A data center built today might require over a decade to recoup its investment; the chips installed within could face competition from newer generations in just a few years. Even if the demand for computing power continues to grow, factors such as the price of compute, equipment utilization, energy costs, and technological substitution will constantly rewrite the initial return calculations.
Reading 'positive' as 'negative'? After tech giants like Google delivered strong earnings reports and NVIDIA proposed a $500 billion industry-financing cooperation plan, the market's response has been nuanced.
In the view of Sonali Basak, Managing Director and Chief Investment Strategist at iCapital, an alternative asset fintech platform, cash flow pressure is raising the bar for AI (Artificial Intelligence) investment scrutiny but hasn't yet constituted a systemic financing crisis for all hyperscalers. The real differentiation lies in balance sheet buffers, vertical integration capabilities, and the speed at which AI revenue materializes.
Perhaps, Wall Street is no longer fixated on whether AI trades have bottomed out or if the bubble has been cleared. After clearing valuations and positions in July, the 'interrogation' in August is delving deeper into income statements, cash flow statements, and even data center floors, power grids, and chip delivery sites. The grand narrative hasn't disappeared; it's simply being placed under a dual 'financial + physical' microscope.
On August 10th, NVIDIA closed down 2.86%, having fallen over 3% intraday, with its market value shedding more than $70 billion in a single day; the Philadelphia Semiconductor Index dropped 2.94%, and shares of optical communication leaders Coherent and Lumentum also retreated significantly.
That same day, NVIDIA announced it had signed memoranda of understanding with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs, and KKR to establish an independent compute financing platform, aiming to gradually mobilize over $500 billion in third-party capital for AI infrastructure construction.
NVIDIA's definition of this endeavor is ambitious: to transform 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 usage-linked revenue.
NVIDIA founder Jensen Huang stated that NVIDIA has evolved from making chips to helping create a new type of productive infrastructure—'AI factories.' He also mentioned on social platforms that NVIDIA could choose to support up to 25% of the collateral residual value in potential transactions.
Notably, specific commitments from the involved institutions, financing costs, final agreements, and deployment timelines have not been disclosed.
The establishment of this financing platform can lower the upfront capital barrier for clients purchasing GPUs (Graphics Processing Units) and building data centers, opening a larger potential market for NVIDIA. At least in short-term trading, the market hasn't directly converted the $500 billion figure into new orders and profits.
The reason might lie within this number itself. The $500 billion is third-party capital intended to be gradually mobilized in the future, not orders already booked; a memorandum of understanding is not equivalent to a final agreement. More importantly, when NVIDIA needs to team up with six global capital giants to build financing channels for clients, the market sees not only robust demand but also asks: Has AI construction become so expensive that it's difficult for tech companies to sustain solely with their own cash flows? Is external capital validating demand or fueling its continuation?
This doesn't simply negate the long-term value of AI but may indicate a change in pricing standards. Financing intent doesn't equal project approval, project approval doesn't guarantee data center completion, and completion doesn't ensure computing power is fully utilized and generates cash returns.
What happened? The answer might be hidden in the Q2 report cards released by tech giants. July 2026 may have become a turning point: the tech giants still presented growth reports, but the market began reading them with a different methodology.
On July 22nd, Google parent Alphabet (using Class A share GOOGL, i.e., 'Google-A,' for stock price data) announced its Q2 2026 results: revenue reached $119.8 billion, a 24% year-over-year increase; Google Cloud revenue grew 82%, and operating profit reached $8.8 billion, more than tripling year-over-year. The next day, Alphabet's stock price fell 7.13%.
A week later, Microsoft reported its quarterly results for the period ending June: revenue reached $90 billion, an 18% year-over-year increase. On July 30th, Microsoft's stock price surged over 15%, marking its largest single-day gain in 18 years, adding approximately $450 billion to its market value in one day.
Both companies are doubling down on AI, both state demand exceeds supply, and their cloud businesses maintain rapid growth. Yet, Wall Street responded with one falling and one rising stock.
The difference lies beyond the income statement. Alphabet's operating cash flow for the quarter was $39.069 billion, while capital expenditures reached $44.924 billion, pushing Free Cash Flow (FCF) down to -$5.855 billion, marking the company's first-ever negative quarterly FCF since going public.
Microsoft also made significant investments, with cash capital expenditures of $35.8 billion for the quarter and an additional $5.6 billion in finance leases. However, its operating cash flow reached $55.4 billion, a 30% year-over-year increase, leaving FCF at $19.6 billion. Microsoft's cloud platform Azure revenue grew 43% year-over-year, commercial remaining performance obligations reached $678 billion, an 84% increase, and the next quarter's Azure revenue growth guidance also exceeded market expectations.
