AIDC Power Supply Under Triple Challenges: SST Leads the Breakthrough

Date: January 18, 2026
Analysts: Zhao Hui (S0910525030003), Zhou Tao (S0910523050001)
Sector: Industrial Technology / Data Center Infrastructure / Power Electronics
Rating: Overweight (Sector Outlook)

Executive Summary

The global proliferation of Artificial Intelligence Data Centers (AIDCs) has triggered an explosive growth in energy consumption, fundamentally reshaping the power infrastructure landscape. As of July 2025, China’s total intelligent computing scale reached 780,000 Pflops, ranking second globally. This expansion has created a critical bottleneck: power supply. Between 2024 and 2030, data center electricity consumption is projected to reach between 405.1 billion and 530.1 billion kWh. Specifically, AIDC energy consumption in 2025 is estimated at 77.7 billion kWh. The industry now faces a "triple challenge" of supply stability, cost control, and carbon emission management. Existing power architectures, primarily based on Uninterruptible Power Supplies (UPS), are struggling to cope with the high volatility of AI workloads (fluctuation rates up to 50%) and stringent environmental regulations limiting diesel generator backups.

In response, the power supply architecture is undergoing a generational evolution from traditional UPS to High-Voltage Direct Current (HVDC), Panama Power solutions, and ultimately, Solid-State Transformers (SST). SST technology emerges as the potential "ultimate solution," offering system efficiencies of up to 98.5%, a footprint reduction of over 50% compared to traditional schemes, and seamless integration with renewable energy sources. We project that by 2030, the domestic market for SST in China will reach RMB 13.27 billion, growing at a Compound Annual Growth Rate (CAGR) of 64.9% from 2024 to 2030.

This report analyzes the structural shifts in AIDC power requirements, details the technological advantages of SST and 800V HVDC architectures, and provides investment recommendations focusing on companies leading the SST, HVDC, and AI server power supply sectors. We recommend investors focus on leaders in SST technology (e.g., Sifang Shares, China XD Electric), 800V HVDC systems (e.g., Zhongheng Electric, Kehua Data), and upstream semiconductor materials.

Key Takeaways

1. The Energy Bottleneck: AIDC Growth Outpaces Infrastructure Capacity

• Explosive Demand: The rise of Large Language Models (LLMs) has driven data training volumes from gigabytes to trillions of tokens. Consequently, single-rack power density is surging from 20-50kW to over 100kW, with future projections reaching 800kW-1MW per rack.
• Power Constraint: Electricity costs account for 57% of data center operating expenses, far exceeding depreciation (25%), rent (8%), and labor (4%). The sheer volume of power required is becoming a primary constraint on AI development, with major projects like OpenAI’s "Stargate" potentially requiring thousands of megawatts.
• Policy Pressure: Regulatory frameworks are shifting from dual-control of energy consumption intensity to carbon emission controls. New national hub nodes are required to achieve a green power proportion of over 80% by 2025 and 90% by 2030. Currently, 63% of data centers have a Power Usage Effectiveness (PUE) above 1.2, indicating significant inefficiency.

2. The Triple Challenge of AIDC Power Management

Survey data indicates that industry participants face three predominant challenges:
1. Supply Stability (93% concern): AI workloads exhibit hour-by-hour power fluctuation rates of up to 50%, ten times that of traditional cloud computing. Traditional grids and diesel backups (facing stricter environmental approvals and capacity limits) struggle to maintain stability.
2. Cost Control (85% concern): With electricity dominating OPEX, even marginal efficiency improvements yield substantial financial benefits.
3. Carbon Management (77% concern): Compliance with green power mandates and ESG goals requires deep integration of renewable energy, which introduces intermittency issues that legacy systems cannot easily manage.

3. Technological Evolution: The Rise of SST and 800V HVDC

The power distribution architecture is evolving through four distinct generations:
* Gen 1: UPS (Uninterruptible Power Supply): The current mainstream but limited by efficiency (92-95% in 2N configuration) and large footprint.
* Gen 2: HVDC (High-Voltage Direct Current): Removes the inversion stage, boosting efficiency to >95%. NVIDIA’s recent endorsement of 800V HVDC marks a pivotal shift, reducing rack volume by 26% and energy consumption by ~8%.
* Gen 3: Panama Power: An integrated medium-voltage DC system that combines transformers and rectifiers, achieving ~97.5% efficiency and reducing footprint by ~64% compared to traditional 240V DC systems.
* Gen 4: SST (Solid-State Transformer): The emerging "ultimate solution." By replacing magnetic transformers with high-frequency power electronics (using SiC/GaN), SST achieves 98.5-99% efficiency, reduces footprint by >50%, and offers millisecond-level response to load fluctuations. It also enables direct DC coupling with renewables (PV/Storage), facilitating "Source-Grid-Load-Storage" integration.

