The current AI boom and AI buildout are both deeply fascinating and causing headaches at the same time. In this article, I will explain some trends that I am seeing and topics that you should be aware of. Among these are rising inference costs, exploding energy costs, Chinese AI models, and geopolitical restrictions, which have a major impact on the world.
About the title image
AI eats the world. But what does AI actually look like? For this image, I have chosen a robot, the next logical step: Physical AI
You will find the first part here:
AI Observations – June 2026 (Part 1): The Boom Continues
This is a follow-up to my AI Eats the World post from November. Roughly half a year later, AI is still the dominant topic of the day, and the new models from Anthropic and OpenAI were another major jump. This is part 1, covering the newest slide deck from Ben Evans and some more interesting observations. Part 2 will move beyond the data, and I will disc…
What does an AI center look like? This short video gives a great overview and also thematizes some of the challenges.
Apart from Nvidia, AMD is the other large player in GPUs. If you invested in AMD in December 2024, when I published my deep dive, you are up +344%. Congrats!
In case you wonder if AI has claimed any victims, Stack Overflow is one of them. The number of questions asked has dropped to almost zero and the event that caused the decline was the release of ChatGPT, in other words. ChatGPT was the asteroid that erased the dinosaurs.
While some people say that Adobe will be gone soon as well, I don’t think so. Learn more about Adobe’s most recent earnings here:
Tokens getting expensive
The cost of AI will continue to rise. Token pricing is one of the major challenges going forward.
Dara Khosrowshahi, CEO of Uber, said the following in an interview with Patrick O’Shaughnessy:
“We blew through our AI budget in a quarter, for the whole year. It is forcing us to adjust.
We are going to meter headcount increases because to the extent that my engineers are getting much more efficient, their throughput is increasing. There’s a cost to that, and it’s a significant cost.
AI adoption has been occurring in all parts of the business –– whether it’s engineers and how they scope projects, how they build, debugging, platform migrations.
I’m pushing the teams to fundamentally use the power of AI to rebuild systems and processes from the bottoms up.
I do think it’s a combination for us right now of encouraging adoption, but then driving efficiency.
We’re using the more expensive models to explore. Once we scale some of these experiences, we’ll look to bring in more efficient models that are more efficient on a token basis or are open source.”
Microsoft is another example. The company with almost infinite cloud computing resources said that the token-based billing made it unjustifiable to run it. At the same time, GutGzv Copilot, which is owned by Microsoft, announced that it will stop providing its flat-rate plans and switch to usage-based billing.
All these signs point in the direction that AI was heavily subsidized by all the VC money, similar to what happened with Uber’s pricing, once the company was public and had to show profits.
Unlike traditional SaaS, adding another user did not cost the company, say Salesforce, any (or close to 0) money. With AI, it is different. Training cost is one thing, but inference, the act of drawing conclusions based on the learned material, does have real economic costs. The question is: How long can OpenAI and Anthropic subsidize this effect, and how long are they willing to do so?
This post by Brian Armstrong, co-founder and CEO of Coinbase, summarizes the points above nicely. You don’t need the most advanced models for every single task. Most of the time, the previous models will do just fine and save you quite some money.
Invest at your own risk; this is not financial advice! This is not a recommendation to buy or sell any securities discussed in the article.
Energy consumption keeps rising
There are reasons for the high prices of tokes. It is not just the expensive hardware (GPU, CPU, RAM, etc) but also the energy that goes into the data centers. Google’s data centers consumed more than 42 million megwatt-hours of electricity in 2025. That is 37% more than in 2024. These are large numbers, but let me give you some perspective. That is more than the entire energy consumption of New Zealand or Denmark! We are talking about whole countries.
Distribution still matters
Distribution is still often overlooked as a crucial factor for successful products. Microsoft Teams dominates the meeting software market because it comes with Windows. That is why it is being used so much more than Slack, even though Slack is by far the better product.
This guy learned the lesson as well. While AI made it so much easier to build something, you still need to find your audience, and your audience must find you. Without any real go-to-market strategy or sales, you are just one of the millions of little fish in a giant ocean. That is why the large incumbents will benefit from AI: They already know their customers and can now ship cheaper (if they control their token spend, see above) and faster than ever before. This is one of the reasons why I believe Salesforce is very attractive these days. You will find my deep dive here:
Salesforce, the unknown agentic AI player
Welcome to the newest deep dive. I am writing about fantastic companies that are trading at a cheap/fair valuation given the quality of the underlying business. Before I look at the price, I make sure that I have a thorough understanding of the company behind the stock, and I share my research with you.
New frontiers in medicine
AI is shaping the world in ways we haven’t seen yet. While the negatives are often making the headlines, there are also very positive developments in terms of medicine. Midjourney, the company behind the image generation tool which I also use for the title images of my Substack, surprised everyone with this news:
This is what Afshine Emrani, MD, FACC, a cardiologist, wrote about this new development on X:
You step into a shallow pool of water. You stand on a platform that slowly descends — about two inches per second — through a ring containing roughly half a million tiny ultrasonic transducers, each the size of a grain of sand. Every one of them acts as both a speaker and a microphone, sending ultrasonic waves through your body from every angle and recording what comes back.
60 seconds later, you step out. The scan is done.
No radiation. No magnets. No claustrophobia. No IV contrast. Just sound, water, and an almost incomprehensible amount of computing power — roughly 2 petaflops processing 17 gigabytes per second of raw acoustic data — reconstructing a 3D map of your entire internal anatomy down to half a millimeter resolution.
Organs. Tissues. Blood vessels. Bones. Muscle. Fat distribution. All segmented by AI in real time.And if it is validated — if the resolution holds up against MRI, if the AI segmentation proves reliable, if the regulatory path clears — then what we’re looking at is the most significant new imaging modality in 50 years.
For my entire career, preventive cardiology has been limited by the fact that seeing inside the body is expensive, slow, uncomfortable, and infrequent. We catch disease late because we image rarely. We image rarely because imaging is hard.
A 60-second, no-radiation, spa-based full-body scan that costs a few dollars would demolish every one of those barriers.
Whether this works or not remains to be seen. The fact that we are entering a golden age of medicine, however, cannot be disputed. So stay healthy, my friends!
The boom of South Korea
South Korea is a fascinating country. While Taiwan (green) is the well-known home of Taiwan Semiconductor, the close neighbour South Korea (orange) is also home to some well-known semiconductor companies. Both Samsung and SK Hynix, two of the top three DRAM companies, are located in South Korea.
The South Korean stock index has been parabolic since 2024.
Samsung’s Device Solutions division estimates that its 2026 operating profit will exceed the cumulative profit of its past 40 years. This massive inflöux to earnings has led to some very interesting second and third order effects:
Samsung workers threatened to strike and halt production during these precious times. To avoid the strike, Samsung pays out a record bonus of $370,000 per person this year alone. Every SK Hynix employee will receive about $477,000 this year and an expected $900,000 next year.
These sums are gigantic. But they get even more interesting. The average memory-chip worker only earns about $51,000 per year. Imagine you are getting a bonus that is more than 7x your normal salary.
And of course, this second-order effect of the boom is causing some more interesting developments: The store Shinsegae South City reported that luxury jewelry sales were up 150% and luxury watch sales +85%. I have also read that some people are wearing SK Hynix-branded polos on dates to improve their success rate. The world is a fascinating place.
The rise of China
The remaining sections examine China’s rapid AI progress, the next phase of AI adoption, AI’s role in geopolitics, and one prediction about the future of the internet that is becoming increasingly difficult to dismiss.
Become a paid subscriber to read the full analysis.
To all existing subscribers: Thank you for your continuous support!
