My basic view is that AI should become infrastructure, not a private toll road.
The models, the tokens, and the services built around them should eventually work more like electricity, water, or the internet: easy to access, easy to distribute, and cheap enough for ordinary users and companies to build on top of. Today’s data centers are not really built for that world yet.
That is why I don’t think the so-called “token apocalypse” is necessarily bad. If AI is going to become national-scale infrastructure, access to compute and models has to become much simpler. In the long run, only open models that are cheap and easy to deploy can support that kind of system.
This is also why I find Alex Karp’s position interesting. I once discussed him with friends who follow U.S. politics closely. He clearly does not like the idea of autonomous AI taking over human analysis and decision-making. And that makes sense: Palantir’s business model depends on keeping humans in the loop, not replacing them with a model. So even if Karp looks like he ate a dozen bars of Safeguard soap, his argument is at least consistent with the company he built.
My problem with Anthropic is different. I don’t dislike the company because it banned my account. I dislike it because its vision feels like trying to monopolize the world’s freshwater supply. If AI is going to become basic infrastructure, that is not just arrogant. It is incredibly stupid.
Thanks for your in-depth comment. There's certainly seems to be a lot of consolidation that's coming and China's stack is a major contribution it turns out. The timing of all this feels fairly premeditated.
Even without China, someone would eventually provide open models as an alternative. Otherwise, we are left with only two possibilities: either AI is too hard to make broadly useful, or AI companies are pricing it in a way that prevents broad adoption. I believe the second explanation is much more likely.
My basic view is that AI should become infrastructure, not a private toll road.
The models, the tokens, and the services built around them should eventually work more like electricity, water, or the internet: easy to access, easy to distribute, and cheap enough for ordinary users and companies to build on top of. Today’s data centers are not really built for that world yet.
That is why I don’t think the so-called “token apocalypse” is necessarily bad. If AI is going to become national-scale infrastructure, access to compute and models has to become much simpler. In the long run, only open models that are cheap and easy to deploy can support that kind of system.
This is also why I find Alex Karp’s position interesting. I once discussed him with friends who follow U.S. politics closely. He clearly does not like the idea of autonomous AI taking over human analysis and decision-making. And that makes sense: Palantir’s business model depends on keeping humans in the loop, not replacing them with a model. So even if Karp looks like he ate a dozen bars of Safeguard soap, his argument is at least consistent with the company he built.
My problem with Anthropic is different. I don’t dislike the company because it banned my account. I dislike it because its vision feels like trying to monopolize the world’s freshwater supply. If AI is going to become basic infrastructure, that is not just arrogant. It is incredibly stupid.
Thanks for your in-depth comment. There's certainly seems to be a lot of consolidation that's coming and China's stack is a major contribution it turns out. The timing of all this feels fairly premeditated.
Even without China, someone would eventually provide open models as an alternative. Otherwise, we are left with only two possibilities: either AI is too hard to make broadly useful, or AI companies are pricing it in a way that prevents broad adoption. I believe the second explanation is much more likely.
"This is leading to a big transfer from closed-source models to open-source models that will make some of China’s AI startups way more profitable"
Using open source models will not, however, improve the accuracy and reliability of the outputs.
I didn't think we'd see this so soon. When Microsoft embeds DeepSeek into their AI strategy, you have to wonder? https://www.axios.com/2026/06/16/microsoft-copilot-cowork-tokenmaxxing-cowork
I thought these people viewed China as the enemy. So I'm more than a little confused.
Karp can hardly hold a thought before he's on to the next one. His idea of a P&L sheet is a little insane as well.
Yes what is that indicative of?
The manic behaviour or the P&L?
The Era of FDE with Open-weight models, this is getting so crazy.
https://en.wikipedia.org/wiki/Forward_Deployed_Engineer
https://x.com/ReadFuturist/status/2073807160958705731
I have a friend who is right into AI. Cannot see a negative side at all.
Karp “what you make less what you lose”
So I guess that is what capital can you pull in less what you actually are losing. The AI version of net profit…
Karp’s arguments are all the more hilarious when you consider how much Palantir is growing relative to Anthropic.
If Anthropic is going to be valued when it goes public in October like Palantir, it would have something like a 6 trillion dollar valuation.
We are just in the early days of this technology enabling any serious work.
If it ever does. I am finding it difficult to get value out of them and just this morning thought how much they helped to fuck up my 2025 for me.