Foreign media reports that large American tech companies are investing tens of billions of dollars in AI, but increased capital expenditures do not necessarily mean that these investments have been translated into returns for shareholders. The article argues that what the market should focus on next is not who spends the most, but who can truly turn GPU, data centers, and cloud infrastructure into revenue, profits, and cash flow.
The article cites data from Coinpaper that in 2026, the combined capital expenditures of the five major hyperscalable cloud providers in the United States could approach $697 billion, and this proportion is also significantly higher than in 2023. Against this backdrop, when judging whether the investment in AI is effective, financial results are more important than the scale of the investment itself.
First, let's see if the income can keep up with the investment.
The article argues that the most direct method of observation is to compare the capital expenditures related to AI with the growth rate of business revenue.
Microsoft's capital expenditure in the June 2026 quarter was approximately $41 billion. During the same period, Azure and other cloud service revenues increased by 43%. Microsoft Cloud revenue rose to $59.3 billion, and the number of paid seats for Copilot exceeded 30 million. In terms of Alphabet, Google Cloud revenue in the second quarter of 2026 increased by 82% to $24.8 billion. The company's capital expenditure for the first half of the year reached $80.6 billion, and it raised its annual capital expenditure forecast to between $195 billion and $205 billion.
The article argues that the key is not just 'income growth', but rather the relationship between new income and new capital expenditure. If capital expenditure rises significantly, but the related business income growth is limited, the subsequent pressure on returns will increase; only if the investment leads to faster income expansion is the investment logic more solid.
Free cash flow is more important than profit.
The article points out that for AI, infrastructure investments are typically made first, with revenue generated later. Therefore, short-term profits do not fully reflect the financial pressures, and free cash flow is a more worthwhile indicator to track.
Meta Revenue grew by 28% to $60.8 billion in the second quarter, but capital expenditures reached $31.1 billion, with free cash flow for the quarter being only $784 million. The company expects capital expenditures for the full year of 2026 to be between $130 billion and $145 billion. Amazon is also facing a similar situation. AWS Revenue increased by 37% to $42.2 billion in the second quarter, and operating profit rose to $16.6 billion, however, free cash flow over the past 12 months was negative by $7.6 billion, mainly due to increased investments in infrastructure, property, and equipment. AI
The article argues that this does not automatically imply that the investment is ineffective, as infrastructure is often built first before generating revenue. However, if free cash flow continues to weaken after the capacity is put into operation, the market will become more cautious.
We also need to consider utilization rates and capital returns.
The article states that building a data center is just the first step; what's more important is whether the expensive GPU is being fully utilized. Although external investors usually cannot see the direct utilization rate, there are still several indicators that can be used to judge this, including whether cloud business revenue is accelerating, whether segment profit margins have improved, whether revenue per unit of infrastructure has increased, whether order reserves have expanded, and whether management has mentioned that demand exceeds current capacity.
Taking Microsoft as an example, the company stated that the demand for Azure still exceeds the available production capacity, and the remaining commercial fulfillment obligations have risen to 678 billion US dollars. The article argues that such data indicates at least that there are customers waiting to absorb the newly added computing power.
In addition, the returns from AI may not necessarily come solely from selling AI products; they could also be reflected in improved efficiency of existing operations and enhanced monetization capabilities. In the second quarter of Meta, ad impressions increased by 14%, and the average price per ad rose by 12%, driving a 28% increase in revenue. Microsoft's returns, on the other hand, may stem from increased Azure usage, more Copilot subscriptions, as well as an increase in revenue per 365 user.
At the end of the article, a simplified framework is provided: dividing the additional after-tax operating profit by the additional AI investment can be used to estimate the return on capital. If a company makes an additional $50 billion in AI investments and ultimately generates an additional $10 billion in after-tax operating profit each year, the implied return rate is about 20%; if it can only generate $2 billion, then the return rate is only 4%.
The article argues that when evaluating AI investments in the future market, the most important indicators to track will be: capital expenditure, revenue, profit margin, cash flow, and return on capital. If the former continues to rise while the latter do not keep up, the persuasiveness of AI investments will diminish.
