Hi Multis
It’s Sunday again and it’s summer. I hope you enjoy both and that this Overview Of The Week even enhances that enjoyment.
Before we go into the rest of the content, two things.
First, you will notice that this Overview Of The Week is early this week. That’s because tonight (here) I want to watch the World Cup final.
Secondly, normally, I would have been at Chapters this week, but the evening before I was traveling to Hamburg, I decided not to go. The reason was, in the first place, my health. I had felt very tired after returning from Vail a few weeks ago and I decided to listen to my body. Secondly, I found out that there was a live stream of the Capital Market Day this year. Of course, I missed the direct contact and conversations with management, but as I had that last year, I thought I could skip a year. I’m feeling less tired now and that is more important. And you can still expect my article on my insights from the Capital Markets Day of Chapters in the upcoming week.
Articles In The Past Week
This is the third article this week.
The first article was the Best Buys Now article.
In the second article, we analyzed the Rubrik earnings.
Memes Of The Week
Maybe you have seen this picture of a 19-year-old Lionel Messi giving a bath to a baby Lamine Yamal. The two protagonists of the World Cup final did a Unicef campaign.
I saw a great parody of this picture.
Interesting Podcasts Or Books
This week, I started a new book called Endure.
The book explores the limits of human performance. It basically challenges that physical barriers are fixed. Think of pain, fatigue, heat and other limiting factors. It says that the brain is much more important than the physical limits and we can endure more than we think. I only started it and I’m very curious to read more.
The markets in the past week
I saw many X accounts blowing up a lot this week. Here are a few.
This again shows how dangerous a non-diversified, leveraged portfolio is.
But the markets didn’t do that badly, of course.
Still, the Russell 2000 dropped 0.52%, the S&P 500 1.55% and the Nasdaq 2.9%, which is quite a bit for an index drop in one week.
The Greed & Fear Index dropped further into FEAR territory.
Quick Facts
1. Kimi K3: DeepSeek 2.0
As a Belgian, the first association I have with the name K3 is this hugely popular children’s music group called, you guessed it, K3 (named after the three original members, whose names all started with a K: Karen, Kristel and Kathleen.
But this week, a different K3 got on the stage, not just in Belgium and the Netherlands, but worldwide. Kimi K3 is a big new AI model from the Chinese company Moonshot AI. It was launched on July 16, 2026. It’s intended to be fully open-source, with weights promised for July 27.
But the main shock came from the fact K3 outperformed American in some tests. Especially the Frontend Code Arena test was a shock to many.
Kimi K3 also came close to Anthropic’s Fable and OpenAI’s Sol overall.
There are several concerns about the frontier models from OpenAI and Anthropic, but it ultimately comes down to lower costs and open-source.
Here you can see that costs are much lower.
In times when AI budgets skyrocket, this may be an important factor for many companies with heavy coding work or running agents.
The open-source part means that everyone with GPUs can adapt it and run it privately, remove usage limits, and build on top of it.
Kimi K3 beats the other models on frontend coding, long coding sessions, at a thrid of the price and soon without limits (because of the open-source part). That means OpenAI and Anthropic will either have to keep dropping their prices, which will hurt them as they already lose money, or make much better models, which is also expensive.
Of course, this shocked the US tech scene, just like DeepSeek did. But it’s probably also very similar to DeepSeek that it is really good for AI. In January 2025, I wrote a full article about DeepSeek.
This is a sentence I wrote in that article: “It’s also clear that DeepSeek did something Chinese companies are good at: copying.” Read that sentence again, but replace DeepSeek with Kimi.
I have not checked it myself this time because it’s probably already fixed anyway, but you could see this everywhere on the net.
This works through distillation. Distillation starts with insights from another model by sending inputs to the teacher model and recording the outputs. Those outputs are then used to train the student model, in this case, Kimi K3.
Probably, Moonshot AI has done this through Anthropic’s API. This clearly violates the terms of service of the models, but you can’t block it.
This is clearly negative for Anthropic and OpenAI. They make a lot of money on their top models, and companies may choose to host their own models based on Kimi. But this is not so bad.
As Gavin Baker reasons, if only OpenAI and Anthropic controlled the whole AI market, they might end up buying too much of the supply. Think of areas where there is more demand than supply: chips, power, data center capacity etc. If there are more players, there’s a bigger market. That’s why Nvidia’s Jensen Huang supports open-source models. They need the same hardware, and more competition means less pricing power from customers.
On top of that, Kimi K3 is not very efficient with tokens. In other words, it uses more words/thinking than some rivals. The real cost per finished task is higher than it looks.
Don’t forget that LLMs (ChatGPT, Claude, Grok, Llama, etc.) have never been seen as the end goal but as an intermediary step on the way to AGI (Artificial General Intelligence).
American companies will now adopt to this new reality as well and get a kick in the behind to keep innovating as efficiently as possible, just like with DeepSeek.
But this kind of X post is very, very misleading.
First, you only see the spending of a few companies. Secondly, this is also about inference, the application of the AI. If you have the top model, but there’s not enough capacity to run it, your lead is completely worthless. The US companies show they lead to China, not that they are less efficient.
What does this mean for chips, memory, and power? That there will only be needed more of it. BlackRock CEO Larry Fink even thinks that there will be futures for those three classes soon. That means they are seen as a new asset class. He thinks it’s important enough to treat it as a scarce commodity essential to the world economy, like oil, and therefore it should be able to be traded and hedged. Weird times.
I also think that there are some deals to be done with the fallen angels in the last weeks. Think of Micron, for example. The stock is off 30% from its highs, but Morgan Stanley thinks the company’s gross margins will jump from a whopping 85% already to 89.5%.
It’s healthy that there is a pullback, but I think that will be it, a pullback, not the start of the ‘popping of the AI bubble’ as many claim. Of course, there’s never certainty in markets, but from following it this closely, I don’t see the signs in reality that many claim the stock prices give.
On an aside, it was also announced that Moonshot AI, the company behind Kimi, wants to list on the Hong Kong Stock Exchange soon, using the momentum of the K3 release. At the same time, it’s collecting money from the private market, by finalizing a round that values it at $30 billion. Its competitor, DeepSeek, is also rumored to be planning an IPO soon.
2. Volatility And What It Might Mean
Below:
Why momentum stocks just had their biggest monthly drop since 2008 (and what it means for the market)
Why I think the “overvalued market” talk is itself overvalued
And much, much more!
If you want to read this, see my Best Buys Now and much more, subscribe now.