Growth in operating cash flow, accelerating cloud revenue, and expanding order backlogs together form a relatively clear path to returns. Microsoft CFO Amy Hood stated the company has increased confidence in its return on invested capital, citing factors including a larger addressable market, improvements in model and chip efficiency, and a broadening AI product portfolio.
Of course, accounting treatment requires discernment. Shifting some data center leases from finance leases to operating leases can change where capital expenditures appear but doesn't eliminate future payment obligations.
However, this doesn't mean the market has formed a simple rule of 'stock up if FCF is positive, down if negative.'
Meta's quarterly FCF plummeted 91% year-over-year, from $8.55 billion to $784 million; $31.1 billion in capital expenditures nearly exhausted its $31.9 billion operating cash flow, and its stock initially fell 10% after-hours. Apple provides a counterexample: for its fiscal Q3 ending June 27, 2026, company revenue grew 16% to $109.4 billion, operating cash flow hit a record for the period, and its AI investment model is far lighter than cloud providers. Yet, its stock still fell over 8% the next day.
data-check-id="478352">The market worries about supply constraints, end-user demand, and future guidance. The Apple case illustrates that while free cash flow is the most prominent clue this quarter, it's not the sole trading trigger. What the market is scrutinizing is a company's ability to provide credible explanations for investment, growth, and valuation simultaneously.The same 'microscope' is also moving towards the hardware end. In June, Micron Technology reported a record $18.3 billion in adjusted free cash flow, and its stock rose 15.7% the day after earnings. By August, when SanDisk and Western Digital announced strong results and revenue guidance above analyst consensus, their stocks initially fell 13.3% and 19.1% intraday, respectively.
Demand remains strong, but questions about how long memory prices can continue rising and whether profit upgrades can catch up with prior stock gains have become new tests. The financial microscope measures cash, while the physical microscope begins measuring prices, capacity, utilization, and technological iteration.
Analysis points out that these seemingly contradictory price reactions actually point to the same change: capital expenditures aren't fully recognized in the income statement during the construction period but are first deducted from cash flow. Profits can prove the business is still growing, while free cash flow exposes the immediate cost paid for that growth; whether orders and guidance can credibly justify this cost determines how long the market is willing to tolerate it.
Simon Taylor, founder of the fintech content platform Fintech Brainfood, aptly summarized the post-earnings stock reaction for Alphabet: the sell-off expresses the market's opinion on future returns, while the backlog comes from signed contracts; between the two, only the latter is binding.
In other words, the market is not repricing 'whether there is AI demand' but rather whether the speed at which demand translates into revenue, profit, and cash can keep pace with the speed of capital investment.
Free cash flow isn't mysterious; it roughly equals cash generated from operating activities minus capital expenditures like 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 is truly left in hand 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 to a year ago, how does the company view the capital return potential and realization timeline for generative AI?
Pichai remains optimistic. He believes AI is still in the early stages of a structural shift, with opportunities for 'extraordinary returns' in both consumer services and enterprise applications, provided execution is on point.
A week later, Goldman Sachs analyst Gabriela Borges posed an almost identical question to Microsoft CFO Amy Hood: Looking at capital expenditures and commercialized revenue together, what has changed in the return on investment for the investments Microsoft is making today compared to a year ago?
Both analysts, independently, pushed the question from model capabilities, cloud revenue, and supply constraints towards return on invested capital.
This may signify that market scrutiny is deepening layer by layer. The first layer is demand: Are there customers willing to buy AI services? The second layer is revenue: Can cloud services, inference, subscriptions, and agents form scaled revenue? The third layer is cash: Can new operating cash cover chip, server, and data center expenditures? The final layer is complete capital return: Can the cash flow generated by these assets over their entire lifecycle cover depreciation, energy, financing costs, and shareholder-required returns?
Accelerating growth in the cloud businesses of Google, Microsoft, and Amazon indicates that AI isn't just about investment without customers.
Revenue is also beginning to materialize. Alphabet's cloud business margin improved noticeably, while Microsoft reports continuous improvements in model, chip, and data center efficiency. However, these companies have not yet separately disclosed complete revenue, profit, and cash flow for their AI businesses, preventing the market from precisely matching each dollar of AI capital expenditure with its corresponding return.
The cash-level stress test has just begun. Investors are distinguishing: Who can still rely on cash from existing businesses to fund construction? Whose capital expenditures are consuming cash more rapidly? And who needs to maintain expansion speed through corporate bonds, leases, and project financing?
As for the ultimate return on capital, there isn't enough long-term data to answer yet.
A data center built today might require over a decade to recoup its investment; the chips installed within could face competition from newer generations in just a few years. Even if the demand for computing power continues to grow, factors such as the price of compute, 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 begin the first round of screening.
This article is from WeChat Official Account: Economic Observer , Author: Ouyang Xiaohong
