Equity Research: Power Equipment & AI Infrastructure

Date: March 1, 2026
Sector: Power Equipment / Electrical Infrastructure
Rating: OVERWEIGHT (Maintained)

Executive Summary

The Chinese power equipment sector is undergoing a structural revaluation, driven not merely by traditional grid expansion but by the explosive convergence of Artificial Intelligence (AI) infrastructure and energy demand. Our analysis indicates a pivotal shift in the domestic AI landscape: the narrative has transitioned from "passive substitution due to external sanctions" to "active selection driven by superior cost-performance and surging global demand."

Data from late February 2026 reveals that Chinese Large Language Models (LLMs) are experiencing exponential growth in token consumption, outpacing their US counterparts in weekly volume. This surge is underpinned by a significant cost advantage—Chinese models offer inference costs nearly 10x lower than leading Western alternatives like Claude Opus 4.6. This "high value-for-money + scaled application" model is driving massive demand for domestic computing power, which in turn acts as a primary catalyst for the Intelligent Data Center (IDC) and high-voltage power equipment sectors.

Key developments supporting our Overweight rating include:
1. Demand Supercycle: Weekly token calls for Chinese models reached 5.16 trillion in the week of Feb 16-22, a 127% increase over three weeks, surpassing the US volume of 2.7 trillion.
2. Infrastructure Bottlenecks & Pricing Power: Severe supply constraints in GPU availability have led to queue times of up to 8 hours for video generation services and prompted price hikes by major providers (e.g., Zhipu AI). This scarcity supports higher utilization rates and improved profitability across the IDC and server supply chain.
3. Technological Shift to Liquid Cooling: The mandate for Power Usage Effectiveness (PUE) ≤ 1.15 in new large-scale data centers by 2026 is accelerating the transition from air to liquid cooling. We project liquid cooling penetration to rise from 8% in 2024 to 35% in 2026, creating a high-value upgrade cycle for thermal management, power distribution, and high-density server components.
4. Grid Intensification: The shift toward high-power density computing is increasing the electrical load per rack and per park, directly benefiting manufacturers of transformers, switchgear, High-Voltage Direct Current (HVDC) systems, and energy storage solutions.

We maintain a Buy rating on key beneficiaries including Liangxin Shares (002706.SZ), Sifang Shares (601126.SH), and Jinpan Technology (688676.SH), while highlighting Runze Technology (300442.SZ) and Dawei Technology (600589.SH) as critical plays in the IDC expansion theme.

Key Takeaways

1. The "Token Export" Phenomenon: Chinese AI Models Go Global

The most significant development in the AI sector this quarter is the rapid internationalization of Chinese LLMs, evidenced by robust API call volumes on global aggregation platforms.

• Volume Surge: According to OpenRouter data, the weekly token invocation volume for Chinese models surged to 5.16 trillion during the week of February 16–22, 2026. This represents a 127% growth over a three-week period. Notably, this volume exceeded the concurrent weekly usage of US models, which stood at 2.7 trillion tokens.
• Market Dominance: Chinese models now occupy four of the top five spots in global invocation rankings. The combined contribution of MiniMax M2.5, Moonshot AI’s Kimi K2.5, Zhipu AI’s GLM-5, and DeepSeek V3.2 accounts for 85.7% of the total volume among the top five models globally.
• Commercial Validation: The monetization potential is becoming evident. Moonshot AI’s Kimi generated revenue in just 20 days that exceeded its entire full-year total for 2025. Similarly, ByteDance’s Seedance 2.0 video generation model faces severe capacity constraints, with over 1,000 users in the queue and wait times extending to 8 hours. This underscores a reality where end-user demand (both B2B and B2C) significantly outstrips current computing supply.

2. Cost Arbitrage: The Engine of Adoption

The primary driver behind this adoption spike is a stark cost advantage derived from the规模化 (scaled) deployment of domestic computing clusters.