4. Market Opportunity: SST to Capture Significant Share

• Market Size: We estimate the domestic AIDC new installed capacity will reach 17.7 GW by 2030.
• SST Penetration: Assuming SST penetration rises from 3% in 2025 to 30% in 2030, and unit prices decline from RMB 5.0/W to RMB 2.5/W due to economies of scale, the SST market size in China will grow from RMB 1.09 billion in 2025 to RMB 13.27 billion in 2030.
• Growth Rate: This represents a CAGR of 64.9%, highlighting SST as a high-growth niche within the broader power infrastructure sector.

5. Investment Strategy

We recommend a barbell strategy focusing on:
1. SST Leaders: Companies with proven technology in solid-state transformation and medium-voltage DC integration.
2. 800V HVDC Ecosystem: Providers of high-voltage direct current systems and components.
3. AI Server Power Supplies: Manufacturers adapting to higher power densities (10kW+ PSU) and higher frequencies.
4. Upstream Materials: Suppliers of wide-bandgap semiconductors (SiC, GaN) and magnetic materials essential for next-gen power electronics.

Detailed Analysis

1. Macro Context: The Intersection of AI and Energy

1.1 The Surge in Intelligent Computing

Artificial Intelligence is no longer just a software trend; it is a physical infrastructure driver. The computational power required for training large models has grown exponentially.
* Data Volume Explosion: From GPT-1’s 4.6GB dataset in 2018 to Qwen2.5Max’s >20 trillion tokens in 2025, the data throughput has increased by orders of magnitude.
* Compute Scale: As of July 2025, China’s intelligent computing scale stands at 780,000 Pflops. The digital economy is expected to contribute ~35% to GDP by the end of 2025.
* The "Compute-Energy" Gap: While compute demand grows exponentially, power infrastructure upgrades lag. The "East Data, West Computing" initiative aims to balance this, but western hubs face challenges in stabilizing intermittent renewable energy (wind/solar) without massive storage or hybrid solutions.

1.2 Energy Consumption Projections

The energy intensity of AIDCs is distinct from traditional cloud data centers due to the nature of GPU-intensive workloads.
* Total Consumption: Under a narrow definition (IT equipment only), data center electricity use will reach 405.1 billion kWh by 2030 (CAGR 16.1%). Under a broad definition (including transmission networks), it will reach 530.1 billion kWh, accounting for 3.1% of total societal electricity consumption.
* AIDC Specifics: In 2025 alone, AIDC energy consumption is projected at 77.7 billion kWh.
* Power Density Shift: Traditional racks operated at 4-8kW. Modern AI racks operate at 20-50kW, with next-generation NVIDIA GB200/GB300 clusters pushing towards 100kW-1MW per rack. This density renders traditional air cooling and low-voltage AC distribution inefficient and physically bulky.

Source: Data Center Conference & Exhibition, Huajin Securities Institute

2. The Triple Challenge: Stability, Cost, and Carbon

The transition to AIDCs is not merely a upgrade in compute; it is a fundamental stress test for energy management. Our analysis identifies three critical pain points.

2.1 Stability: The Volatility Problem

Traditional data centers had relatively stable, predictable loads. AIDCs are different.
* Load Fluctuation: AI training and inference involve bursty computations. The power load can fluctuate by 50% on an hourly basis, which is 10x the volatility of traditional cloud services.
* Grid Impact: Such rapid fluctuations cause voltage and frequency instability if not buffered effectively.
* Backup Limitations: Historically, diesel generators provided backup. However:
* Environmental Regulations: 56% of enterprises cite strict environmental assessments, emissions (CO2, pollutants), and noise as major hurdles for diesel expansion.
* Capacity Constraints: 40% of firms report transformer capacity limits restricting diesel output.
* Risk: 28% worry about fuel storage and leakage risks.
* Conclusion: Diesel is increasingly seen as a non-viable long-term primary backup for green, high-density AIDCs.

2.2 Cost: The OPEX Dominance

Energy cost is the single largest operational expense.
* Cost Structure: Electricity accounts for 57% of OPEX. Depreciation is 25%, Rent 8%, Labor 4%.
* Efficiency Sensitivity: A 1% improvement in power distribution efficiency translates directly to the bottom line. For a 2.5MW system running at 90% load, a 3% efficiency gain (achievable with SST vs. UPS) saves approximately 591,300 kWh annually. At industrial electricity rates, this is a significant recurring saving that offsets higher CapEx over time.
* Copper Costs: Traditional low-voltage high-current systems require massive amounts of copper. A 1MW rack at 54V DC requires ~200kg of copper cabling. A 1GW facility could require 500,000 tons of busbar copper. This is economically and physically unsustainable. Higher voltage (800V) reduces current, thereby reducing copper cross-section requirements by ~66% (to one-third of AC 380/220V usage).

2.3 Carbon: The Regulatory Mandate

China’s "Dual Carbon" goals and specific directives like the Data Center Green Low-Carbon Development Special Action Plan are forcing a transition.
* Green Power Quotas: National hub nodes must source >80% of power from renewables by 2025.
* PUE Targets: 63% of existing data centers have a PUE > 1.2. New standards demand lower PUEs, driving adoption of liquid cooling and waste heat recovery.
* ESG Pressure: 56% of enterprises feel pressure from both policy and ESG ratings to reduce carbon footprints. Global tech giants (Google, Microsoft) have set carbon-neutral timelines, pressuring their supply chains and partners in China to follow suit.

