AI sector sharply diverges: Meta leads gains, chip stocks plunge, what is the market repricing?

On July 1, 2026, Eastern Time, the U.S. stock market appeared calm on the surface—the Dow edged down 0.03%, the S&P 500 fell 0.22%, and the Nasdaq dropped 0.66%. But beneath the surface, a dramatic value reassessment was underway.

The Philadelphia Semiconductor Index plummeted over 6%. Micron Technology (MU) and SanDisk (SNDK) both fell more than 10%, Corning dropped over 13%, Intel fell 9.03%, AMD dropped 6.89%, and TSMC fell 6.98%. NVIDIA (NVDA) closed at $197.58, down 1.25%, with a market cap of $4.781 trillion.

However, the source of this sell-off—Meta (META)—soared 8.81%, closing at $612.91. Microsoft (MSFT) bucked the trend, rising 3.02% to $384.28.

The direct trigger for this divergence was the news that Meta is planning to launch a cloud infrastructure business, offering to sell its excess AI computing power externally. The market quickly interpreted this as: the narrative of the AI industry is shifting from a "capability race" to a "monetization battle."

$725 Billion in Capital Expenditure and a Fundamental Question

To understand the logic behind this sell-off, one must first grasp the scale of AI infrastructure investment.

In 2026, the combined capital expenditure of the four major tech giants—Meta, Microsoft, Alphabet, and Amazon—is expected to total approximately $725 billion, up 77% from about $410 billion in 2025. Meta alone has raised its capital expenditure guidance to between $125 billion and $145 billion.

But there is a fundamental difference between Meta and the other three: Microsoft has Azure, Google has GCP, and Amazon has AWS—their massive capital expenditures are directly offset by mature cloud service revenues. Meta does not have that. Every dollar it spends on infrastructure has been a pure cost item.

This explains an apparently anomalous phenomenon: Meta has beaten Wall Street earnings expectations for two consecutive quarters in 2026, yet its stock is still down about 7% year-to-date. The market's core question is: with $135 billion spent annually on building data centers, where is the return?

The answer Meta provided is essentially buying itself a "put option"—if internal AI monetization succeeds, all computing power is used internally; if internal consumption falls short, the excess computing power can generate revenue. If it works, it's a great innovation; if not, it can still collect rent.

The AI Industry Has Crossed the First Threshold of "Self-Sustaining"

The reason Meta's "computing power leasing" plan triggered such a strong market reaction is that it touches the core issue of the AI industry's cycle shift: from "infrastructure-driven" to "revenue realization-driven."

Exponential View, a research institution founded by renowned investor Azeem Azhar, recently released its "State of the AI Economy Report," providing key data support. The report shows that as of June 2026, the global generative AI industry (excluding China) has achieved a real annualized revenue of approximately $175 billion, with actual revenue of about $110 billion over the past 12 months.

More importantly, in the first quarter of 2026, the industry's quarterly revenue exceeded the depreciation expenses of AI infrastructure for the first time. This means that the cash flow generated by current AI operations can already cover the accounting depreciation costs of servers, GPUs, and data centers. The AI industry has crossed the first threshold of "being able to support itself."

However, this does not mean that AI investment has entered a "harvest period." The report estimates that by the end of 2026, the cumulative AI-related capital expenditure of global hyperscale cloud providers and emerging AI cloud platforms will reach about $2 trillion. The annual depreciation expenses for AI infrastructure in 2026 are expected to approach $111 billion. Although quarterly revenue can cover depreciation, cumulative revenue has not yet fully covered the depreciation pressure from historical cumulative capital investment.

Generative AI revenue still maintains a year-over-year growth rate of about 200%, roughly three times the speed of any previous IT platform upgrade, and the overall development trajectory has surpassed the early stages of the internet, cloud computing, and smartphones. According to the revenue growth curve, in 2023, it took about 180 days for the AI industry to add $1 billion in cumulative revenue; today, that process has been shortened to less than 2 days.

