A data center built today may need to recoup its investment over the next decade or more; the chips installed in it may face competition from new-generation products within a few years. Even if computing demand continues to grow, computing prices, equipment utilization, energy costs, and technological substitution will constantly rewrite the initial return estimates.
Is "good news" also read as "bad news"? First, tech giants like Google delivered strong earnings, then Nvidia unveiled a $500 billion industrial-financial cooperation plan, but the market's response is hard to summarize.
In the view of Sonali Basak, Managing Director and Chief Investment Strategist at U.S. alternative asset fintech platform iCapital, cash flow pressure is raising the bar for AI investment evaluation, but it has not yet constituted a systemic financing crisis for all hyperscalers; what really needs to be distinguished is balance sheet buffers, vertical integration capabilities, and the speed at which AI revenue is realized.
Perhaps Wall Street is no longer fixated on whether AI trades have bottomed out or whether the bubble has cleared. After the July liquidation of valuations and positions, August's "interrogation" has delved into income statements, cash flow statements, and even the sites of data centers, power, and chip delivery. The grand narrative has not disappeared, but it has been placed under a dual microscope of "finance and physics."
On August 10, Nvidia closed down 2.86%, at one point falling more than 3% during the session, with a single-day market value loss exceeding $70 billion; the Philadelphia Semiconductor Index fell 2.94%, and shares of optical communication leaders Coherent and Lumentum also dropped significantly.
That day, Nvidia announced it had signed memorandums of understanding with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs, and KKR to establish an independent computing financing platform, gradually mobilizing over $500 billion in third-party capital for AI infrastructure construction.
Nvidia's definition of this matter is quite ambitious: transforming Nvidia's computing power and full-stack AI infrastructure into assets investable by global capital, attracting insurance, asset management, private credit, and infrastructure funds through long-term revenue linked to usage.
Nvidia founder Jensen Huang said that Nvidia has moved from manufacturing chips to helping create a new kind of productive infrastructure—"AI factories." He also stated on social platforms that Nvidia may choose to support up to 25% of the residual value of collateral in potential transactions.
Notably, the specific commitments of the relevant institutions, financing costs, final agreements, and deployment timelines have not yet been disclosed.
The establishment of this financing platform can lower the capital threshold for customers to purchase GPUs and build data centers in one go, and open up a larger potential market for Nvidia. At least in short-term trading, the market has not directly converted the $500 billion into new orders and profits.
The reason may lie within this number. The $500 billion is third-party capital to be gradually mobilized in the future, not orders already received; a memorandum of understanding is not a final agreement. More importantly, when Nvidia needs to unite six global capital giants to build financing channels for customers, the market sees not only strong demand but also asks: Has AI construction become so expensive that it can no longer be supported solely by tech companies' own cash flows? Is external capital validating demand, or is it sustaining demand?
This may not simply negate AI's long-term value, but rather indicate that pricing standards have changed. Financing intention does not equal project approval, project approval does not equal data center completion, and data center completion does not equal computing power being fully utilized and generating cash returns.
What happened? The answer may lie in the second-quarter results released by tech giants. July 2026 may become a turning point: tech giants still delivered growth reports, but the market began to read them with a different method.
On July 22, Google parent Alphabet (stock price data using Class A shares GOOGL, i.e., "Google-A" caliber) announced its Q2 2026 results: revenue reached $119.8 billion, up 24% year-over-year; Google Cloud revenue increased 82% year-over-year, with operating profit reaching $8.8 billion, more than doubling from the same period last year. The next day, Alphabet's stock fell 7.13%.
A week later, Microsoft reported quarterly results for the period ending June 30: revenue reached $90 billion, up 18% year-over-year. On July 30, Microsoft's stock rose more than 15%, marking its biggest single-day gain in 18 years, with a single-day market value increase of about $450 billion.
Both companies are doubling down on AI, both say demand exceeds supply, and both cloud businesses are growing rapidly. Wall Street's answers, however, were one down and one up.
The difference lies behind the income statement. Alphabet's quarterly operating cash flow was $39.069 billion, capital expenditures reached $44.924 billion, and free cash flow (FCF) thus fell to -$5.855 billion, the first quarterly negative since the company went public.
Microsoft also invested heavily, with quarterly cash capital expenditures of $35.8 billion and another $5.6 billion in finance leases; but its operating cash flow reached $55.4 billion, up 30% year-over-year, and free cash flow remained at $19.6 billion. Microsoft's cloud platform Azure revenue increased 43% year-over-year, and commercial remaining performance obligations reached $678 billion, up 84% year-over-year, with next quarter's Azure revenue growth guidance also above market expectations.
Operating cash flow growth, accelerating cloud revenue, and expanding order backlog together constitute a relatively clear return path. Microsoft CFO Amy Hood said the company is more confident in its return on invested capital, citing reasons including expanding market space, improved model and chip efficiency, and a broadening AI product portfolio.
