After watching Jane Stree's live action video of a globally renowned quantitative trading company's data center today, I truly felt that the next stage will be the rotation of AI's essential sectors, most likely electricity and liquid cooling! Although the comparison chart between Nvidia VR200 and GB300 has been hotly hyped recently, including storage, MLCC, PCB, etc. In fact, they have all gone up once, and the magnitude is not small, and the valuation is not cheap. And liquid cooling and electricity, as the most important links in data centers, are not allowed to have any faults, otherwise the $7.8 million VR200 will be scrapped on site, and no one can afford this cost. Therefore, their importance and irreplaceability will be even higher. The key point is that these companies are not too expensive at the moment, and the safety cushion is high enough! In the video, those neatly arranged GB300 cabinets have a peak power consumption of 140 kilowatts and cannot be suppressed by air cooling. And the next generation Vera Rubin Ultra can run directly to 600 kilowatts per rack, equivalent to the electricity consumption of 500 American households, all packed into a 7-foot high iron cabinet. This kind of leapfrog upgrade is basically at the physical limit. Now the entire Wall Street quantitative agencies and Silicon Valley giants are crazily expanding their AI data centers. At Microsoft's Q1 earnings conference, Nadella said that he added 1GW of capacity in a single quarter and plans to double it within two years. The combined Capex of Google and Meta in 2026 exceeds $700 billion. And one-third of this money is not for buying GPUs, but for buying transformers, bus ducts, liquid cooling systems, backup generators, etc. What's even more interesting is liquid cooling. Jane Stree and her team talked about a very funny thing. In the past, they used to just check if the fan could turn, but now they have to worry about the server growing moss every day. 90% of the heat in the computer room is carried away by the liquid in the cold plate. In order to prevent micron sized gaps from being blocked by bacteria, the water needs to be filtered to 25 microns and mixed with 25% propylene glycol. Jane Stree, the technical director, also spoke the truth that hardware is indeed expensive, but what they are most afraid of is opportunity cost. Due to the scarcity of computing power, various internal trading teams are queuing up to compete for quotas. If the circuit breaker trips due to the power load not being accounted for, and the training task is rolled back directly, the missed trading strategy and real money are the most fatal. This is not only for quantitative institutions, but also for major manufacturers. The opportunity cost far exceeds the hardware cost, and ensuring the safety and reliability of long-term operation and maintenance is the most critical. And some of these liquid cooling operation and maintenance companies have gradually developed a lifelong SaaS model similar to the elevator business, which may become the potential hidden cash flow king of AI services. I will publish another tweet in the future, detailing this type of listed company. Put the original video in the comments section!