Dario Amodei’s call to slow down the pace of AI development—backed by Elon Musk and Sam Altman—has sparked a new concern in the market: Will AI investments lose momentum?

I do not view these statements as inherently negative for AI stocks. However, dismissing them by claiming they will have "no impact on infrastructure investments" would be overly optimistic.

First, it is necessary to distinguish between two different types of demand. On one hand, there is the massive spending dedicated to developing more powerful models. On the other, there is the usage of existing models—specifically, inference. Even if the development of new models slows down, companies will continue to utilize AI, generate code, and automate business processes. All of these activities require computing power, memory, and electricity.

However, Dario’s proposals go beyond merely delaying the market launch of new models; they also raise the possibility of imposing limits on the computing power used for training. If such an approach were adopted, the timelines for certain hardware orders and capacity investments could be affected. We do not yet know how quickly growth driven by actual usage might offset this impact.

On the other hand, Dario shows no signs of slowing down himself. According to *The Information*, the total cost of Anthropic’s computing capacity agreements could reach $517 billion. It is quite striking that a company calling for a slowdown is simultaneously seeking to expand its capacity on such a massive scale—a clear case of "preaching one thing while doing another."

Of course, this does not represent $517 billion in immediate spending; the agreements are spread over several years, with a significant portion covering the coming decade. In other words, this figure is nearly three times the previously announced projection of $180 billion in spending through 2029. My conclusion is this: Anthropic is preparing for a future where the need for computing power will grow significantly. It is possible to have both more controlled model development and expanding commercial use simultaneously. Therefore, there is insufficient basis to jump directly from a call for a slowdown to the conclusion that "data center and energy demand will collapse."

However, one must not overlook the role of expectations in share prices. A company’s stock price can fall even while the company continues to grow, simply because the price already factored in much faster growth. Orders do not even need to be cancelled; a delay of a few quarters in expected revenue is enough to call a high valuation multiple into question.

That is why I do not approach all AI companies in the same way. The risk profile differs between a cash-generating company with full capacity and a diverse client base, and a company that builds capacity through debt and relies on just a few major customers. Large capacity agreements are positive, but the timing of their conversion into revenue and the method of financing are equally important.

This is precisely the angle I take regarding the AI ​​energy bottleneck thesis I am monitoring. If data centers come online as planned and electrical grid connections and capacity contracts proceed smoothly, there is no issue. Trouble arises only if projects begin to face delays.

In the event of a broad market sell-off this week, I would not buy stocks indiscriminately just because prices have dropped. However, if companies with strong order books, growth prospects, and balance sheets experience a sharp sell-off due to this specific issue, I would look for buying opportunities.

I remain optimistic about the long-term potential of artificial intelligence, and Anthropic’s capacity agreements reinforce this view. Nevertheless, it appears we are facing a highly volatile market.