Reviewing the 2000 foam of SciDev. Net: Four signals, which AI is repeating? Everyone is asking whether AI is a foam, but they are asking the wrong question. The real question is: if it is a foam, in what form will it burst? The biggest similarity between the foam of Scinet in 2000 and the AI market today is not technology, but humanity. The biggest difference is not technology, but the source of money. If you understand this, you won't have to worry about whether Nvidia's next quarter financial report will beat or not, and you won't have to keep an eye on the news about Meta's rental computing power. Those are just appearances. The financing structure and the game of people's hearts behind the appearance are the core to determine whether and how the foam will break. 01) foam is not technology, but financing structure Many people will only say that Internet companies have no income and high valuation after reviewing the year 2000. This is treating the result as the reason. The foam in 2000 was tragic because the entire Internet industry was kidnapped by the financing capacity of the stock market. Companies without income rely on IPO for blood transfusion, VCs rely on IPO for exit, listed companies rely on refinancing and acquisition for expansion, and communication equipment manufacturers rely on the purchase orders of these non income companies to survive. The expansion of upstream fiber optic cables, optical modules, and routers is not due to real demand, but because there is unlimited ammunition burning money to buy the future downstream. The stock market is the real engine of this foam. Once the stock market stops financing, the entire industry ecosystem will instantly become ischemic. So the year 2000 was not a simple foam burst, but an engine shutdown. Today's AI market is different. The money burning companies now are wealthy giants such as Microsoft, Google, Amazon, and Meta. They do not rely on stock market financing, but on corporate bonds, operating cash flow, and balance sheets. The fuel for AI infrastructure is not IPO, it's Capex. This means two things: first, if the AI foam bursts, it will not be presented in the way of IPO suspension and batch death of start-ups; Secondly, it is more likely to gradually shrink from the enterprise end in a way that does not meet the expected return on capital expenditures. This kind of contraction will not be vigorous, but it will be very abrasive. Like a high-speed train, slowly releasing the accelerator instead of sudden braking. 02) The bursting of foam is not bad news, but narrative collapse The foam of 2000 was not punctured by a bad thing. What really killed it was a series of 'not good enough' news. When the market rises to that position, what is needed is not good news, but increasingly good news. Any neutral or 'normalized' statement will be interpreted as bad news. Because expectations have already been suspended in the sky, only exceeding expectations can be maintained, and in reality, there is no such thing as always exceeding expectations. This is the cruelty of the foam: it does not need real bad news, only good news is not strong enough. So you see, Cisco CEO Chambers said 'normalizing', which means the company is entering normal growth, but the market sounds like 'enough network equipment'. If Huang Renxun were to say at the performance meeting today that 'GPU supply and demand are now approaching balance', how would the market react? There are only two states in the industrial chain: shortage and surplus. There is no 'just right'. "Balance" at the peak of the foam is the prelude to excess. 03) The moment when four beliefs collapse If I had to organize the year 2000 into a signal, I would summarize it in four words: enough, no longer buying, slow down, and idle. That's enough. At first, some companies said that the bandwidth was already sufficient and there were enough fiber optic cables laid. This is not negative, but it undermines the narrative of 'perpetual shortage'. This round corresponds to the narrative of 'computing power is no longer scarce' - Meta renting computing power, shortening GPU delivery cycles, and increasing cloud vendor reuse rates are essentially variants of 'enough'. When the belief in scarcity is shaken, the valuation foundation of the entire industry chain will loosen. Because all the stories of the past two years were based on the assumption that 'computing power will never be enough'. Once this assumption starts to crack, funds will reassess every aspect. Not buying anymore. Telecom operators are starting to cut capital expenditures. They were the largest buyers back then, equivalent to today's super large-scale cloud vendors. Their procurement has slowed down, indicating that buyers are starting to run out of ammunition. If Microsoft, Google, and Amazon start to slow down data center construction or reduce capital expenditure growth from 50% to 20% today, the market will immediately become tight. What is even more alarming is that the financial fraud of these operators in 2000 was essentially because they could no longer meet their real needs and could only use accounting methods to disguise them. In this round of AI infrastructure, bears are already telling similar stories: Large scale cloud vendors include unused GPUs and data centers under construction in CIP without depreciation; There is an 18 month gap between GPU