• Input Efficiency: Chinese models offer input processing at $0.30 per million tokens, compared to $5.00 for Claude Opus 4.6—a 16x cost advantage.
• Output Efficiency: Output costs range from $1.10 to $2.55, whereas Claude Opus 4.6 charges $25.00—nearly a 10x difference.
• Strategic Implication: This cost structure allows Chinese developers and enterprises to deploy AI applications at a scale previously economically unviable. The "High Cost-Performance + Scaled Application" model is creating a virtuous cycle: lower costs drive higher volume, which justifies further investment in domestic GPU clusters, further lowering unit costs.

3. Infrastructure Bottlenecks Driving Pricing Power

The surge in demand has exposed critical bottlenecks in the supply of high-end computing power, shifting the industry dynamic from a buyer’s market to a seller’s market.

• Supply Constraints: Zhipu AI has initiated phased price increases and launched a "Computing Power Partner" recruitment program to alleviate resource bottlenecks. This confirms that computing scarcity is now the core variable limiting the commercialization speed of AI models.
• Profitability Improvement: In this tight supply environment, IDC operators and server manufacturers are gaining pricing power. We anticipate that rising utilization rates and modest price hikes will support margin expansion across the产业链 (industry chain), particularly for companies with existing capacity and access to domestic GPU supplies.
• Short-Term Outlook: The domestic computing power gap is unlikely to be filled in the short term. Therefore, the combination of price increases and high utilization rates will provide strong support for industry profits throughout 2026.

4. The Liquid Cooling Imperative: From Optional to Standard

As AI workloads become more intensive, the thermal design power (TDP) of servers is rising sharply. This necessitates a fundamental shift in data center cooling architectures.

• Regulatory Push: New regulations mandate that the Power Usage Effectiveness (PUE) of newly built large and ultra-large data centers must be ≤ 1.15 by 2026. Traditional air-cooling systems struggle to meet this efficiency target under high-load AI scenarios.
• Penetration Forecast: Consequently, liquid cooling is transitioning from an "optional" feature to a "standard" configuration. We estimate that the penetration rate of liquid cooling solutions will skyrocket from 8% in 2024 to 35% in 2026.
• Value Chain Impact: This transition boosts the value content per rack for:
  • Thermal Management: Cold plates, immersion cooling tanks, and coolant distribution units (CDUs).
  • Power Systems: High-efficiency power supply units (PSUs) and busbars.
  • Components: High-layer count PCBs and optical modules required for high-speed interconnects in dense server configurations.

5. Power Equipment: The Hidden Beneficiary of AI

The narrative for power equipment is evolving from "grid modernization" to "AI-enabling infrastructure." The shift of domestic computing from a "supplementary role" to a "main platform" means that AI data centers are becoming major industrial loads.

• Load Intensity: High-power density servers significantly increase the power consumption per cabinet and the overall electrical intensity of industrial parks.
• Equipment Demand: This drives incremental demand for:
  • Transformers: Especially high-efficiency, low-loss units capable of handling fluctuating AI loads.
  • Switchgear & HVDC: For reliable power distribution and long-distance transmission of renewable energy to data hubs.
  • Energy Storage: To manage peak shaving and ensure uninterrupted power supply (UPS) for critical AI workloads.
  • Distribution Automation: For real-time monitoring and optimization of complex microgrids within data center parks.

6. Sector Performance & Market Sentiment

• Relative Strength: The Power Equipment sector (Shenwan Index) demonstrated resilience and outperformance in early 2026.
  • 1-Month Return: +5.3% (vs. CSI 300: +0.1%)
  • 3-Month Return: +10.3% (vs. CSI 300: +2.9%)
  • 12-Month Return: +51.6% (vs. CSI 300: +21.1%)
• Recent Weekly Performance: In the week of Feb 23–27, the sector rose 1.89%, ranking 13th out of 28 first-level industries. It outperformed the CSI 300 by 0.81 percentage points.
• Top Performers: Stocks linked to smart grid and high-voltage technologies led gains, including Anke Zhidian (+27.54%), Beijing Kerui (+26.75%), and Hangdian Shares (+25.09%), reflecting investor focus on grid flexibility and AI-related power needs.

Industry Dynamics & Cross-Sector Analysis

While the primary thesis focuses on AI-driven power demand, we monitor adjacent sectors for broader macro-industrial signals.

