Zheshang Securities: AI Downstream Commercialization Takes Priority Over Capex
N.R. Finch
Zheshang Securities argued on July 2 that the biggest structural risk in the U.S. AI chain is that upstream hardware momentum depends on aggressive cloud capex, while downstream apps have yet to prove they can absorb the demand — making monetization data the top tracking priority.
How weak is downstream monetization, exactly?
Salesforce's remaining performance obligations (RPO — contracted but undelivered revenue) growth fell from 21% in 2022 to 12%, and is still declining. This means → enterprise appetite for long-term AI subscriptions is cooling, not building.
C3.AI's RPO has actually dropped since the AI boom began; North American cloud providers' marginal return on investment has trended lower since 2024, with Amazon now below breakeven.
In plain terms = upstream is building compute at full speed, but the enterprises meant to pay for it are shifting from "keep investing" to "wait and see" — money is going out, but the revenue loop hasn't closed.
What is the price action signaling?
Microsoft has broken below the "888-day" moving average — a long-cycle indicator Zheshang uses to verify industrial trends. Netflix broke it on June 23; Oracle is testing it now.
This means → multiple flagship AI-application stocks approaching or losing the same long-term support at the same time is not a single-name issue. It is a market-level signal that the "end demand → compute procurement → hardware earnings" chain is still in calibration, not yet running smoothly.
Where is the cloud capex money coming from?
Microsoft's Q4 2025 capex hit 637.54% of free cash flow. Amazon's Q1 2026 figure reached 3,587.91%. In plain terms = for every dollar of free cash flow generated, these companies are spending multiples — sometimes dozens of times more — on data centers, funded almost entirely by external financing.
This reflects a deeper risk: cloud capex has decoupled from operating profit. If financing conditions tighten or downstream returns disappoint, the expansion pace could brake hard.
Can upstream hardware hold up?
Nvidia's data-center revenue surged from $2.048 billion in Q1 2022 to $193.737 billion in Q4 2026 — financial fundamentals remain strong.
But the trading picture is splitting: Nvidia, Broadcom, and Lumentum's 20-day moving averages have turned down, while Applied Materials, Micron, and SanDisk's are still rising. This means → capital within the hardware sector is narrowing its focus, no longer chasing all compute names indiscriminately.
Is the macro backdrop helping or hurting?
The 10-year Treasury yield is running between 4.65% and 4.85%, near the top of its range since 2023. Core CPI has beaten expectations for three consecutive months. The probability of a Fed rate cut has dropped from 60% to below 30%.
This means → rates staying elevated don't just suppress valuation expansion — they are forcing global long-term capital to reassess how much weight AI deserves in portfolio allocation.
What happens next?
Bull case: if enterprise AI agents and vertical SaaS enter large-scale deployment in coming months, the growth driver shifts from supply to demand. Hardware momentum extends; investment opportunities spread from compute to the model and application layers.
Bear case: if monetization progress stalls over the next three to six months and enterprise willingness to pay disappoints, the market will zero in on cloud capex returns. Once capex growth slows, upstream hardware faces simultaneous compression of earnings forecasts and valuations — repricing from growth multiples back to cyclical-manufacturing logic.
Zheshang's current read: "the risk of a double downgrade exists, but the cycle hasn't fully ended." The key checkpoint: whether AI subscription ARR growth and vertical-app paid penetration show material improvement in the months ahead. China's daily token call volume has risen from 0.1 trillion in early 2024 to 180 trillion by June 2026 — the absolute number is still growing, but what matters is whether it converts into revenue.
Content is for reference only, not financial advice.