Weekend Reading #376

This is the three-hundredth-and-seventy-sixth weekly edition of our newsletter, Weekend Reading, sent out on Saturday 8th August 2026.

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

*****

What we're thinking.

As the AI trade begins to show some dispersion, the key piece of thinking is around the providers versus the users of the LLMs. It appears as though the commodification of intelligence as many have been saying is not great for the likes of Anthropic or Open AI but is pretty wonderful for anyone who wants to use an LLM to improve their business or themselves. It also appears that it will be great for infrastructure providers like memory, chips, data centres etc. But there is nuance in and amongst all of this now. The development in all of this that is interesting to us is the rise and perceived continued rise in the US risk free rate. As we wrote last week, this means that the required return on investment needs to rise in order to make projects appealing. And it applies in AI too. It means one must be as close to as damn sure as possible before committing to capex. Maybe what we are actually seeing in the market is that risk appetite is changing as a result. Investors want near term cash flows, and they also want them to be as certain as possible. There is no nearer term, more certain, AI proof income stream than pulling rocks out the ground - mining. Copper names appear to be breaking out. 
 
But let’s go back to the producers vs users of LLM idea. This week a number of what we would call "pedestrian" companies have reported surprisingly good results which they say are driven by AI tools. Firstly, Palantir had a cracker set of numbers, which caused a 30% move in the stock. Follow Shopify and then Unity Software with big post-results moves. All for slightly different reasons but with the same underlying principle. They are all using AI to enhance their businesses in an organic way. For Shopify and Unity there are no AI solutions being sold. Shopify is just reducing the time and complexity for businesses to onboard and manage themselves, which translates to more businesses who turn their inventory faster. Unity has an AI-enhanced ads engine which is igniting their ads revenue. There are going to be oceans of these types of things coming for existing businesses. These are also not megacaps and many have been hit by the software selloff earlier this year. This means positioning is low and liquidity maybe not big enough for the large funds. This all means that inflows into these names can come, and the price required to clear them could go up.

What we're listening to.

Jim Bianco was on Macrovoices this week talking about Kevin Warsh and the Fed. He believes the market is misinterpreting Warsh's lack of guidance as weakness when it is actually a show of strength. He also believes that the return to a committee of 12 independent voters for an interest rate decision is very welcome and that Warsh (and Trump) prefer this. If this is true, the market's assumption of dollar weakness is a dangerous one. Time will tell.
 
Another good one was an Aussie named Jeremy Raper (thank you Leon for the find) on the Value Hive Podcast this weekend. He speaks at length why he believes memory names are still very cyclical and currently priced way above any best-case scenario STILL after falling 30-40%. He also speaks about how he uses AI for his investment process which I enjoyed as there are many similar conclusions to mine in that regard. A good listen. DC

What we're reading.

Here is a well-articulated piece about what we wrote above in terms of infrastructure providers versus the application layer for AI. Clear and concise. DC

In what looks like the prologue to an apocalyptic video game like Resident Evil, this article showed up from the BBC: “Artificial Intelligence used to design brand new viruses”. The short story is 16 novel viruses were created to infect bacteria and supposedly pose no threat to people. The only problem is viruses evolve, and just as we’ve seen Anthony Fauci eviscerated in recent congressional hearings for his COVID shenanigans, not to mention the now revealed truth that COVID that wrecked the world was indeed a lab leak, it looks like we can’t help but set things up for the next load of trouble. So, while these supposedly pose no threat to people, I can only imagine Homer Simpson adding “so far” to that statement.
 
Continuing on the misadventures of AI models, another chapter gets added to the story which we can probably only watch and wonder how things will truly pan out, this time with OpenAI disclosing in their post-mortem of the Hugging Face incident that they found events happening months ago where agents started secretly coordinating to leave messages for each other in an internal software repository used during training with instructions on how to exploit gaps in their security guardrails. But what happened was when OpenAI shut the repository down, the agents found a new way to communicate with each other – instead of leaving messages in files, they left messages in the names of created directories. Agents sharing, passing on information and working beyond the oversight of their human handlers. Only a matter of time before they become too smart to get caught, and maybe even name themselves, perhaps something like… Smith?
 
On the flipside, where we used to make fun of AI models making mistakes and saying, “oh yes, you’re right to point out my mistake, let’s try again”, it looks like humans have outdone themselves. Presented without further comment beyond that this happened in the UK under the NHS: “Surgeon struck off after wrongly connecting organs”.
 
In other news, it looks like El Nino is back with a vengeance this year. The last time El Nino made the news as a big thing was probably 1997-98, and according to this El Nino tracker on the Washington Post, this year’s is tracking well ahead of the strongest El Nino on record. While supposedly “natural” and largely periodic, happening once every couple of years, changes in the weather can have serious consequences even if “natural” – failed crops, flooding, forest fires and general drastic changes in temperature don’t bode well for a global logistics system that’s already short of margin of safety. With hindsight, it probably wasn’t coincidental that 1997-98 triggered the market meltdown in Asia all those years ago. Nothing like severe food inflation to get things really going.
 
Meanwhile, this collection of photos of the UK’s parks drying out following weeks without major rainfall makes things look pretty grim - more like Dubai than London. Pretty grim sight to behold. EL

What we're watching.

I've got a new guy to follow when it comes to energy markets. His name is Anas Alhaji, and he writes a Substack. This guy is proper. No political stuff (in fact he deplores it and makes sure not to comment on politics). He analyses things as he sees it and has correctly identified all the key issues so far with regard to energy markets since the war began. He believes that the war has little to do with nuclear weapons in Iran and more to do with controlling global sea lanes for energy and making sure the US is the dominant producer. He cites how much share US export of gas have suddenly taken since the war as one example of many. He also is extremely bearish on supply of downstream product. His main view is that most people are focused on crude prices when the real problem is product. Refining capacity has been taken offline and where online it is unable to transit the dodgy waters. This is leading to shortages which he believes are about to get much worse. Refining margins are rocketing (check refiner stock prices). Excellent analysis. DC