Second half of the AI bull market: downstream commercialization may be more important than Capex

```

After three years of a major industry bull market, the US AI industry is facing a critical structural inflection point—the fissure between highly prosperous upstream computing hardware and lagging downstream application commercialization is becoming increasingly evident, and the direction in which this contradiction is resolved will determine whether this round of AI market enters a long-cycle expansion or comes to an end with the capital expenditure turning point.

The Zheshang Securities strategy team pointed out in their July 2 report that currently the biggest structural contradiction in the US AI industry chain is: upstream equipment and materials' prosperity depends heavily on aggressive capital expenditure expansion by midstream cloud vendors, while effective downstream application demand has not yet fully materialized. The stock prices of many representative AI application companies have fallen below or are approaching the "888-day" moving average, an important technical support level, and market worries about mid- and downstream sectors continue to ferment.

From the data, the absence of a closed commercial loop in downstream application layers is forming substantive pressure. The growth rate of Salesforce's remaining performance obligations (RPO) has dropped from 21% in 2022 to 12% in Q3 2026 and continues to decline; C3.AI's RPO has steadily decreased after the AI boom; North American cloud vendors' marginal ROI has steadily decreased since 2024, and Amazon has already fallen below breakeven.

Zheshang Securities believes the tracking priority should be: downstream commercialization data > cloud vendor Capex guidance > US Treasury rate fluctuations.

AI Year One Review: Earnings Resilience, Not Loose Liquidity, Drives Gains

To understand the current predicament, it is necessary to review the start of this AI bull market. In 2022, the US Fed’s aggressive rate hikes pushed the 10-year Treasury yield up from 1.3% at the start of the year to 4.24%, the highest since 2008. The Nasdaq index fell 34.02% for the year, with a maximum drawdown of 36.36%. Microsoft saw a cumulative annual drop of 29.32%, and Netflix's maximum drawdown exceeded 70%. At the time, the tech sector suffered a systemic "valuation kill."

However, starting in 2023, the situation reversed. Even with Treasury yields remaining over 4% for the long term, AI upstream hardware realized volume and price gains driven by rigid demand from large model training, forming molecular-side support to counter high interest rates. NVIDIA's data center business revenue soared from $2.048 billion in Q1 2022 to $193.737 billion in Q4 2026; Google Cloud's 2025 revenue was $58.7 billion, up 35.8% year-on-year. This round of tech gains was not due to loose liquidity, but was achieved by industrial earnings resilience overcoming liquidity tightening.

Zheshang Securities referenced the "888-day" moving average as an industrial cycle verification tool in their report. This indicator corresponds to about 42.67 months (888 trading days), embedding the evolution of a complete inventory cycle. History shows that when the industry index historically crosses this moving average, it often marks substantial completion of capacity clearing or significant fundamental revaluation. Zhongji Xuchuang and Industrial Fulian both broke above the 888-day average at the start of 2023, confirming a new round of prosperity for upstream computing hardware.

Upstream Divergence: Midstream Capex-Driven Boom Starts to Crack

The current high prosperity of upstream computing hardware essentially mirrors aggressive capital expenditure by midstream cloud vendors, rather than real transmission of downstream demand. This logic is now being tested.

From financial data, US and Chinese upstream companies' revenue and gross margin remain high. NVIDIA’s gross margin has stayed above 70% since Q4 2024, and most leading Chinese companies maintain gross margins above 50%. However, differentiation in stock prices has emerged—Lumentum Holdings, NVIDIA, and Broadcom's 20-day moving averages have recently turned downwards, while Applied Materials, Micron Technology, and SanDisk's 20-day averages continue to rise, showing evident contraction in trading patterns.

Midstream cloud vendors’ financials reveal deeper hidden dangers. According to Zheshang Securities, Microsoft’s capital expenditures in Q4 2025 surged to 637.54% of free cash flow, and Amazon reached a peak of 3587.91% in Q1 2026—this means for every $1 in free cash flow, cloud vendors face several or even dozens of times that in capital expenditure. Capital investment is highly detached from operating profit and heavily reliant on external financing. If the financing environment tightens, midstream expansion may converge, directly affecting upstream hardware demand.

At the same time, North American cloud vendors’ marginal ROI has continued to decline since 2024. For example, Amazon's marginal ROI has dropped from its historical high of 44.93 below breakeven, indicating incremental investment cannot guarantee profitability. There is a clear timing mismatch between hardware depreciation and demand release.

