Market Forecaster Weekly Call | Differentiation·Validation·Competition: High-probability Investment Strategies in the AI Sector

Key Takeaways: AI transitions from 'beta-driven generalized gains' to 'return validation + stock selection,' with computational power expanding from GPUs to storage/interconnect/materials/power constraints.

I. Key Conclusions:

Joe Yu's macroeconomic assessment: Weakening employment but resilient wages, tariff disruptions coupled with rising oil prices, and market risk appetite remains cautious; in the short term, a 'defensive + wait-for-catalysts' approach is more suitable, while medium-term focus should be onNVIDIA GTCandBroadcomand other key earnings validations.

Yu Dingheng’s definition of the AI cycle: The phase of 'aggressive advance' in valuation expansion ended in 2023, with the capital market now demanding that cloud providers prove their worth.CapEx can drive cloud revenue and free cash flow (FCF), the industry is entering a period of differentiation and validation.

Changes in key factors at the model level: shifting focus from 'entry subsidies/conversational capabilities' to 'token costs, engineering delivery, Agent functionality, and closed-loop scenario data'; Chinese manufacturers may gain an advantage window through the 'cost-performance route' (engineering efficiency + power/computing cost).

Structural opportunities in computing power: Yu Dingheng is more optimistic aboutstorage (especially HBM)'s supply-demand mismatch bringing vibrancy and flexibility; interconnection (optical modules/CPO) and high-frequency, high-speed materials emphasize more on 'choosing companies rather than buying sectors'; long-term constraints come fromelectricity costs and supply。

Configuration methods: Transitioning from 'allocating AI tracks' to a combination strategy of 'core positions (certainty) + flexible positions (supply-demand mismatch/technical path) + thematic speculation (security, etc.)'.

II. Macro and Market: Why More Defensive in the Short Term (Joe Yu)

Joe Yu emphasized three key points when reviewing the US stock market：Coexistence of slowing growth and sticky inflation: February employment data missed expectations, unemployment rate ticked up, but wages remained resilient, making the market more sensitive to 'stagflation-like pricing'.

Exogenous shocks raise risk premiums: Tariff expectations combined with tensions in the Middle East pushed oil prices above $90 per barrel; if oil prices rise further, it could create a dual drag on valuations and earnings.

At the trading level, investors are getting 'more picky about cash flows': During the phase where rate cut expectations are wavering, capital places greater emphasis on the quality and certainty of corporate cash flows, and sustained rebounds require catalysts for confirmation.

Strategic implications (Joe Yu’s suggested approach): In the short term, useEnergy/ResourcesImplement some defensive hedging; wait for developments in the AI sectorGTC/Critical earnings reportsReassess the pace of adding positions after implementation; on the individual stock level, he mentioned paying attention toCrowdStrike (CRWD)(compliance and security demands) andSino Biopharm (1177.HK)(BD/pipeline implementation expectations) as alternative options based on different investment theses.

III. The AI industry enters the 'validation phase': three layers of divergence are occurring (Yu Dingheng)

1) Models and applications: shifting focus from 'entry competition' to 'cost-effectiveness and functionality'

Yu Dingheng summarized the current model competition with the term 'differentiation validation'

The overseas approach emphasizes greater division of labor: OpenAI leans towards a 'super ecosystem/platformization', while Anthropic places more emphasis on enterprise services and plugin capabilities; Google, on the other hand, advances multimodal and product integrationGemini, representing an effort to promote multi-modal and product synergy

The domestic market is still experiencing diminishing marginal returns from entry subsidies: New user acquisition subsidies during the Spring Festival period led to a short-term spike in DAU, but the subsequent decline was rapid, indicating that 'subsidies alone' are increasingly unable to constitute a competitive moat

The introduction of open-source Agent frameworks, such as OpenClaw, has brought about a shift: moving from 'conversation' to 'task execution' (processes, permissions, auditing, payments/authorization, tool invocation), making it easier for enterprises to convert computing power investments into billable business value; it may also reshape the entry logic of traditional super apps