A. Photovoltaic (PV) Industry: Consolidation and Price Bottoming

The PV sector is currently in a phase of inventory digestion and price stabilization following a period of aggressive expansion.

• Policy Headwind: On February 10, 2026, the US Department of Commerce issued affirmative final determinations in the second sunset review of anti-dumping and countervailing duties on crystalline silicon PV products from China and Taiwan.
  • China Dumping Margin: 165.04%
  • China Subsidy Rate: 29.72% – 41.57%
  • Implication: Trade barriers remain high, reinforcing the need for Chinese manufacturers to focus on domestic demand and non-US overseas markets (Europe, Middle East, Africa).
• Installation Data: In 2025, China added 317 GW of new PV capacity (164 GW centralized, 153 GW distributed), a 14% YoY increase. Total installed capacity reached 1.2 TW, with PV generation exceeding wind power for the first time (1.17 TWh, +40% YoY).
• Price Trends (Feb 2026):
  • Polysilicon: Prices remain depressed. Dense material trades at RMB 42-49/kg. Manufacturers are cutting operating rates in Q1 to manage inventory, with global monthly output expected to hover around 80-100k tons.
  • Wafers: Prices are nearing cyclical lows. 183N wafers traded at ~RMB 1.08/piece; 210N at ~RMB 1.40/piece. Volume has increased slightly as manufacturers clear inventory pre-holiday.
  • Cells & Modules: N-type cell prices dropped to ~RMB 0.44/W. Module prices for distributed projects are quoted at RMB 0.80-0.88/W, with transaction prices around RMB 0.75-0.80/W. Overseas TOPCon module averages are ~$0.096/W.
• Outlook: We expect prices to stabilize in Q2 2026 as production cuts take effect and seasonal demand picks up ("Golden March, Silver April"). However, immediate upside is limited by high inventory levels.

B. Energy Storage: Robust Overseas Expansion

Despite PV headwinds, the energy storage sector shows strong momentum, particularly in overseas markets.

• January 2026 Orders: 27 Chinese enterprises signed or completed 36 overseas projects/orders, totaling 29.16 GWh.
• Regional Breakdown:
  • Europe: 17.62 GWh (Led by UK deals).
  • Africa: 6.15 GWh (Led by Egypt deals).
  • Asia/South America/Oceania: Each >1 GWh.
• Key Winners:
  • CATL: Signed a 10 GWh MOU with European partners (including Schroders Green Investment), dominating the monthly rankings.
  • Chuneng New Energy: Secured a 6 GWh strategic agreement in Egypt.
  • Sungrow: Achieved 4.12 GWh in orders/cooperations.
  • Trina Storage: Delivered and signed nearly 4 GWh across South America, Europe, and APAC.
• Investment Implication: The globalization of Chinese storage firms is accelerating. Companies with strong overseas channels and brand recognition (CATL, Sungrow, Trina) are capturing high-margin markets, offsetting domestic competition.

C. Robotics: Early Stage Industrialization

• UBS Forecast: Global humanoid robot demand is projected to reach 30,000 units in 2026, with mass adoption likely in 2027-2028.
• Tesla Optimus: Plans for a 1 million-unit annual production line are on track for late 2026 completion.
• Xiaomi Robotics: Released the open-source Xiaomi-Robotics-0 VLA (Vision-Language-Action) model (4.7 billion parameters), achieving SOTA results in simulation and real-world tasks. This highlights the rapid progress in embodied AI, which will eventually drive additional power and compute demands for training and edge inference.

Investment Logic & Strategic Recommendations

Our investment strategy is anchored in the conviction that AI Infrastructure is the most deterministic growth vector in the current market. The logic flows from Application Demand -> Compute Scarcity -> IDC Expansion -> Power Equipment Upgrades.

Core Thesis: From "Substitution" to "Active Choice"

Previously, investments in domestic computing were driven by the necessity to replace banned foreign chips. Today, the driver is economic superiority. Chinese models are cheaper, faster to iterate, and increasingly competitive in performance. This organic demand creates a sustainable, long-term capex cycle for infrastructure, unlike the one-off replacement cycles of the past.

