[https://arena.ai/leaderboard/code/webdev](https://arena.ai/leaderboard/code/webdev)
[https://www.kimi.com/blog/kimi-k3](https://www.kimi.com/blog/kimi-k3)
With Chinese firm, Moonshot AI, releasing open weight (which means freely downloadable to run yourself locally on a server) of Kimi K3 which is now better than Fable 5 for many coding and is top 2 for most things, this pales even to the first Deepseek moment.
Chinese open weights are not even months behind anymore. It's days behind if not beginning to get ahead of leading proprietary models from OpenAI and Anthropic.
Won't this crash the AI LLM thesis? Tokens are 30% the cost of Fable 5. The models themselves are freely downloadable. The cutting edge LLM model is basically free.
How will cloud services justify themselves to keep buying cutting edge hardwares at losses when the available software is essentially free while being the leading model?
The justification for frontier AI labs in US was that Chinese AI labs were months behind and Chinese AI labs can never be ahead because they can only distill and perform 90% of the US models afterwards.
Well how are investors supposed to price in this now? The world in which freely downloadable on the Internet models from Chinese AI labs are potentially the leaders.
What's also the justification for big tech to throw endless money to a pit for GPUs, etc? If the frontier AI Labs model are going to compete against practically free (Anthropic and OpenAI are almost a trillion in valuation while Moonshot AI which created Kimi is only 20\~30 billion) and the token costs are fractions of fractions of what is already an endless money losing value.... then isn't this a whole valuation risk to all the insane valuation of recent AI related stocks?
So the hardware costs a fortune for the Cloud providers. The Cloud providers already lose heck a lot of money hosting and/or investing on LLMs. But now the frontier AI Labs have major pressure to lower token cost to a third since open weight is only days behind. Hosting LLMs on servers are fine but then the problem is the hardware needs to be re-updated every few years. So where's the insane profit of LLM?
How should we value AI related stocks now?
