Although I have previously expressed optimism about liquid cooling, it is indeed not the case that all can deliver performance, and there is certainly a risk of overheating in the short term.
The fundamental reason is that this is indeed not an industry we dominate, unlike PCB and optical interconnects. Some upstream PCB companies moved last week, and I tend to believe that products like electronic cloth can deliver performance (FLH has been working on it for a while), because the downstream is our comparatively advantageous industry.
This is also why I look forward to the domestic computing power beta; many niches can drive dozens of stocks with real performance improvements (such as switches, liquid cooling, engines).
The reason why robots can drive up hundreds of stocks is also because this is completely our dominant industry, and the downstream mainframe manufacturers are all insiders; it is normal for capital to act in advance. On robots, I would like to say a bit more: I predict there will be a significant breakthrough in 2026. Currently, robots are divided into two camps: one is the synthetic data faction represented by NV, and the other is the real data collection faction represented by Tesla and ZhiYuan. ZhiYuan currently has 1,000 robots collecting data, which can reach tens of trillions of tokens in a year, and achieving this scale with synthetic data is also not a problem. According to scaling law, when the data scale reaches the level of tens of trillions of tokens (whichever route it takes, this will be achieved in 2026), the robots' 'GPT-3' moment will emerge, so physical AI will become a new scaling direction, and of course, those selling shovels like robots and NV will benefit.
Let’s also talk about consumer electronics; Apple is estimated to launch three AI smart hardware products within two years: a lamp robot, HomePod, and AI camera. Cook has positioned the AI hardware strategy very high, unlike Vision, which is actually a departmental strategy that Cook does not prioritize. Many teachers back then were misled into thinking that Apple could create anything, without understanding Apple's internal strategic priorities. Last night, Ultraman also mentioned entering the consumer electronics market, indicating that edge AI hardware is actually a consensus for AI applications (robots are also edge, and imagination needs to be unleashed).
[Dog head]
). Additionally, Tesla's Grok4 has a slightly different route; Musk comes from a product background, and the internal heavy bet is on the emotional companionship route (which also needs to be realized at the edge).
I don't want to write too much about software (to avoid being scolded by the teachers); the viewpoint from Silicon Valley is that 'AI will swallow all software.' The term 'swallow' may be exaggerated; my view is that if SaaS in North America can prove it hasn't benefited from AI, it will be enough to kill the valuation.