Sector Allocation Strategy

1. Intelligent Data Centers (IDC)

• Logic: Direct beneficiaries of token volume growth. As model providers scramble for capacity, IDC operators with available racks, particularly those equipped for high-density and liquid cooling, will see improved occupancy rates and rental yields.
• Key Catalysts: Acceleration of domestic AIDC tenders; PUE regulatory compliance driving retrofitting and new builds.
• Top Picks:
  • Runze Technology (300442.SZ): A leader in large-scale data center operations with a strong footprint in core hub regions. Its scale allows it to absorb capital expenditures for liquid cooling upgrades more efficiently.
  • Dawei Technology (600589.SH): Positioned to benefit from the initial wave of AIDC construction and server hosting services.
  • Kehua Data (002335.SZ): Strong presence in UPS and data center infrastructure, offering a integrated solution play.

2. High-Voltage & Power Distribution (HVDC/SST)

• Logic: AI data centers are essentially massive, concentrated electrical loads. They require robust, high-capacity power entry and distribution. The trend toward Higher Voltage Direct Current (HVDC) and Solid State Transformers (SST) is driven by the need for efficiency and space savings in dense urban or park environments.
• Key Catalysts: Grid approval of new UHV lines (e.g., Dalat-Mengxi 1000kV); increased capex by State Grid on distribution networks (+120% YoY investment in Feb 2026).
• Top Picks:
  • Jinpan Technology (688676.SH): A global leader in dry-type transformers, with significant exposure to data center and new energy sectors. Its digital manufacturing capabilities allow for rapid customization for AI park requirements.
  • Sifang Shares (601126.SH): A dominant player in relay protection and automation. As grids become more complex with bidirectional flows from storage and renewables powering data centers, advanced automation software and hardware become critical.

3. Low-Voltage Electrical Components

• Logic: Within the data center and downstream distribution network, the reliability of circuit breaking and control is paramount. High-frequency switching and high loads require premium low-voltage appliances.
• Top Pick:
  • Liangxin Shares (002706.SZ): Specializes in high-end low-voltage electrical appliances. Its products are essential for the final stage of power distribution in sensitive IT environments. The company benefits from the "quality over price" trend in critical infrastructure.

Financial Analysis & Valuation

We analyze the valuation metrics of our core recommended stocks against their earnings growth trajectories. The market is currently pricing in significant growth expectations for 2025-2026, reflecting the anticipated AI infrastructure boom.

Valuation Table: Key Coverage Universe

(Source: Wind, Huaxin Securities Research. Note: Unrated earnings forecasts are based on Wind consensus estimates.)

Detailed Company Analysis

1. Liangxin Shares (002706.SZ) – BUY

• Investment Thesis: Liangxin is the premier domestic supplier of high-end low-voltage electrical appliances. Its customer base includes leading data center operators, telecommunications giants, and industrial automation firms.
• Financial Outlook: We forecast EPS to grow from RMB 0.28 in 2024 to RMB 0.44 in 2026, representing a CAGR of ~25%.
• Valuation: Trading at 26.2x 2026E PE, the stock offers a reasonable entry point given its defensive characteristics and steady growth in the AI infrastructure niche. The multiple compression from 41x (2024) to 26x (2026) reflects the market's expectation of earnings normalization and volume growth.
• Catalyst: Increased specification of smart breakers in new AIDC designs; market share gains in the nuclear power and high-end manufacturing sectors.

2. Jinpan Technology (688676.SH) – BUY

• Investment Thesis: Jinpan is a global leader in dry-type transformers, a critical component for data centers due to their safety and fire-resistance profiles. The company is aggressively expanding its overseas presence, particularly in Europe and the US, where grid modernization is also underway.
• Financial Outlook: EPS is projected to rise from RMB 1.26 (2024) to RMB 2.20 (2026), a robust 32% CAGR. This growth is driven by both volume expansion in the renewable/data center sector and product mix improvement (higher margin digital transformers).
• Valuation: At 45.8x 2026E PE, Jinpan commands a premium due to its high growth visibility and export exposure. However, given the secular tailwinds in global grid investment and AI power needs, this premium is justified.
• Catalyst: Execution of large overseas orders; successful rollout of energy storage integrated transformer solutions.