procurement and power on operation, with no indication of technological depreciation or hardware aging; Capitalization of expenditures results in a reduction of current expenses. All these have resulted in abnormal earnings growth of the S&P 500, which is a typical financial feature of the foam period. Slow down. The growth rate of orders from leading companies has slowed down. Cisco's performance was still growing back then, but the growth rate slowed down, and the market went from 120 times PE to 15 times. If Nvidia and Micron also experience "good growth but not as expected" today, the same valuation killing will happen again. In the foam stage, the second derivative is 100 times more important than the first derivative. This is the cruellest part of valuation in the foam period. The company's performance may still be growing rapidly, but the pricing in the capital market has shifted from "high growth" to "slow growth". Once the second-order derivative turns negative, the speed of valuation will far exceed the speed of performance changes. Idle. Iconic overcapacity has been discovered. In 2000, it was dark fiber, and a large number of pre laid fibers were unused. If someone were to confirm today that there is a large amount of idle GPU or HBM in the data center, or that the utilization rate of the training model's computing power is extremely low, it would be the fourth type of signal. This is not an expected issue, it is a factual issue. The reason why Dark Fiber became a landmark event is that it completely falsified the narrative of 'long-term shortage'. If words like "Dark GPU" or "Dark HBM" appear in the future, the main uptrend of this AI market is likely to end. These four signals progress layer by layer, from narrative wavering to buyer contraction, then to order slowdown, and finally to the confirmation of overcapacity. The further it goes, the less it is a guess, but a real hammer. In 2000, it took about six months to a year to transition from the first type of signal to the fourth type of signal. If AI takes a similar path in this round, the time window may also be similar. Why is it more covert this time The foam of 2000 was very transparent. The signals of no income, high valuation, and crazy IPO can even be seen by ordinary retail investors. When it finally broke, it was also a landslide, and Nasdaq fell 80%. If this round of AI foam is established, it will be much more hidden. Because the players have changed. The protagonists back then were start-up companies without income and retail investors chasing after gains and losses. Today is a time for tech giants, professional institutions, and global sovereign funds with abundant cash flow. They will not sell within a day, but will re evaluate the return on capital expenditure and gradually adjust the budget. This leads to a strange phenomenon: the AI industry may continue to grow rapidly, but related stocks may remain stagnant or even decline for a long time. Because market expectations have been raised too high, no matter how fast the actual growth is, it cannot keep up with expectations. This state of 'industry prosperity and stagnant stock prices' is the most torturous. It is also the most hidden way of foam bursting - not collapse, but slowly deflate. Many people will continue to copy the bottom in this state, because their underlying logic is correct - AI is indeed developing, but the foam of stock prices has not ended. 05) Survival strategies for ordinary people If you are not a sell side analyst, not an industry insider, and do not have the ability to track GPU inventory, HBM production capacity, cloud vendor CIP, and capital expenditure plans in real-time, then do not attempt to be a smart person in this game. The judgment of foam is too difficult. Even if you guess the direction correctly, it is extremely difficult to guess the time and magnitude. In 2000, there were many people who were short of the Internet, but many people were short too early and were crushed to bankruptcy. The market can persist in an irrational state for a long time, to the point where you doubt life. The only thing ordinary people can do is admit that they can't see clearly, and then use a long enough time window to blur the correct information. This tool is an index. Buy Nasdaq 100, buy S&P 500. You are not betting on whether a certain company will become the next Cisco, you are betting on the long-term upward trend of the technological revolution. Even if the foam really bursts, the index will also metabolize: bad companies will be eliminated, and new companies will make up. After the foam in 2000, the Nasdaq 100 rose more than ten times from the bottom. Moreover, compared with individual stocks, the index has a huge psychological advantage: you don't need to judge whether the foam has burst. Because regardless of whether it is broken or not, your strategy is to hold on. This will help you make 80% fewer mistakes. The temptation to guess the top and bottom is great, but history has repeatedly proven that those who hold indices for a long time outperform the vast majority of those who try to be smarter. History does not repeat itself, but human hearts do. In the previous round, people died from greed and fear, and this round is no exception. The difference is that smart people predict foam, and the real winner is just to buy the whole market and live long enough.