Kevin hitting it with a 44 minute dunk on Murican AI
The economic model for the AI industry brought to us by Wall Street and Silicon Valley is falling apart, with subscription fees paid by users which are far below the companies’ cost of compute.
The companies also face severe blowback for new data center construction almost everywhere, and constraints on power grids and capital budgets have delayed dozens of projects.
Top managers of Silicon Valley AI companies are also, finally, facing harsh public scrutiny, particularly after the disastrous opening days of the War on Iran.
The Artificial Intelligence industry is comprised of five layers: energy, chips, infrastructure, models, and applications.
Now that Chinese Large Language Models are optimized to run on Chinese-built chips, are are soaring in popularity across the world by enterprise users, even top US industry insiders admit that Chinese tech will dominate going forward.
Haven’t watched the vid so maybe I will repeat some points.
Consider that Deepseek’s main business is trading. They made deepseek as a side project because they had spare compute power. Don’t forget that in China deepseek’s main company is a pretty major trading company - it’s one of the big four of hedge funds in China.
This means that deepseek, and others such as alibaba (qwen family of models and so on), have no funding problems. They have a business that sustains their venture into new unstable tech, basically. Whereas in the west, companies leading AI were founded specifically for AI - OpenAI and Anthropic (Claude).
It’s very very different and much more sustainable. In the west the big players that have other sources of funding can’t seem to make interesting models. Gemini (Google) is not even talked about anymore, even if their Pro model, when it came out, was pretty impressive. IBM makes the Granite family of open source models made for local inference - not competitive whatsoever. Meta is completely out of the game at this point, LLama was good in its time to fine-tune but now everyone is using Chinese models for that.
In the long run they will survive and persevere, and there will come a tipping point eventually where quantitative change leads to qualitative change (or vice versa). They also understand in China that AI is a bubble - but bubble doesn’t mean “once it explodes it goes away”, don’t make that mistake. The internet was also a bubble initially and yet here we are talking to each other online. It just means it’s completely overvalued, and the 20$/month membership tiers on GPT, Gemini or Claude (Claude is actually much more expensive on API lol) are not covering expenses, not at all. They barely make a dent in fact - these companies are not sustainable and are only being handed out pile after pile of cash because the US Government believes it to be almost existential at this point. They wouldn’t be entirely wrong, seeing how China is just running laps around them in every other area.
But deepseek doesn’t need to be profitable, because the parent company is already profitable and sustainable, so they get to offer deepseek 100% for free and with an API so cheap, you now have people switching over from Claude because there’s no rate limits and 1$ gets you a full suite of scripts coded.
As time goes my essay becomes more and more vindicated lol, I keep making additions to it every time I get proven right about the assertions I made back in November.