Storage Chip and Compute Power产业链爆发：China's Domestic Substitution and AI Devices Fuel New Quality Productive Forces
Counter-Cyclical Surge in Memory Chips and Compute Infrastructure: Validating the “New Quality Productive Forces” Logic Behind a Structural Market Rally
Against the backdrop of declining trading volume and cautious investor sentiment—exemplified by the ChiNext Index’s downward movement—April 22 saw pronounced market divergence: Dapu Micro (301332.SZ) and Tongyou Technology (300302.SZ) both hit their daily trading limits, while Hongxin Electronics (300657.SZ) and Huagong Technology (000988.SZ) also surged strongly to涨停 (daily upper limit). This counter-cyclical rally, led by memory chips and compute infrastructure stocks, is no isolated event—it reflects a confluence of multiple industrial signals. It marks a pivotal transition: domestic substitution has moved beyond policy-driven momentum into an early phase of earnings realization—and, more profoundly, affirms the real-world traction of “New Quality Productive Forces” as the core investment theme driving China’s capital markets today.
Industry Data Reinforces a Fundamental Turning Point
The most robust foundation underpinning this rally lies in accelerating, tangible industry data. According to the latest release from China’s National Bureau of Statistics, China’s integrated circuit output reached 34.2 billion units in April 2024—a 22.1% year-on-year increase, the highest monthly growth rate in the past 18 months. Notably, shipments of memory chips (DRAM/NAND Flash) and their accompanying controller ICs significantly outpaced the industry average—reflecting rising utilization rates and improved order visibility for domestic manufacturers, fueled by surging downstream demand from servers, AI PCs, and smart terminals. Crucially, this growth is not attributable to short-term inventory buildup or reliance on a single customer. Rather, it aligns closely with a 35% quarter-on-quarter increase in Q2 server shipments by leading domestic OEMs—and with the initial mass production and delivery schedules of AI laptops, such as Huawei’s MateBook X Pro AI Edition and Lenovo’s Yoga Slim 7i Gen 9.
Even more significantly, technological breakthroughs are rapidly transitioning from lab validation to engineering implementation. Recent major progress in the “Chu Ri” (Sun-Tracking) Space Solar Power Station project exemplifies this: its core energy transmission module employs domestically developed high-frequency RF chips and ultra-low-latency optical interconnect solutions—achieving microsecond-level power response at ground-based receivers. This milestone not only validates China’s self-reliance in high-speed SerDes interfaces and advanced packaging technologies (e.g., Chiplet heterogeneous integration), but also extends the performance frontier of AI compute infrastructure—from terrestrial data centers to space-ground collaborative systems. A virtuous cycle is now emerging: technological generational leaps reinforce rising localization rates, decisively overturning the prior market perception of the semiconductor sector as merely “theme-driven speculation with hollow earnings.”
Global AI Hardware Cycle Shifts Toward the Edge—Fueling Structural Incremental Demand
Elon Musk recently signaled two critical developments: first, Tesla’s FSD V12.5 has entered large-scale global road testing, compressing visual perception model inference latency to just 83 ms; second, Neuralink’s first human clinical trial participant—three months post-implantation—has achieved a mind-controlled typing speed of 23 words per minute. Together, these developments point to an irreversible trend: the global AI hardware cycle is strategically pivoting from “cloud-based training” toward “edge-side inference.”
This shift is fundamentally reshaping value distribution across the semiconductor supply chain. While cloud computing relies on high-performance GPU clusters, edge applications impose extreme requirements on low-power consumption, high energy efficiency, and ultra-low real-time latency—demanding purpose-built chips. Consequently, traditional memory chips are evolving into “compute-in-memory” architectures (e.g., memristor arrays); high-speed interface standards are rapidly advancing from PCIe 4.0 to CXL 3.0; and adoption of advanced packaging technologies—including Through-Silicon Via (TSV) and Fan-Out Panel-Level Packaging (FOPLP)—is accelerating. SEMI forecasts that China’s advanced packaging market will reach USD 18.2 billion in 2024, up 29% year-on-year, with Chiplet-based packaging serving AI endpoints accounting for over 65% of that total. Investment logic has thus evolved—from asking “Is it domestically produced?” to demanding “Can it meet the stringent technical specifications required for edge AI?”—simultaneously raising both technological barriers and commercial certainty.
The Underlying Logic Driving Capital Allocation Around “New Quality Productive Forces”
Sustained institutional buying into memory and compute infrastructure stocks reflects deep market consensus around the substance of “New Quality Productive Forces.” Distinct from traditional factor-driven growth, its essence lies in enhancing total factor productivity (TFP)—a goal realized only through capital-intensive, knowledge-intensive, and data-intensive hard-tech industries. Memory chips and compute infrastructure constitute the most fundamental “productive tools” of the digital era: without high-bandwidth, low-latency memory pools, large language models cannot dynamically access vast parameter sets in real time; without highly reliable distributed storage, petabyte-scale time-series data generated by industrial internet platforms lose analytical value.
Current capital flows reveal three distinctive patterns:
- Increased holding concentration: In Q1 2024, public mutual funds’ semiconductor holdings allocated 41% to memory controllers, high-speed interconnect chips, and advanced packaging equipment—up 12 percentage points from end-2023.
- Deeper, earlier-stage due diligence: Institutional investors have intensively visited ChangXin Memory’s Hefei facility and Cambricon’s Shanghai R&D center, focusing on concrete process metrics—such as HBM2e yield ramp-up progress and breakthroughs in 3D NAND stacking layers.
- Evolving valuation anchors: Traditional P/E ratios are losing weight; instead, investors increasingly rely on PS (Price-to-Sales) and EV/Sales, adjusted for R&D capitalization rates—precisely reflecting the long-cycle, high-barrier, R&D-intensive nature of “New Quality Productive Forces.”
Risk Warnings and Allocation Recommendations
Investors must remain clear-eyed about three key challenges facing the sector:
- Global memory price volatility: Although NAND Flash contract prices have stabilized after recent declines, sluggish recovery in consumer electronics could reignite pricing pressure in Q3.
- Geopolitical headwinds: While ongoing Pakistan-mediated talks between Iran and the U.S. signal easing tensions, uncertainty remains regarding outcomes of negotiations over Strait of Hormuz transit mechanisms—posing potential constraints on imports of advanced-process manufacturing equipment.
- Technology roadmap competition: Alternative paradigms—such as in-memory computing and photonic computing—have yet to establish definitive dominance; excessive concentration on any single path carries opportunity-cost risks.
In summary, the memory and compute infrastructure industry has entered a virtuous cycle of “policy support → technological breakthrough → order fulfillment → earnings validation.” We recommend investors focus on two strategic themes:
First, deepening domestic substitution in “choke-point” segments: prioritize memory interface IP vendors capable of mass-producing HBM2e, and OSAT leaders mastering TSV core processes.
Second, emerging sectors catalyzed by edge-AI volume ramp-up: select best-in-class suppliers of automotive-grade LPDDR5X controllers and intelligent NICs compliant with the CXL 3.0 protocol.
When capital begins paying—not for abstract narratives—but for every 0.3 TOPS improvement in compute-per-watt, “New Quality Productive Forces” ceases to be grand rhetoric and becomes a tangible, measurable growth curve on corporate financial statements.