Chang Jingwen Column: Earnings Up, Stock Prices Down – Is AI Real, But Are the Prices?

"Earnings up, stock prices down." This is the strangest scene in the latest U.S. earnings season. Market sentiment appears to be changing — capital markets are no longer asking "Does AI work?" but instead starting to settle the accounts: "After spending so much money, who will actually make profits?"

This isn't denying the revolutionary nature of AI technology, but posing a sharper question: When major tech giants are setting new records in AI investment, are current high stock valuations overly optimistic — or even completely ignoring the long timeframes and uncertainties required to recoup these investments?

Investor Shift: From Growth to Cash Flow

For the past two years, whenever a company announced increased GPU purchases or data center construction, its stock price would typically rise immediately. But judging from Q2 2026 earnings reports, the rules of the game have changed.

First, companies with unlimited spending are being punished. Google's parent company Alphabet recorded its first-ever quarterly "negative free cash flow" since going public. Meta's cash flow plummeted from $8.5 billion the previous year to under $800 million. Amazon's cash flow also approached negative territory. Oracle's capital expenditures reached 174% of its operating cash flow. Stocks of these companies, which continue aggressive spending, have all come under pressure.

Second, even companies with strong revenue but excessive spending have seen stock declines. The latest examples are AMD and SpaceX.

AMD reported stellar results: data center revenue doubled, and overall revenue grew 50% year-on-year. Yet its stock plunged after hours, primarily due to excessively high capital expenditures and Elon Musk's explicit statement during the SpaceX earnings call that future AI infrastructure would prioritize NVIDIA, severely undermining AMD's position as a challenger.

Meanwhile, SpaceX released its first post-IPO financial report: Q2 revenue reached $7.8 billion (92% YoY growth), adjusted EBITDA hit $3.5 billion (nearly tripled year-on-year), and net loss narrowed to $541 million — all exceeding expectations. However, due to massive expansion of Starlink, advancement of Starship, and AI infrastructure development, quarterly capital expenditures soared to $18.4 billion. Combined with upcoming insider share unlocks, volatility risk from Bitcoin holdings, and warnings from JPMorgan that annual capex could approach $200 billion between 2027 and 2028 — continuously eroding free cash flow — SpaceX's stock dropped 13.61% on Wednesday, closing at $108.27.

The main reason for both companies' stock declines was capital expenditures far exceeding expectations, with future outlooks insufficient to justify previously inflated market expectations.

Notably, companies focused on efficiency have shown relative resilience. Microsoft, which invests more conservatively and can demonstrate AI-driven revenue acceleration, received better market evaluations.

This signals a clear shift in market sentiment: investors no longer just ask "How impressive are the revenue numbers?" but are now demanding: "When will the money spent turn into actual cash returns?"

Data Center Accounting: Payback Period May Exceed Equipment Lifespan

The cost of building data centers is staggering, with a significant gap between setup costs and revenue generation. According to NVIDIA CEO Jensen Huang's estimates, a large-scale AI data center can cost between $80 billion and $100 billion. However, industry estimates suggest such facilities generate only about $10 billion to $12 billion in annual computing revenue.

Additionally, the payback period is extremely long. Rough calculations indicate it would take 8 to 10 years just to break even, yet current chips and hardware may require upgrades within a few years — meaning equipment becomes obsolete before full depreciation. Rising electricity, cooling, and memory costs further reduce the actual computing power per dollar spent.

The Bank for International Settlements (BIS) used the term "AI exuberance" in its June report, warning that this mirrors historical bubbles like canals, railways, and the internet — where excessive optimism drives capital inflows, but final returns fall short of expectations.

Technology may be useful, but corporate adoption is another matter entirely.

Recent studies show that approximately 95% of generative AI pilot projects fail to bring tangible improvements to corporate bottom lines. Particularly, return on investment (ROI) remains low. Surveys indicate that over half of enterprises find AI benefits do not outweigh costs. In India, only 12% of companies with large-scale AI adoption can prove economic benefits.

Moreover, corporations are beginning to cut costs. Major companies like Uber have publicly stated that AI costs are "increasingly difficult to justify," with some limiting usage or canceling expensive licenses, demanding stricter cost reviews.

From "Blind Belief" to "Merit-Based Performance"

AI development and application are undeniably real. What's now being questioned is the "price" — whether stock valuations have prematurely priced in future profits.

As total AI investments by major tech giants surpass $700 billion, straining cash flows while enterprise profitability lags behind expectations, markets are naturally reassessing risks.

This doesn't mean the end of AI development, but rather a shift in investment logic: in the coming quarters, stock prices will no longer be determined by "who spends the most on equipment," but by "who can turn investments into stable profits and cash flow."