To understand the "time difference" framework from Amazon CEO's speech during the earnings call, what Amazon CEO Andy Jassy paid the most attention to today's earnings call was not his statement that AWS will reach trillions of dollars in revenue in the future, which is also important but not the most crucial. The most crucial thing is that he finally provided a timetable for capital expenditure investment returns, which was completely absent from the Google call. The "time difference" framework we discussed earlier can be well understood from Jassy's speech. Andy Jassy believes that capital expenditure investment mainly consists of two parts: data centers, as well as the servers and network equipment placed within them. They have different capital cycles: 1) The construction period of a data center is generally one to two years, and once the data center is opened and servers are inserted, it will immediately start generating considerable revenue. The service life of data centers is about 30 years, which means that these data centers can continue to generate revenue for more than 30 years without the need to invest the startup capital again. 2) The cycle of servers and network devices is relatively short. These devices are usually purchased a few months before the data center is put into use, so there needs to be strong visibility into customer needs before triggering expenses. If there is no demand, Amazon will not spend this capital. For servers and network equipment, on average, this investment takes less than three years to achieve breakeven. The service life of servers is currently at least five to six years, while most of Amazon's AI capacity contracts have a minimum term of five years. This means that within two to three years after achieving breakeven, Amazon will generate significant free cash flow on servers and network devices. And AWS has a good track record in shortening the break even cycle of server equipment. We have made meaningful progress without affecting customer experience and found ways to extend the service life of these devices. 3) For data centers with a lifespan of over 30 years, AWS should at least benefit from the economic benefits of five to six generations of servers as explained before, and the subsequent generations after the first generation have even better overall economic benefits because they do not require repeated previous data center investments. When demand forces so many data centers to be built simultaneously before monetization begins, it will incur significant capital expenditures in the short term and face free cash flow headwinds until these data centers go online and can be monetized, and after these servers are utilized for several years. As years pass and income growth exceeds incremental growth in capital expenditures (which will occur at some point), the resulting income, free cash flow, and return on investment capital will be significant. AWS experienced this situation in the first era of cloud computing, but over a longer period of time because the accumulation of demand was more gradual than in the AI era. But now we see that the profit margin and return rate in the AI field are consistent with those seen at the same stage of core business development, and in fact, they are still slightly ahead. From the beginning of the second paragraph until here, it's basically Andy Jassy's exact words. According to Andy Jassy mentioned that servers and network equipment can achieve investment recovery in less than three years on average, while the service life of servers is at least five to six years, and most AI capacity contracts are at least five years. If this statement can be cashed out based on real cash flow, then the approximate lifecycle of a server batch is: Year 0: Procurement of servers and network equipment Year 1-3: Accumulated recovery of initial investment Years 4-5/6: Entering a phase of significant positive free cash flow Subsequently, the server will be updated, but the original data center, power, and land will continue to be used. This model can theoretically generate ROIC higher than the cost of capital. This is actually the core essence of the time difference we talked about before The cash first payment (capital expenditure money is spent first), production capacity is put online later, orders are converted into revenue, and profits and free cash flow are finally realized "model, spending money first and then looking at the efficiency and speed of making money