Of course, accounting treatments still need to be discerned. Some data center leases have been converted from finance leases to operating leases, which can change where capital expenditures appear but do not eliminate future payment obligations.
However, this does not mean the market has formed a simple rule that "positive free cash flow leads to gains, negative leads to losses."
Meta's quarterly free cash flow fell 91% year-over-year, from $8.55 billion to $784 million; capital expenditures of $31.1 billion almost used up its $31.9 billion operating cash flow, and its stock fell up to 10% after hours. Apple, however, provided a counterexample: in the company's fiscal third quarter ended June 27, 2026, revenue grew 16% to $109.4 billion, operating cash flow set a record for the period, and its AI investment model is much lighter than cloud vendors, yet its stock still fell more than 8% at one point the next day.
The market is worried about supply constraints, end demand, and future guidance. Apple shows that free cash flow is the most prominent clue this quarter, but it is not the only trading switch. What the market scrutinizes is whether a company can provide a credible explanation for investment, growth, and valuation simultaneously.
The same "microscope" is also moving to the hardware side. In June, Micron Technology delivered a record $18.3 billion in adjusted free cash flow, and its stock rose 15.7% the day after earnings; by August, SanDisk and Western Digital reported strong results with revenue guidance above analyst consensus, but their stocks fell as much as 13.3% and 19.1% respectively during the session.
Demand remains strong, but how much longer storage prices can rise and whether earnings upgrades can catch up with previous gains have become new questions. The financial microscope measures cash, while the physical microscope begins to measure prices, capacity, utilization, and technological iteration.
Some analysts point out that these seemingly contradictory price reactions actually point to the same change: capital expenditures do not all enter the income statement in the period of construction, but they first flow out of cash flow. Profits can prove that the business is still growing, while free cash flow exposes the immediate cost paid for growth; whether orders and guidance can provide a credible defense for this cost determines how long the market is willing to tolerate it.
Simon Taylor, founder of fintech content platform Fintech Brainfood, gave a vivid summary of Alphabet's stock performance after its earnings release: the sell-off expresses the market's opinion on future returns, while backlog orders come from contracts already signed; between the two, only the latter is binding.
In other words, the market is not repricing "whether there is demand for AI," but whether the speed at which demand converts into revenue, profit, and cash can keep up with the speed of capital investment.
Free cash flow is not mysterious; it roughly equals cash generated from operating activities minus capital expenditures such as purchases of property, plant, and equipment. The income statement records how much money a company earned in an accounting sense; free cash flow asks how much cash actually remains after paying for this period's construction investments.
This becomes especially important in the AI era. Morgan Stanley analyst Brian Nowak asked Alphabet CEO Sundar Pichai: Compared with a year ago, how does the company assess the capital return potential and realization timeline for generative AI?
Pichai remains optimistic. He believes AI is still in the early stages of structural change, and whether in consumer services or enterprise applications, there are opportunities for "extraordinary returns," provided execution is solid.
A week later, Goldman Sachs analyst Gabriela Borges asked Microsoft CFO Amy Hood almost the same question: If you look at capital expenditures and commercialized revenue together, how has the return on investment of the investments Microsoft is making today changed compared with a year ago?
The two analysts coincidentally pushed the question from model capabilities, cloud revenue, and supply constraints to return on invested capital.
This may mean that the market's test is deepening layer by layer. The first layer is demand: whether customers are willing to purchase AI services. The second layer is revenue: whether cloud services, inference, subscriptions, and agents can generate scale revenue. The third layer is cash: whether new operating cash can cover chip, server, and data center expenditures. Finally, there is the full return on capital: whether the cash flow generated by these assets over their entire lifecycle can cover depreciation, energy, financing costs, and the returns required by shareholders.
The accelerating growth of Google, Microsoft, and Amazon's cloud businesses shows that AI is not just about investment without customers.
Revenue is also beginning to show. Alphabet's cloud business profit margin has risen significantly, while Microsoft reports continuous improvements in the efficiency of models, chips, and data centers. However, these companies have not yet separately disclosed the full revenue, profit, and cash flow of their AI businesses, so the market still cannot match every dollar of AI capital expenditure with corresponding returns.
The stress test at the cash level is just beginning. Investors are distinguishing: who can still rely on cash from existing businesses to complete construction, whose capital expenditure is eating up cash faster, and who needs to maintain expansion pace through corporate bonds, leases, and project financing.
As for the ultimate return on capital, there is not yet enough data to answer.
A data center built today may need to recover its investment over the next decade or more; the chips installed in it may face competition from new generations within a few years. Even if computing demand continues to grow, computing prices, equipment utilization, energy costs, and technological substitution will constantly rewrite the initial return calculations.
What Wall Street is doing now is using free cash flow to start the first round of screening.
This article is from WeChat public account: Economic Observer , author: Ouyang Xiaohong
