I wrote a post on the free bulletin board and moved it here..... I thought it was meant for stock discussion, but ㅠㅠ since the tone fits better with stock discussion, I ended up deleting the post from the free board.
Due to the AI speed regulation theory, SOX has undergone an adjustment of more than 5.3%, and domestic markets are also experiencing significant adjustments.
The rising channel has not only been broken at the bottom but has shifted into a sideways pattern, and if it drops to the 6400 level from here, it is expected to transition into a falling channel.
In the midst of this turbulent market, VIX is still below 20 and the fear index is at the Fear stage. One might think, shouldn't this level be extreme fear? But this is only the case for tech stocks, while other sectors are in a relatively calm state. Even looking at just the S&P 500 index, it has only dropped very slightly.
With the conclusion that AI speed regulation = reduced competition = the end of the chip competition for learning, tech stocks including semiconductor stocks are all declining.
So what is the reason for AI speed regulation.... If we look at it,
Mankind has been regulating with the goal of reducing greenhouse gas emissions since 1990, but ultimately failed. To avoid repeating the history where control becomes impossible once the critical point shown by the hysteresis graph is breached, this means creating controllable AI through speed regulation before the critical point is breached.
The problem is that this is just superficial packaging, and it's also a venue where the three major AI companies boast to each other, saying "The AI we developed is uncontrollable."
As a result, the Citi Green report warning about the creep effect—where rapid AI growth shortens the connected cycle of production-distribution-expenditure and triggers economic recession—has also been highlighted.
I see this as noise. The reason is, if I first question whether semiconductors are no longer needed just because AI speed regulation is being implemented, it goes like this:
Hyperscalers' Investment Plans
Investment plans are already confirmed and in progress through 2027. These are plans that cannot be further reduced. The problem is that they could be halted for political reasons, or CAPEX investment through borrowing could become burdensome due to rising government bond rates. However, some experts explain that even if government bond rates rise, the return rate is higher, so the burden of CAPEX investment is lighter.
Business Strategy to Monetize by Converting Training Servers to Actual Operational Stage
It could be a business strategy to convert training servers to inference servers that cannot be increased in the short term even if substantial costs are incurred now due to increased usage, to cover the demand. Training servers must always use the latest hardware, and it is a strategy to postpone AI training until Nvidia's next-generation Vera Rubin is fully deployed.
Anyway, with the chips they currently have, further training will yield similar results, so until better results are achieved with the latest chips from Vera Rubin, it means converting existing training servers to inference servers, monetizing them, and then rushing to implement Vera Rubin to compete.
The Demand for Hardware Continues
https://contents.premium.naver.com/0301/gadonet/contents/260908062444074my
Just looking at Nvidia's plan for 400,000 additional units shows how severe the shortage is right now. The fact that they are restricting Astra's top-tier subscription signups shows that insufficient AI computing capacity is forcing these restrictions, and to meet this demand immediately, they need to redirect training servers. Anyway, OpenAI also agrees, as they also need time to achieve better results with Vera Rubin and train with improved models.
Now, semiconductors, which briefly led the sector's rise, are taking a break, and rotational buying is shifting toward security software. This is because controlling AI = strengthening security. To control AI more precisely, all logs from the inference process must be visible, and ultimately they must be stored and output in memory, making memory an indispensable necessity.
Unless we actually see AI infrastructure investment decreasing in the numbers, I think it's correct to view this as a market adjustment phase due to overreaction.
Until we see Micron's Q3 earnings announcement and the direction of additional guidance from hyperscalers at the end of this month, this turbulent market is expected to continue, interest rates and inflation will continue to shake the market, and oil prices will further complicate the situation.
I thought the fog was clearing, but instead it has grown thicker, and now it is raining. But once the rain stops and the fog disappears, when the dark, gloomy drizzle ceases, the path ahead will become clear.
Right now is not the end of the AI cycle but a period of reallocating its uses. Ultimately, this competition will intensify again when signs of China's advancement become visible. As soon as the speed regulation theory emerged, China was the first to oppose it, and besides, China is not just sitting idle.
Now is the time to view the market coldly. Although things are difficult now, I think this is the time to catch our breath while viewing the future optimistically.