3. Sifang Shares (601126.SH) – BUY

• Investment Thesis: Sifang is a cornerstone of China’s power grid automation. As the grid integrates more volatile renewable sources and concentrated AI loads, the need for advanced relay protection, substations automation, and flexible AC/DC transmission control grows.
• Financial Outlook: Steady EPS growth from RMB 0.86 (2024) to RMB 1.19 (2026). While growth is slower than Jinpan, it is highly predictable and backed by State Grid’s consistent capex.
• Valuation: Trading at 37.6x 2026E PE. The stock offers a balance of growth and stability. Its involvement in HVDC and flexible interconnection projects (like the Hainan-Guangdong link) positions it well for the next phase of grid development.
• Catalyst: Acceleration of UHV project approvals; increased spending on distribution network automation.

4. Runze Technology (300442.SZ) – Monitor / High Growth Potential

• Investment Thesis: Runze operates some of the largest hyperscale data centers in China. It is a direct proxy for AI compute demand.
• Financial Volatility: Note the significant jump in EPS from 1.04 (2024) to 2.77 (2025E), followed by a drop to 1.94 (2026E) in consensus estimates. This volatility may reflect one-time gains, accounting adjustments, or heavy depreciation cycles from new builds. Investors should scrutinize the quality of earnings and free cash flow.
• Valuation: The high PE ratios (37x-53x) reflect the market’s willingness to pay for scarcity in high-quality IDC assets.
• Risk: High leverage associated with rapid expansion; potential margin pressure if electricity costs rise without corresponding rent increases.

5. Dawei Technology (600589.SH) – Speculative Turnaround

• Investment Thesis: Dawei is emerging from a loss-making position (EPS -0.01 in 2024) to projected profitability (EPS 0.07 in 2026). It represents a turnaround play in the IDC/server space.
• Valuation: Extremely high PE multiples (185x in 2026E) indicate that the stock is priced on future growth potential rather than current earnings. It is suitable for risk-tolerant investors betting on the early-stage explosion of AIDC services.
• Risk: Execution risk in turning profitable; intense competition in the server hosting market.

Risks / Headwinds

While the outlook is positive, institutional investors must consider the following risks:

1. Technological Development Risks

• AI Model Iteration Slowdown: If the pace of algorithmic efficiency improvements slows, the demand for raw compute power might plateau. However, current trends suggest the opposite (scaling laws still holding).
• Liquid Cooling Adoption Delays: If technical issues with leakage or maintenance complexity hinder the widespread adoption of liquid cooling, the expected value uplift for thermal management suppliers may be delayed.

2. Competitive Intensification

• IDC Oversupply: There is a risk that excessive capital inflow into IDC construction could lead to localized oversupply, particularly in non-core regions, pressuring rental yields.
• Price Wars in Power Equipment: As more players enter the high-voltage and transformer markets attracted by AI margins, competition could erode gross margins for companies like Jinpan and Liangxin.

3. Macro & Systemic Risks

• Grid Connection Bottlenecks: The rapid addition of AI loads and renewable generation may outpace the grid’s ability to transmit power, leading to curtailment or connection delays for new data centers.
• Interest Rate Environment: High capital intensity in IDC and power infrastructure makes these sectors sensitive to interest rate fluctuations. Higher financing costs could dampen ROI for new projects.
• Geopolitical Trade Barriers: Further restrictions on semiconductor exports or tariffs on Chinese electrical equipment (similar to the PV sector) could impact supply chains or overseas revenue streams for companies like Jinpan.

4. Company-Specific Execution Risks

• Earnings Misses: The high valuations of stocks like Runze and Dawei leave little room for error. Any miss in quarterly earnings or guidance could result in significant multiple contraction.
• Raw Material Price Volatility: Fluctuations in copper, aluminum, and silver prices (impacting PV and electrical components) can squeeze margins if not fully passed through to customers.

