The core issue regarding AI investment now is the time gap, which can also be said to be one of the most important frameworks for understanding the AI capital expenditure cycle. Why do you say that? Looking at Google's financial report, and even the financial reports of Microsoft, Amazon, and Meta next week, there may be a pattern of "cash first payment (capital expenditure money is spent first), production capacity is launched later, orders are converted into revenue, and profits and free cash flow are finally realized". Spending money first and then looking at the efficiency and speed of making money is the essence of time difference. 1. Let's take a look at Google's financial report again. The second quarter of 2026 is a very typical case. This season: The total revenue was 119.8 billion US dollars, a year-on-year increase of 24%; Operating cash flow of 39.1 billion US dollars; Capital expenditure of 44.9 billion US dollars; Free cash flow is approximately negative $5.9 billion; Google Cloud revenue reached 24.8 billion US dollars, a year-on-year increase of 82%; The operating profit of cloud business was 8.8 billion US dollars, with an operating profit margin of 35.6%; The backlog of cloud business orders is $514 billion, with an increase of approximately $50 billion in a single quarter; More than half of the backlog orders are expected to be confirmed in the next 24 months. On the surface, the company's negative free cash flow seems to indicate a deterioration in AI investment returns. But the data of cloud business is accelerating at the same time: Cloud revenue growth of 82%+cloud profit margin of 35.6%+Backlog continues to increase This indicates that capital expenditure is not blindly built without demand, but rather expanded when existing demand exceeds supply. The management has also raised the 2026 capital expenditure guidance from $180 billion to $190 billion to $195 billion to $205 billion, also due to the accelerated delivery of equipment and data center capacity 2. The time difference here is divided into four steps: Step 1: Pay for the server and data center fees first Approximately 60% of Alphabet's capital expenditures this quarter were spent on servers, and approximately 40% were spent on data centers and networks. Server payments, land construction, power access, and network equipment all occur before revenue recognition. Google disclosed its own capital expenditure construction cycle of several months to several years. That is to say, the newly added revenue of Google Cloud in the second quarter of this year is basically brought by last year's capital expenditures, and the revenue generated by this year's capital expenditures will not be gradually realized until next year. Step 2: Capacity deployment and acceptance Buying a chip does not mean immediate billing, it also requires installation, networking, debugging, software adaptation, and customer deployment. Step 3: The order begins to be converted into revenue Even if the customer has signed a multi-year contract, revenue will not be recognized at the time of signing, but will be recognized gradually after the customer continues to use cloud services, computing power, or hardware delivery. Google has disclosed that some TPU system sales will only recognize a small amount of revenue in 2026, and the vast majority of related revenue will not be recognized until 2027. Step 4: Convert income into free cash flow Even if revenue begins to be recognized, it may still be suppressed in the short term by the following factors: Increase in depreciation of new assets; Increased electricity and network costs; The initial utilization rate of the new data center has not yet reached a mature level; Renting third-party capacity to alleviate insufficient computing power incurs higher costs than self built capacity; Customer migration and deployment require time. Alphabet has indicated that using third-party capacity as a transitional solution will put pressure on short-term profit margins. Therefore, Alphabet is currently not 'without returns', but rather: We have seen revenue, orders, and profit margin returns, but the growth of new capital expenditures has been faster, resulting in a temporary lag in free cash flow at the company level. 3. Several key points to pay attention to regarding time difference 1) Cannot directly compare the growth rate of capital expenditures with the growth rate of current income Capital expenditures purchase assets that can be used for many years, while income is the flow of a quarter or a year. For example, a data center may invest $10 billion at once this year, but generate revenue annually for the next decade. Comparing this year's $10 billion capital expenditure directly with this year's new revenue naturally underestimates project returns. The correct approach is to model each batch of assets separately: Accumulated project revenue - Accumulated project operating costs - Maintenance capital expenditures - Taxes - Financing costs. Then calculate the project IRR and ROIC. The best model is naturally analyzed using capital batches, production lag, and full lifecycle cash flow. 2) High growth in backlog is a good thing, but it also depends on the pace and quality of backlog to actual revenue conversion Order backlog can only prove demand intention, but cannot automatically prove final profit. To further assess: How many will be confirmed in the next 12 months and 24 months respectively; Can the contract be cancelled; Is capacity delivery a prerequisite; Whether the customer has made a prepayment; Is the contract price fixed; Can the cost increase be passed on; Is the customer's minimum usage commitment rigid. 3) High quality time difference High quality time difference should see conversions every quarter Orders → Under construction capacity → Online capacity → Revenue → Segment profit → Operating cash flow → Free cash flow If you only see: Capital expenditure increases+Backlog increases But without seeing the capacity increase, revenue recognition, and profit improvement, the time difference may have shifted from a normal construction cycle to a construction delay or deterioration of capital efficiency. Google's Q2 financial report shows us a signal of high-quality time lag Summary The current AI investment cycle can indeed be summarized as: Capital expenditures have entered an explosive stage, and revenue, orders, and profits are also accelerating, but most of the newly built production capacity has not yet been fully put into operation, so the realization of free cash flow lags behind capital investment. The future contract gross profit, capacity utilization rate, and cash recovery rate corresponding to each dollar of new capital expenditure that we need to pay attention to are improving, and the recovery period is shorter than the asset obsolescence period and financing period. Only when the following relationships are met simultaneously, is the time difference worth waiting for: Increased demand certainty+accelerated order conversion+improved capacity utilization+incremental ROIC>WACC+balance sheet can withstand Conversely, once: Capital expenditures continue to increase+production capacity is no longer scarce+Backlog growth rate decreases+confirmation cycle extends+segment profit margin declines+financing costs rise So the market will quickly reprice the 'normal time difference' as' excessive AI construction '. Simply put Time difference is the most important starting point for analyzing AI capital expenditures; The endpoint that truly determines investment returns is the ROIC, payback period, and balance sheet safety of the capital batch.