The financial reports of super large CSPs (Microsoft, Amazon, Google, Meta) are the second watershed of the market in July. What are the concerns and worries of the market? Their importance is discussed in the referenced tweet, where the performance of the super large CSP is already a key indicator for measuring AI commercialization (with weights consistent with the ARR of the large model). The core still depends on the growth rate of AI demand, income quality, and unit economy. Can it catch up with the rising speed of capital expenditure, depreciation, energy costs, and financing demand? Simply put, the commercialization of AI cannot keep up with the speed of AI capital investment and cost growth, and the free cash flow of big technology will also be put on the table. Let's talk about my thoughts. 1. Market concerns can be roughly divided into four layers: 1) High expectations and crowded transactions The market has already included the sustained Capex, chip demand, and cloud growth for the next few years in the price. Several major technology companies will have a combined Capex of approximately 730 billion US dollars by 2026. As long as the financial report is good but does not continue to exceed expectations, crowded semiconductor, storage, data center, and electricity trading may also decline. But it cannot be simply said that the entire major technology sector has already been included in the 'perfect expectations for many years'. Microsoft, Alphabet, Amazon, Meta, as well as Office, search advertising, e-commerce, social advertising, and other large non AI cash flows. The supply chain companies that are highly tied to AI Capex in terms of revenue and have greater cyclical elasticity are usually the ones that are most easily included in multi-year expectations. The recent deleveraging has greatly reduced the level of trading that is higher than it and crowded trading 2) Mismatch of capital expenditures, depreciation, and free cash flow Revenue can increase immediately, but data centers GPU、 The depreciation caused by power supply and network will last for many years. The current market is shifting from "who spends the most" to "who can quickly enter paid utilization of newly added capacity and generate gross profit sufficient to cover depreciation and capital costs". Amazon's negative free cash flow in the first quarter, Microsoft Cloud's gross profit margin under AI investment pressure, and Alphabet's clear warning of rising depreciation and energy costs indicate that this concern is not purely emotional. 3) The market is concerned about both "too much Capex" and "Capex being reduced" This is not contradictory, as the affected objects are different: For CSP shareholders, an increase in Capex without returns is negative; For chips, storage, data centers, and power supply chains, Capex's cuts are negative for revenue; Reducing Capex due to improved efficiency while maintaining revenue and capacity is positive; Reducing Capex due to weakened demand is negative for the entire AI industry chain. So what the market really cares about is not whether Capex absolutely increases or decreases, but rather: How much sustainable gross profit and free cash flow can each additional Capex bring 4) There is a time lag between the revenue and investment cycle of AI commercialization The current demand for cloud AI is strong, and some CSPs are still constrained by capacity. However, it will take longer for enterprises to shift from pilot projects to production, renewal, and expansion on a large scale. The market is concerned about whether enterprise AI revenue can quickly catch up after depreciation and financing costs first enter the financial statements. In addition, interest rates, energy prices, and macro risks can also affect long-term technology assets, and not all pullbacks can be attributed to AI. Investors are oscillating between the two narratives of "AI infrastructure being undervalued" and "investing too much and too early". 2. What can truly reverse market sentiment Simply releasing more models, increasing the number of tokens, or raising Capex again are not enough to alleviate concerns. The market needs to see a continuous chain of evidence. Firstly, AI revenue has breadth, rather than relying on a few strategic customers It must be seen that the number of enterprise customers, production workloads, paid seats, renewals, and per customer consumption are growing synchronously, and after excluding a few super large contracts such as OpenAI and Anthropic, demand remains strong. Secondly, AI gross profit growth is faster than depreciation and operating costs Short term gross profit margin reduction is allowed, but the newly added AI gross profit must significantly exceed the newly added depreciation, energy, network, and talent investment. It would be best to see a decrease in unit inference costs while the total gross profit continues to increase. Thirdly, the backlog can be converted into recent revenue A large contract of only three or five years is not enough. The market will place greater emphasis on the revenue recognition ratio for the next 12-24 months, the extent to which consumption exceeds initial commitments, and the immediate utilization rate after the launch of new capacity. Fourthly, self-developed chips and model optimization bring verifiable economic benefits What the market needs is not just 'Trainium, TPU, MAI are cheaper', but also: an increase in the proportion of self-developed chips, a decrease in the cost per token or successful task, achieving a price reduction smaller than the cost reduction, an increase in gross profit per accelerator hour, and continued growth in customer consumption and platform ancillary revenue. Fifth, the trough of free cash flow becomes visible The market should see that the growth rate of Capex is beginning to be controllable, new Capex is supported by contract demand, operating cash flow growth can cover more and more capital expenditures, repurchases and balance sheets will not be squeezed in the long run, depreciation peaks and investment payback periods can be explained clearly. Whether free cash flow has turned negative is just an appearance, what the market really trades is the reason, magnitude, duration, and whether it can recover in the future. Sixth, enterprise customers can prove the true ROI The most powerful evidence is not model evaluation or token count, but customer disclosure: increased revenue and conversion rates, decreased labor and processing time, transition from pilot to production, renewal and expansion of deployment, all AI costs included, and investment payback period still attractive. What the market most hopes to see for the ideal "blonde girl combination" in the second quarter financial report is: Cloud and AI revenue is higher than expected+profit margin is roughly stable+backlog conversion begins+Capex has not unexpectedly lost control, or new Capex is clearly corresponding to signed demand+free cash flow is no longer deteriorating. Conversely, the most dangerous combination is: Token and usage have surged, but there has been no improvement in per customer spending, gross profit, and free cash flow; RPO growth relies on a single strategic model company; Capex has once again seen a significant increase, while profit margins and revenue guidance have decreased