Whether AI is already out of control from human？

Author : James Madison | Published On : 15 Sep 2026

In last Friday and Saturday, Dario released an article "We Must Pace the Frontier", ask frontier AI develop organizations slow the developing speed of AI. He suggested that we should take three measurements to control the AI development. 1. Bringing in Third-Party Organizations to Assess the Risks of Large AI Models 2. Coordination and Unified Action in the AI Industry 3. Sign a global agreement to ensure the security of AI research and development. His suggestion was quickly endorsed by Sam Altman and Elon Musk.

Prior to this, Jacob Coxon—a former researcher at Anthropic—said on X that AI could kill all humans within a decade, and that this was not just a marketing gimmick. These events led to a sharp decline in AI hardware and tech stocks on Monday.  I think this is excessive panic.  The main reason is that many AI applications that currently consume large amounts of tokens do not use the most advanced models, but rather the most cost-effective ones. This is why chinese open-source AI models are so popular in world. One month ago, the CEO of Deepseek Wenjie Lian said that the most shortage thing for DeepSeek is the GPU chip. Anthropic and OpenAI's largest sources of revenue are not their most advanced, cutting-edge models.

Among the factors influencing demand for AI infrastructure, the most important is still the demand for AI from businesses and individuals. We have not yet seen any signs of a slowdown in demand. For me personally, the development of AI agents has enabled me to use them to complete many tasks that I simply couldn't have done before. As a result, I now rely on AI even more, and my consumption of computing power has increased. I don't think I'm the only one who feels this way. The token consumption had a surge risen in past six months as the ability of AI agent improves. For example, we can generate a new website in one hour by AI agent now, so we can develop many websites for the off-line shops. We can also develop many Customized Personal Apps for someone.

So, the price of AI hardware and AI infrastructure will rebound in follow several days. There has been a persistent shortage of AI hardware in recent. Unless chip production at wafer fabs increases significantly, this shortage is unlikely to ease. So, for AI hardware and AI infrastructure stocks right now, every major dip presents a buying opportunity.

For Coreweave and Nebius, in pursuit of rapid expansion, they have all raised substantial amounts of capital, resulting in high interest costs. Their EBITDA margin are higher than many public companies. For Maase Inc, as their own the low cost electric and employee salary expenses make it can get higher EBITDA margin once its revenue reaches a certain level.

At present, stocks in the artificial intelligence infrastructure sector remain good buying opportunities.