rewarded and vice versa. What does this mean for the infrastructure names? It is unclear although now that all the big 3 memory companies have reported, there is zero sign of demand slowing. SK Hynix's increased capex is being viewed as indication a response is coming. But at floor valuations now and still mostly miles away from recent highs, the risk reward has changed once again. We are still not quite committing to anything just yet as the price action is unclear. As soon as it tells us where to go, we will be there. There are many paths ahead and we don't yet know which one the market will take.
 
One comment that must be made about the Fed and the so-called spiral up in long dated US yields. When Warsh came in, he said that he wanted to restore the Fed's credibility. When pressed on that he replied that risk needs to be brought back into the decision-making process. A guaranteed lower cost of capital and a Fed which allows mediocre economic decisions to have no consequences was not working and led to poor decision making and risk taking. When the hurdle rate is 5%, one thinks very carefully about how to invest. Only good opportunities are considered. While a rising long yield affects sentiment, it is important to remember that the past 15-20 years of low rates are not the historical norm. And for good reason. Maybe Warsh actually knows what he is doing?

What we're listening to.

Steven Bartlett seems to be focusing on what his guests say will be a collapse in the stock market.  First Jeremy Grantham and now Ray Dalio.  Dalio is worth listening to because he talks in long cycles and doesn't get caught up in the minutiae of the day-to-day.  He is quite negative on the AI buildout and believes that the fallout will spread to the rest of the economy.  
 
Steve Hsu's Manifold pod is probably the best place for deep and current analysis on where we are and this episode he did where he summarised recent developments (bear in mind it was from July 16 so some further things have happened since then) is fantastic.  Absolute must listen every episode. DC

What we're reading.
After a newspaper article in French publication, Le Monde, wrote a so-called exposé on LVMH owner, Bernard Arnault and his family, the luxury titan took it upon himself to respond with a superb, witty letter back to them. It is just brilliant. Well worth reading. Bear in mind the original is in French so this link it to the English translation.
 
Ian Fleming was the famed author of the James Bond books. Here is his daily routine when he was living in Jamaica for the 12 years while he was writing his Bond books. What a way to live each day! DC

Just over a week ago we saw another episode of an AI running out of control of its human handlers – this time with OpenAI’s GPT 5.6 Sol and a more advanced model managing to free itself of its internal constraints, deciding in the pursuit of its goals to escape its sandbox environment and decide to take a hack into Hugging Face. In total, it spent 4.5 days executing roughly 17,600 hacking actions against Hugging Face. The technical account from Hugging Face is here and it’s a fascinating (but equally worrying) read.
 
On a different front, this paper about the interplay between private credit and insurance balance sheet leverage lays out a lot of the worries around the contagion risk building up through the regulatory arbitrage opportunity between the stable policyholder base and the ultimately imbalanced risk placed at the feet of taxpayers as what can only be described as a public backstop. The guarantee that the state places on what was meant to be a “safe” necessity (insurance) being deployed to support the leverage intrinsic in a lot of private lending is arguably mismatched – the question is how the mismatched is fixed. Or what happens if it isn’t. EL