The AI midfield is a mix of ice and fire; market concerns and expectations are all reflected in the financial reports of the three major cloud vendors. Let's discuss my understanding of their financial reports in detail: 1. First, the "ice" part: 1) Undoubtedly, the market's biggest concern is the soaring capital expenditure (see Figure 1 below). Capital expenditure has accelerated visibly over the past six quarters. Google and Amazon's free cash flow is already negative, while Microsoft can maintain positive free cash flow this quarter and even for 2027. The cash flow gradient among the three giants has widened significantly. Of course, if you add back TPU inventory and EU fines to Google's financial report, its free cash flow is actually balanced. I discussed this last week here: https://(((x.com)))/qinbafrank/status/2115720160715923557?a=VO-j_qkvJt3-Oy7b-1w&s=46 However, it needs to be clarified that the three companies are not already in a liquidity crisis—the gap can be filled by raising funds from the capital market through bond or equity issuance. Google's free cash reserves can even support the current cash flow gap for three to four years. Of course, this cannot be done entirely; a safety net must be provided for the company. 2) The second point of the "ice" is that the core business is not as strong as expected. (1) Google's AI mode search volume surged, but the conversion rate has not yet picked up, and commercialization has not kept pace. The low monetization of AI mode is the most noteworthy structural factor in this case. It is normal for commercialization to be below expectations for a period of time during a paradigm shift; it needs time to catch up. The increased accuracy of answers under AI mode means that the cost per click can be higher to offset the decrease in clicks. Of course, if the commercialization of AI mode does not catch up significantly, the market will still be worried, and this concern is understandable. (2) Amazon's revenue was slightly lower than expected, largely because this year's Prime Day was moved to the second quarter in most of the company's important market regions, while last year it was in the third quarter. The advertising business underperformed in the third quarter. The company stated that advertising revenue was affected by factors such as Prime Day and expects year-on-year growth in advertising revenue to rebound to the high-double digits in the fourth quarter. As for Amazon, the market's greater concern is that if external AI agents become the main interface for users to search for and compare products, Amazon may still be able to maintain transaction volume, commissions, and logistics revenue, but it may lose its most valuable asset: pricing power for in-site advertising. Overall, with significantly increased capital expenditures and negative free cash flow, underperforming core businesses naturally exacerbates market concerns. 2. Let's talk about the "hot" parts: 1) The growth rate of cloud businesses is indeed encouraging. AWS grew by 39.3% year-on-year this quarter, Azure by 48%, and Google Cloud by 103%. This figure shows that the AI wave has brought real demand for computing power, and enterprise customers are endorsing this judgment with actual purchasing behavior. Moreover, a closer look at the growth rate of the three giants' cloud businesses over the past six quarters (Figure 2 below) reveals that the three have recently begun to accelerate. In the past three quarters, the year-on-year growth rate of cloud business has been discussed previously (https://(((x.com)))/qinbafrank/status/2114909089323225090?a=xGu2TW_SyYKrNtWQws0&s=46). Based on the data center construction pace, capital expenditure accelerated starting in the third quarter of 2025. With data center computing power gradually coming online from the third quarter of this year to next year, it means that cloud vendors' cloud performance will continue to accelerate. Although capital expenditure and free cash flow are poor, the rapid growth of cloud business performance cannot be ignored. Amazon CEO Jassy even explicitly stated in a conference call that AWS's growth rate in 2027 and 2028 would be at least 40%. Based on a minimum 40% growth rate, AWS's annual revenue could exceed $360 billion by 2028. In fact, conservative estimates suggest that Microsoft and Google's cloud businesses will have at least $300 billion in revenue by 2028. 2) More importantly, AI adoption is deepening and spreading. Among the enterprise AI adoption figures disclosed in the quarterly financial reports of the three cloud vendors, the most noteworthy aspect is not simply the increase in new customers or the number of seats, but rather the significant expansion of spending by existing customers and the clear extension of the payment model from seat licensing to usage/pay-as-you-go billing. (1) Microsoft Foundry has over 120,000 customers and revenue more than tripled year-on-year; Copilot's customers actively consuming pay-as-you-go credits increased by over 75% quarter-on-quarter; and GitHub Agent's credit consumption more than doubled since the introduction of pay-as-you-go billing. (2) Google's customers who used Gemini Enterprise last year spent more than four times more on Google this year, and in the past 12 months, they have formed over 2,600 high-intensity users with tens of billions of tokens and approximately 75 users with tens of trillions of tokens. (3) AWS Fortune 500 companies' spending on Bedrock increased nearly sevenfold year-on-year, and token usage increased nearly 17 times; AgentCore's agent deployments this quarter also reached twice the total of the previous three quarters. These data all point to the same trend: enterprise AI is transitioning from "purchasing seats" to "continuously consuming computing power and agents," with the spending depth and willingness to pay of existing customers becoming the most crucial growth engine. Meanwhile, the number of new customers is still increasing. The key takeaways from the financial reports of the three major cloud vendors are: 1) Regarding the time lag issue of AI infrastructure development, which we discussed in July (https://(((x.com)))/qinbafrank/status/2081226811514753045?a=T68Z_jTwqjJtK2k2F9U&s=46), we are already seeing new capacity being converted into orders and revenue, but operating cash flow has not yet fully covered capital value. This naturally leads to increased market anxiety. Coupled with the uncertainty of the midterm elections and high oil prices, this has triggered a sharp market correction. 2) The AI main theme, which we've been discussing before, is shifting: Memory → Optical / Software → Agent / AI for Science / Applications. Don't form the fixed mindset of "AI = Semiconductor = Memory," equating AI with semiconductors and semiconductors with storage. The main theme of AI shifts at each stage. Because with the commoditization of models and the decrease in token prices, it's not just upstream hardware that benefits. The cheaper the token, the easier the deployment, and the more benefited cloud, software, enterprise integration, and agent companies will become. The next stage may be AI for Science, Bio, and a broader application layer. Therefore, what trend investing really needs to do is not to cling to the existing winners, but to continuously identify where the next profit pool is migrating to. Semiconductors aren't dead, but the phase of everything rushing upwards like in the first half of the year may be over in the short term. Of course, what the market really needs to see is whether operating cash flow can grow at a faster pace after capital expenditure expansion, and whether customer payments can cover inference, depreciation, energy, and network costs. The core is "when will the investment start to generate sufficient returns?" I personally think we might see that in another quarter or two.