Downstream Stalling: RPO Weakness and 888 Line Warning Appear Together

Downstream application layer is the terminal outlet for the AI industry and currently the weakest link. Whether domestically or internationally, the C-end is dominated by free tools, and the B-end by enterprise pilot projects. Replicable, scalable paid scenarios have not yet matured.

RPO (remaining performance obligation) data is a leading indicator of downstream payment willingness. Salesforce’s RPO growth rate has fallen from 21% in 2022 to 12% in Q3 2026, still declining; Adobe’s RPO percentage growth has dropped fluctuant since 2022, with new contracted speed lower than RPO consumption speed; C3.AI’s RPO has fallen continuously since the AI explosion, with many projects stuck at the pilot stage. Zheshang Securities pointed out: This means corporate clients' attitudes toward long-term AI subscription services are shifting from committed investment to watch-and-wait.

Technical signals further confirm these concerns. Microsoft's share price has broken below the 888-day moving average, Netflix broke below the line on June 23, and Oracle is also testing the validity of this support line. Zheshang Securities believes this is a market confirmation that a closed loop of application layer commercialization has not yet formed, and the transmission path from "terminal demand → computing procurement → hardware profitability" is still being refined.

Risk Signal System: Macro, Industry, Demand—Three Layers of Observation

Zheshang Securities constructs a "macro—industry—demand" three-layer observation system to quantify current risk levels.

Macro layer: The 10-year US Treasury yield has stayed in the 4.65%-4.85% range, nearing the top of the box since 2023; US core CPI has exceeded expectations for three consecutive months; the probability of Fed rate cuts has dropped from 60% to less than 30%, and the dot plot has shifted toward further hikes. This combination not only curbs momentum for valuation expansion, but also prompts global long-term capital to reevaluate AI allocation weights.

Industry layer: North American cloud vendors’ combined capital expenditure growth has notably slowed, with deceleration emerging; incremental AI server shipments are slowing; the Philadelphia Semiconductor Index’s second derivative has seen a phase turning point, suggesting the industry is transitioning from rapid recovery to moderate growth—a structural inflection.

Demand layer: Global AI vendors’ forecast ARR growth continues to slow, and C-end paid ratios still need improvement. However, Zheshang Securities also noted that daily token calls in China climbed from 1 trillion at the start of 2024 to 180 trillion in June 2026, and absolute ARR values of major AI model companies remain on the rise, indicating the medium-term trend of industry prosperity still continues. Currently, it is a state where "dual-kill risk exists but the rally has not yet ended."

Two Major Scenarios: Application Landing Determines Market Endgame

Zheshang Securities presents two contrasting outlooks.

Optimistic scenario: If enterprise-level AI agents, vertical SaaS, and industry applications achieve large-scale implementation, and user payment and inference demand continue to rise, the AI industry could enter a second growth phase driven by demand. Industrial driving force would shift from supply-side to demand-side, market attention would move from capital expenditure growth rates to application revenue and user paid rates, upstream hardware prosperity would continue, and investment opportunities would spread from computing hardware to models and applications, forming a self-reinforcing positive cycle capable of offsetting the valuation pressure from high interest rates.

Pessimistic scenario: If AI application commercialization progress is slow over the next three to six months, and corporate willingness to pay and AI budget growth underperform expectations against the backdrop of persistent high US Treasury rates, the market will focus more on cloud vendors’ capital expenditure return rates. Once capital expenditure growth slows, upstream industry chain demand forecasts will be revised down, and the hardware sector faces simultaneous compression of profit forecasts and valuations. Upstream hardware will return from growth-expectation pricing to cyclical manufacturing pricing logic, and funds will concentrate on leaders with strong order certainty. Thematic concept stocks face significant valuation pullback risk.

Zheshang Securities emphasizes that the current AI group market is unlikely to end due to a single unconfirmed risk; more likely it will manifest as increased volatility. The priority order for core tracking indicators is: AI paid ARR growth rate and vertical application payment penetration, cloud vendor Capex guidance and global semiconductor sales year-on-year, US Treasury rates and Fed policy shift expectations.

Risk Disclosure & DisclaimerThe market involves risk; invest cautiously. This article does not constitute personal investment advice, nor does it account for individual users' special investment objectives, financial status, or needs. Users should consider whether any opinions, views, or conclusions herein fit their circumstances. Invest accordingly, at your own risk. ```