Rating / Sector Outlook

Sector Rating: OVERWEIGHT (Recommended)

We maintain our Overweight rating on the Power Equipment sector. The convergence of AI-driven compute demand and energy infrastructure upgrades provides a dual-engine growth model that is rare in mature industrial sectors.

• Short-Term (1-3 Months): Expect volatility as the market digests Q4 2025 earnings and Q1 2026 guidance. However, any pullback should be viewed as a buying opportunity, particularly for names with direct AI exposure (IDC, Liquid Cooling).
• Medium-Term (6-12 Months): The fundamental drivers (token growth, PUE mandates, grid capex) will strengthen. We expect the sector to continue outperforming the broader CSI 300 index, driven by earnings visibility and policy support.
• Long-Term (1-3 Years): The structural shift of AI from "experiment" to "utility" will cement the role of power equipment providers as critical enablers of the digital economy. Companies that successfully integrate digital technologies (smart grids, automated manufacturing) into their offerings will command sustained premium valuations.

Investment Themes to Watch

1. "AI-Power" Nexus: Companies that explicitly bridge the gap between computing and energy (e.g., integrated DC power solutions, microgrid controllers for data centers).
2. Overseas Expansion: Manufacturers with proven success in exporting to Europe, the Middle East, and Southeast Asia, diversifying away from domestic cyclicality.
3. Green Computing: Solutions that directly contribute to PUE reduction, including advanced liquid cooling, waste heat recovery, and renewable energy integration for IDCs.

Investment View & Conclusion

The data from February 2026 confirms a paradigm shift in the Chinese technology and industrial landscape. The explosion in token usage by Chinese LLMs is not a transient hype cycle but a structural change in how digital services are consumed and produced. This demand is physically manifesting in the form of server racks, cooling pipes, and high-voltage cables.

For institutional investors, the implication is clear: The AI revolution is an infrastructure revolution.

While software and algorithm companies capture headlines, the Power Equipment and IDC sectors offer a more tangible, measurable, and arguably less risky pathway to participating in this growth. The moats in this sector are built on engineering expertise, manufacturing scale, and regulatory compliance (PUE standards), which are harder to disrupt than pure software code.

We recommend constructing a portfolio that balances high-growth IDC operators (Runze, Dawei) with high-certainty power equipment leaders (Jinpan, Sifang, Liangxin). This barbell approach captures the upside of the AI boom while mitigating risk through exposure to essential, regulated utility-like infrastructure assets.

The transition of domestic computing from a "backup" option to the "primary platform" is complete. The subsequent build-out of the physical layer—power and cooling—is just beginning. We advise investors to accumulate positions in these foundational assets as the market continues to recognize the depth and durability of this demand supercycle.

Appendix: Detailed Industry Data & Charts Reference

(Note: The following details are derived from the source report's data sections for reference.)

A. Market Performance Metrics (Week of Feb 23-27, 2026)

• Sector Return: +1.89% (Rank 13/28)
• Outperformance: +0.81% vs. CSI 300
• Top Gainers:
  1. Anke Zhidian: +27.54%
  2. Beijing Kerui: +26.75%
  3. Hangdian Shares: +25.09%
  4. Jiawei New Energy: +24.07%
  5. Hanlan Shares: +22.01%
• Top Losers:
  1. Penghui Energy: -8.84%
  2. Fangzheng Motor: -6.73%
  3. Kedali: -6.60%
  4. CATL: -6.39%
  5. Zhaowei Electromechanical: -6.20%

B. Photovoltaic Price Tracking (Feb 2026)

C. Overseas Storage Orders (Jan 2026 Snapshot)

• Total Volume: 29.16 GWh
• Leading Region: Europe (17.62 GWh)
• Leading Company: CATL (10.03 GWh)
• Key Deal: CATL x Schroders Green Investment (UK) - 10 GWh MOU.
• Emerging Market: Africa (6.15 GWh), led by Chuneng New Energy in Egypt (6 GWh).

Analyst Certification & Disclosures

Analysts:
* Fu Honghao (S1050521120004): Chief Analyst, Power Equipment. M.Eng, Chinese Academy of Sciences. 6 years of new energy research experience.
* Zang Tianlv (S1050522120001): CFA, FRM. M.Fin, Shanghai Jiao Tong University. 4 years of financial research experience, covering PV and Energy Storage.