Yu Dingheng: Users may not always pursue the 'strongest model'; in more scenarios, 'good enough' suffices. Thus, the focus of competition will shift towards token costs, reasoning efficiency, and engineering delivery capabilities; second- and third-tier vendors can gain a competitive edge on the 'cost-performance curve' through distillation and engineering optimization

Cloud providers: CapEx needs to be justified by revenue and FCF

Yu Dingheng believes that the market has shifted from 'who invests more, who has greater room for imagination' to questioning:

whether massive CapEx can lead toaccelerated cloud revenue, increased inference calls, and improvements in subscription and free cash flow (FCF)；

financing capability and financial structurewill impact valuation resilience (weaker balance sheets are more likely to face market punishment first);

if the Agent ecosystem matures rapidly, cloud providers must take a more proactive approach in acquiring customers and scenario data to feed back into model and product closed-loop development.

3) Computing power: opportunities have overflowed from GPUs to storage/interconnect/materials/power

Yu Dingheng breaks down the main theme of the computing power side into four variables:

storage (HBM): He mentioned there are clear signs of shortages and price hikes emerging across the supply chain. Additionally, the demand for storage capacity and bandwidth will rise during the inference era, making storage one of the segments with the highest growth momentum. At the same time, he reminded that as the cost of storage in AI servers increases, downstream players may control costs by managing procurement cycles or adjusting configurations, which could create short-term fluctuations — storage resembles a direction with 'high momentum + strong trading attributes'.

Interconnect (Optical Modules / CPO)He believes that the optical module supply chain is concentrated in China and overall valuations are not low. Investment should focus on 'who can continuously secure core customer share, who has more comprehensive chip and production capacity, and who can expand within a multi-customer system.' He cited examples such as InnoLight Technology, Eoptolink, and Dongsan Precision, which hold different positions and customer value within the industry chain. Additionally, overseas markets are showing increasing interest in the CPO pathway.

Materials and PCBYu Dingheng used NVIDIARubinas an example to highlight that inference-driven energy efficiency optimization and higher bandwidth requirements may drivedemand for upstream components like high-frequency and high-speed PCB materials (e.g., Megtron 9 / M9) and electronic-grade low-dielectric quartz clothfor validation of demand.

Power ConstraintsHe emphasized that the 'ultimate physical limit' of computing power expansion may come from electricity costs and stable supply. Using Google and Meta's differing approaches to securing power as examples, he suggested this variable will become more prominent after 2026.

Four, Portfolio Implementation: How to turn 'differentiation validation' into investment actions (Yu Dingheng + Joe Yu)

Core positions (certainty): Yu Dingheng still categorizesNVIDIAas core assets defined as 'selling shovels in the computing power sector'; cloud platforms favor companies with strong overall capabilities, fewer weaknesses, and the ability to demonstrate Free Cash Flow (FCF).

Flexible positions (supply-demand mismatch / technological pathway)：Storage chain: More valuable during pullbacks but requires dynamic tracking of cloud vendors' capital expenditures (CapEx), server configurations, and delivery schedules.Interconnection chain selection: Make trade-offs based on market share, customer structure, chips/production capacity, and technology iteration positioning.High-frequency high-speed materials / PCB: Awaiting the transition from expectations to validation regarding technological pathways and order flows.

Application-side and security (theme shifts to monetizable): JudgmentB-end takes the lead(Cost reduction and efficiency improvement in knowledge-based roles such as coding, legal, and finance are easier to quantify);C-endFocus more on Alibaba (e-commerce/service closed-loop) and ByteDance (video generation and content production chain);AI SecurityConsidered a 'must-solve problem' (privacy, payment authorization, hallucination, compliance), better suited for lying low and waiting for valuation catalysts from landmark events or regulatory implementation.

V. Key tracking list for the next 4–8 weeks (for 'validation period' pricing)

NVIDIA GTCProduct roadmap, commercialization on the inference side, and supply chain guidance;

Key earnings/guidance from Broadcom and others;Validate the transmission of AI orders into revenue and profit;

HBM/DRAM pricing and lead times;Changes in cloud vendors' procurement pace;

Oil price and tariff developments;Impact on valuation levels and risk appetite;

Electricity/grid connection/approval and data center delivery schedules;Whether order fulfillment starts to face constraints.

136K Views