My view: bullish in the medium term, but it’s no longer the stage of “buy AI blindly and it always goes up.”
After AI stocks keep rallying, opportunities are shifting from “trading the concept” to “looking at performance + looking at execution + looking at scarce segments.” The next areas that are more worth watching include the supporting links for computing power infrastructure, domestic substitution, on-device AI, Agent applications, and the lines of compute-power synergy.
What investment opportunities are there as AI stocks keep rising?
1. I’m still bullish, but I want to see the “real beneficiaries”
The AI supercycle hasn’t ended, but the differentiation will be extremely severe:
There are orders, gross profit, and overseas/major-manufacturer partnerships → it still has room to move up
Just labeling something “AI concept” with no revenue → it’s easy to get hit again
If a leading stock has run up too much → see whether earnings can support the valuation
II. The next 5 directions worth paying more attention to
1) The “water-sell-and-electricity-sell” segment of compute (the steadiest)
The model arms race doesn’t stop; beyond chips, the tightest part is the supporting ecosystem:
Optical modules / CPO / 800G / 1.6T
AI server PCBs and high-speed switching
Liquid cooling, temperature control, high-power power supplies, HVDC
Data center power electronics, energy storage, and grid supporting infrastructure
Logic: compute expansion = chips + networks + electricity + cooling all expand together.
2) Domestic compute and semiconductor self-reliance (lots of elasticity)
Overseas restrictions + domestic intelligent computing center construction—benefits:
Domestic AI chips
Advanced packaging, HBM, and chip testing/assembly
Semiconductor equipment and materials
Innovative industry IT servers, domestic large-model ecosystem
Suitable for funds that can withstand high volatility.
3) On-device AI (phones/PCs/glasses/IoT)
After models get smaller and costs decline, the edge will take off:
AI smartphones, AI PCs
AR/AI glasses
On-device NPU, storage, and sensors
Lightweight models, edge-cloud collaboration
After 2026, on-device may shift from “a story” to “a replacement-cycle logic.”
4) AI applications: agents, office productivity, marketing, games, film/TV, education
Downstream applications are a direction where capital can spill over, but you must look at revenue:
Enterprise-grade agents / process automation
AI + office, tax & finance, legal services, customer service
AI + games, advertising, short dramas, e-commerce
AI for pharmaceutical R&D and industrial quality inspection
It’s not that you can’t buy the application end, but don’t buy things that “look like AI” by name—buy what can be validated by ARPU/orders/renewal rate.
5) Physical AI and new quality productivity
AI is more than software:
Humanoid robots / embodied intelligence
Intelligent driving
Industrial robots and machine vision
AI + defense industry, satellites, commercial space
These directions are volatile, but the imagination space is also big.
III. Configuration thinking (not stock-picking)
Defensive type: compute infrastructure leaders + performance delivered through liquid cooling/power supplies/optical modules
Aggressive approach: domestic AI chips + on-device AI + agent applications
Defensive: use banks/coal/high free-cash-flow assets to hedge volatility
Don’t do: thematic stocks driven only by announcements of “strategic cooperation,” with no revenue, and with high-position accelerated shrinkage in volume
IV. Risk points
Overseas cloud providers’ capex is below expectations
Semiconductor high-margin returns reverting to the mean
Application end still not making money
Leaders’ valuations have priced in too much; the sector’s high volatility leads to pullbacks
In one sentence: AI hasn’t finished its run yet, but the next phase’s money is in “performance + scarce segments + execution/implementation capability,” not in the breadth of the concept.