CITIC Securities: Bullish on US Tech as Core Theme, Favoring Mid-to-Downstream of the Computing Power Chain
Taylor Wilson
CITIC Securities remains bullish on US tech but favors mid- and downstream segments within the AI compute chain — the core logic being that greater compute efficiency expands total demand, making downstream beneficiaries more certain.
Why did upstream hardware surge — and what's the catch?
The strong Q2 performance in compute hardware was driven mainly by price increases, not volume growth.
South Korea's semiconductor export volume fell roughly 25% year-on-year in May, signaling that the rally is built on scarcity premiums.
This means → the trade looks structurally similar to past memory-chip cycles and commodity squeezes — once the market shifts focus from "how scarce" to "how sustainable," upstream names like Korean chip stocks face correction risk.
What is a "token budget" — and how does it cap upstream pricing power?
A token budget — the ceiling an enterprise sets for spending on AI inference — is the key variable governing end-user AI investment behavior.
AI models perform better with more compute, but most enterprise users will not chase peak performance regardless of cost.
In plain terms = companies buy AI services the way they buy utilities — useful, but if the bill gets too high they cut consumption. That cost discipline flows through to cloud providers' capex willingness, ultimately squeezing upstream hardware's pricing power.
Why does CITIC prefer mid- and downstream over upstream?
CITIC invokes the Jevons paradox — an economic observation that improving the efficiency of a resource often increases its total consumption rather than reducing it.
In plain terms = as the unit cost of compute falls, enterprises use more AI services and total demand expands — mid- and downstream players (cloud services, AI applications, model layer) benefit with greater certainty.
This reflects CITIC's core judgment: upstream expectations for simultaneous volume and price gains are already fully priced, while downstream demand expansion is only beginning to be valued.
What makes US equities more attractive than Japanese and Korean markets?
Traditional US cyclical sectors are constrained by high interest rates, but tech investment is less rate-sensitive — the tech theme remains the most compelling opportunity.
Among global tech stocks, hardware names have mostly rallied significantly; non-hardware names have lagged — CITIC sees the latter as more likely to deliver fundamental upside surprises.
This means → if the upstream supply-demand gap eases and falling token prices drive higher end-user adoption, the US tech chain could outperform Japanese and Korean markets, with allocation value versus US Treasuries still worth watching.
Where is the biggest uncertainty in this call?
Whether upstream pricing power stays under pressure depends on two factors: cloud providers' capex willingness and actual progress in end-user AI commercialization.
If AI commercialization lands slower than expected, token-budget constraints could bite earlier and harder than CITIC anticipates.
This reflects the central open question in the AI investment chain: not "is AI useful?" but "how much will enterprises pay for it?" — the answer will determine how profits are distributed across the entire value chain.
Content is for reference only, not financial advice.