From "Expanding CapEx" to "Token Production and Commercialization"

In a research report released on July 2, 2026, China Merchants Securities clearly stated that the main theme of the AI industry in 2026 is shifting from "expanding CapEx" to "token production and commercialization." Market focus is moving from "who builds more GPUs and data centers" to "who can produce and monetize more tokens at lower cost and lower latency."

This shift has been fully validated in the secondary market. In a July 2 note, Goldman Sachs stated that it is too early to call the current AI trade a bubble, and the related rally is still more akin to a profit-driven bull market rather than a speculative frenzy relying solely on valuation expansion. Goldman continues to bet on companies that can directly generate revenue and profit growth from AI capital expenditure.

Huaan Securities also believes that in the first half of 2026, the global AI industry has moved from a "technology explosion period" to a "rational implementation period." Computing power supply is becoming more diversified, model capabilities are continuously iterating, application layers are beginning to realize revenue and profits, and the token economy is transitioning from a hidden cost to a visible operational variable.

A Pioneer in Revenue Realization: Microsoft's $37 Billion ARR and $627 Billion Backlog

In terms of specific progress in AI commercialization, Microsoft provides a highly valuable reference.

Microsoft's AI division has reached an annualized revenue run rate (ARR) of $37 billion. More importantly, its commercial remaining performance obligations (i.e., contract order backlog) surged 99% to $627 billion. This massive backlog ensures a highly visible revenue stream for years to come once enterprise customers lock in cloud services.

Although Microsoft's stock price has fallen as much as 22% year-to-date, its core business engine remains strong—Azure cloud services revenue growth continues at a 40% pace. Based on current valuations, Microsoft's forward P/E ratio is about 22 times, a significant discount compared to its 10-year average of 31 times. A recent survey shows that 35 analysts rate Microsoft as a "buy," with an average price target of $562.10.

Microsoft's case illustrates that the market is punishing capital expenditure models that "only spend money without making money," while rewarding business models that can convert AI investment into verifiable revenue.

Accelerated Commercialization at the Application Layer

The most intuitive manifestation of the AI industry cycle shift is the accelerating commercialization at the application layer.

Barclays estimates that the AI industry's ARR was $44 billion at the end of 2025 and is expected to reach $200 billion by the end of 2026. Citigroup predicts that global AI revenue from 2026 to 2030 will be significantly raised from the previous $2.8 trillion to $3.3 trillion.

In the enterprise AI market, revenue growth for leading companies is particularly prominent. Anthropic is expected to achieve an annualized revenue of $26 billion in 2026, while OpenAI's annualized revenue has already surpassed $25 billion. HSBC predicts that between 2026 and 2030, B2B AI industry revenue forecasts have been revised up by 74%, primarily driven by the rise of agentic AI and the continuous expansion of enterprise application scenarios.

IDC's latest survey shows that in 2026, 72% of global enterprises have deployed AI agents into production, and 51.6% have embedded agents into core business processes. AI agents are becoming the "new entry point" for enterprise software.

The differentiation path at the application layer is also becoming clearer: on one side are C-end products targeting the mass market, responsible for educating the market and vying for entry points; on the other side are B-end services targeting enterprises, which are beginning to take on more direct commercialization goals and gradually becoming the key to industry profit realization. The essence of enterprises purchasing AI services is to improve business outcomes—reducing costs, increasing efficiency, optimizing processes, and improving decision quality.

Reconstruction of the Valuation System: From "Growth Expectations" to "Profitability"

The shift in the AI industry cycle ultimately points to a deeper change: the reconstruction of tech companies' valuation logic.

In the "infrastructure-driven" phase, the market paid a "growth expectation premium" for AI concept stocks—the larger the GPU cluster built, the higher the capital expenditure, the higher the valuation. NVIDIA's valuation expansion in 2024-2025 was the ultimate expression of this logic.