Certification:
The analysts named in this report hereby certify that all of the views expressed herein accurately reflect their personal views about the subject securities or issuers. No part of the compensation of any of the analysts was, is, or will be directly or indirectly related to the specific recommendations or views expressed in this report.

Disclaimer:
This report is prepared by Huaxin Securities Co., Ltd. for informational purposes only and does not constitute an offer to sell or a solicitation of an offer to buy any securities. The information contained herein is based on sources believed to be reliable, but Huaxin Securities makes no representation or warranty, express or implied, as to its accuracy or completeness. Investors should conduct their own independent assessment and consult with professional advisors before making any investment decisions. Past performance is not indicative of future results.

Strategic Deep Dive: Why Power Equipment is the "Pick-and-Shovel" Play of AI

To further elaborate on the investment thesis for institutional clients, we dissect the specific mechanisms by which AI drives power equipment value.

1. The Physics of AI Compute: Power Density Matters

Traditional data centers operated at power densities of 4-6 kW per rack. Modern AI training clusters, utilizing high-end GPUs (such as Huawei Ascend or NVIDIA H-series equivalents), operate at 20-40 kW per rack, with next-generation architectures pushing towards 100 kW per rack.

• Implication for Transformers: Standard distribution transformers cannot handle such concentrated loads without significant overheating and efficiency losses. This necessitates the deployment of high-efficiency, cast-resin dry-type transformers (Jinpan Technology’s specialty) which offer better thermal management and safety in confined spaces.
• Implication for Switchgear: The rapid fluctuation in power draw during AI training spikes requires faster-acting, more intelligent switchgear and circuit breakers (Liangxin Shares) to prevent tripping and ensure uptime.

2. The PUE Regulatory Cliff

The Chinese government’s mandate for PUE ≤ 1.15 is not a suggestion; it is a hard constraint for new large-scale projects.

• Air Cooling Limit: Traditional air cooling typically achieves a PUE of 1.4-1.6. It is physically impossible to reach 1.15 with air cooling in high-density environments.
• Liquid Cooling Solution: Liquid cooling (cold plate or immersion) can achieve PUEs of 1.1-1.2.
• Market Size Calculation: If 35% of new data centers adopt liquid cooling by 2026, and the average value of liquid cooling infrastructure per rack is 3-5x that of air cooling, the total addressable market (TAM) for thermal management expands disproportionately to the number of racks. This is a value-multiple expansion, not just a volume expansion.

3. The Grid Edge: Microgrids and Storage

AI data centers are increasingly being paired with on-site energy storage and renewable generation to manage costs and ensure reliability.

• Peak Shaving: Electricity prices often vary by time of day. AI training jobs can be scheduled flexibly. Energy storage systems (ESS) allow data centers to charge batteries when power is cheap and discharge during peak hours. This drives demand for PCS (Power Conversion Systems) and integrated ESS solutions (Sungrow, CATL).
• Grid Stability: The sudden load changes from AI clusters can destabilize the local grid. Advanced SVG (Static Var Generators) and APF (Active Power Filters) are required to maintain power quality. Sifang Shares and similar automation firms provide the control logic for these systems.

4. The "Domestic Substitution" 2.0

In the previous cycle (2019-2022), substitution was about replacing foreign chips. In this cycle (2024-2026), substitution is about replacing foreign infrastructure standards with domestic ones.

• Standards Setting: Chinese companies are setting the standards for liquid cooling interfaces, high-voltage DC distribution in data centers, and smart grid protocols.
• Export Potential: As Chinese AI models go global (the "Token Export"), the underlying infrastructure stack is also being exported. Chinese IDC operators and power equipment manufacturers are winning bids in Southeast Asia, the Middle East, and even Europe, leveraging their cost advantage and proven scale at home.

Final Recommendation Summary

Conclusion: The intersection of AI and Energy is the most compelling investment theme in the Chinese industrial sector for 2026. We urge investors to overweight high-quality power equipment and IDC assets that are demonstrably benefiting from the token-driven compute supercycle.