But in the "revenue realization-driven" phase, the valuation anchor is shifting from "scale of capital expenditure" to "revenue quality and profitability sustainability." The reason Meta's "computing power leasing" plan was interpreted by the market as positive rather than negative is precisely because it demonstrates a commitment to financial discipline over massive capital expenditure.

This shift was vividly reflected in the market action on July 2: hardware stocks plummeted, while application and platform stocks rose against the trend. Microsoft's rise, Palantir (PLTR) surging 7.77%, and Meta's sharp rally stood in stark contrast to the chip stock crash. Capital is flowing from those who "sell shovels" to those who "dig out gold with shovels."

The AI theme is shifting toward earnings-driven performance. Many public fund management companies believe that as the industry enters the earnings realization phase, the investment logic is shifting from valuation-driven to earnings-driven.

Conclusion

The market anomaly on July 2, 2026, was not a random sector rotation but a concentrated signal release of the AI industry cycle shift.

From Meta's "computing power leasing" to AI quarterly revenue surpassing depreciation costs for the first time, from China Merchants Securities' "token production and commercialization" theme to Microsoft's $627 billion order backlog—all signals point to the same conclusion: the AI industry is moving from the first phase of "infrastructure-driven" to the second phase of "revenue realization-driven."

In this new phase, the market's core question is no longer "who built the largest data center," but "who can convert AI capabilities into sustainable cash flow." The reconstruction of the valuation system has begun: valuation pressure is building in the hardware segment, while revenue validation at the application and platform layers is becoming the new pricing anchor.

For investors, this means a fundamental change in AI investment logic—from chasing the scale of capital expenditure to tracking the quality of revenue realization. AI's "money-making ability" is becoming the core pricing variable in this industry cycle.

FAQ

Q1: What does it mean that AI industry "quarterly revenue exceeded depreciation costs for the first time"?

This is a key indicator that the AI industry is moving from a "burning cash stage" to a "self-sustaining stage." As of the first quarter of 2026, the cash flow generated by AI operations can already cover the accounting depreciation costs of servers, GPUs, and data centers. However, there is still a gap to fully recover all historical investments, with cumulative capital expenditure at about $2 trillion and annual depreciation at about $111 billion.

Q2: Why did Meta selling computing power trigger a chip stock sell-off?

The market interpreted Meta's move as a signal that AI infrastructure capital expenditure may have peaked. If hyperscalers start selling excess computing power externally instead of continuing to purchase new hardware, the supply-demand dynamics for GPUs, memory chips, and other hardware could reverse. This directly triggered a revaluation of future earnings expectations for semiconductor stocks.

Q3: What does the shift in AI investment logic from "infrastructure" to "application layer" mean?

It means the market's pricing anchor is shifting from "scale of capital expenditure" to "quality of revenue realization." The valuation premium for hardware is narrowing, while companies at the application and platform layers that can convert AI capabilities into verifiable revenue are being repriced. China Merchants Securities calls this a shift from "expanding CapEx" to "token production and commercialization."

Q4: What types of companies are more likely to benefit in the second phase of AI commercialization?

Companies with clear AI revenue models are more likely to benefit, including: cloud platforms with a large enterprise customer base (e.g., Microsoft Azure), software companies that can embed AI capabilities into core business processes, and leading AI companies that have validated paths for large model commercialization (e.g., OpenAI, Anthropic). B-end scenarios, due to stronger willingness to pay and verifiable results, are seen as the main battlefield for large-scale AI application deployment.

Q5: What implications does the second phase of AI commercialization have for the crypto industry?

The integration of AI and the crypto industry is deepening. The Gate platform has launched a "Gate for AI" infrastructure layer, integrating AI into core areas such as trading, risk management, and data analysis. AI agents are shifting from information retrieval to executing economic activities—calling paid APIs, executing on-chain transactions, and purchasing computing resources. AI commercialization logic applies equally to the crypto space: whoever can convert AI capabilities into verifiable on-chain revenue will gain valuation repricing in the new cycle.