3. Solutions: Compute-Power Synergy and Flexible Dispatch

To overcome these challenges, the industry is moving towards "Compute-Power Synergy" (算电协同), integrating energy generation, storage, and consumption dynamically.

3.1 Multi-Energy Supply Networks

Data centers are evolving from passive consumers to active energy nodes.
* Diverse Sources: Integration of Photovoltaic (PV), Wind, Energy Storage, and even Nuclear (in specific zones) to create a resilient microgrid.
* Adoption Rate: 56% of data centers now use some form of new energy, with 40% deploying distributed generation onsite.
* Models:
1. Comprehensive Energy Base: Large-scale integration in resource-rich areas (e.g., Western China).
2. Distributed/Microgrid: On-site PV/Storage for medium-sized DCs, enhancing self-sufficiency.
3. Virtual Power Plant (VPP): Aggregating DC loads and storage to participate in electricity market trading, optimizing costs via arbitrage.

3.2 Flexible Load Dispatch

Instead of just supplying more power, the industry is making the demand flexible.
* IT Load Flexibility:
* GPU Frequency Scaling: AI models can dynamically adjust GPU clock speeds based on real-time power availability and carbon intensity signals. This allows shifting computation to times when green power is abundant (e.g., midday solar peak) without impacting total training time significantly.
* Spatial Migration: Workloads can be migrated across geographically dispersed data centers. If one region has excess wind power, tasks are routed there. This acts as a "virtual transmission line," alleviating grid congestion.
* Non-IT Load Optimization:
* Cooling Systems: Accounting for ~30-40% of non-IT energy use.
* AI-Driven Cooling: Using AI algorithms to predict heat loads and adjust chiller pumps, fans, and valves in real-time.
* Liquid Cooling: Transitioning from air to cold-plate or immersion liquid cooling reduces fan energy and allows higher density.

3.3 Waste Heat Recovery

Data centers generate vast amounts of low-grade heat (20-90°C).
* Policy Support: Government guidelines encourage using this heat for district heating, agriculture, or industrial processes.
* Technologies:
* ORC (Organic Rankine Cycle): Generates electricity from low-temp heat.
* Absorption Chillers: Uses waste heat to drive cooling cycles, further lowering PUE.
* Direct Heating: Supplying nearby residential or commercial heating networks.

4. Technological Evolution of Power Architecture

The core of this report focuses on the hardware evolution of the power distribution system. We identify four distinct generations.

4.1 Generation 1: UPS (Uninterruptible Power Supply)

• Status: Current mainstream, especially in legacy and conservative sectors (finance, government).
• Architecture: AC Input -> Rectifier -> Battery -> Inverter -> AC Output.
• Pros: Mature ecosystem, proven reliability, standard AC output compatible with all servers.
• Cons:
  • Efficiency: Double conversion losses. Typical 2N system efficiency is 92-95%.
  • Footprint: Bulky transformers and batteries require dedicated power rooms.
  • Response: Slow to adapt to rapid load changes compared to solid-state solutions.
• Trend: Moving towards modular, high-density UPS (e.g., 600kVA units shrinking from 1200mm to 600mm width), but physical limits are being reached.

4.2 Generation 2: HVDC (High-Voltage Direct Current)

• Status: Growing adoption, particularly in telecom and internet giant data centers.
• Architecture: AC Input -> Rectifier -> DC Bus (240V/336V) -> DC/DC Converter in Server.
• Pros:
  • Efficiency: Eliminates the final inversion stage. System efficiency >95%.
  • Reliability: Simpler structure, fewer failure points.
  • Cost: Lower cable costs due to higher voltage/lower current compared to low-voltage AC.
• Cons: Still relies on line-frequency transformers; limited scalability for MW-level racks without significant bulk.

4.3 Generation 3: Panama Power (Integrated Medium-Voltage DC)

• Status: Innovative solution deployed by major Chinese internet firms (e.g., Alibaba, Tencent).
• Architecture: Integrates the medium-voltage transformer, rectifier, and DC distribution into a single modular unit. Outputs 240V DC for IT and 540V DC for motors/cooling.
• Pros:
  • Footprint: Reduces space by ~64% compared to traditional 240V DC systems (e.g., from 300m² to 110m² for a specific module).
  • Efficiency: Peak efficiency ~97.5%.
  • Modularity: Prefabricated, faster deployment.
• Cons: Still uses magnetic transformers, which are heavy and have physical size limits proportional to power.

4.4 Generation 4: SST (Solid-State Transformer) – The Ultimate Solution?

• Status: Emerging, with recent breakthroughs announced by Delta Electronics and others in late 2025.
• Technology: Replaces the bulky iron-core transformer with high-frequency power electronic converters using Wide Bandgap (WBG) semiconductors (SiC/GaN).
• Key Advantages:
  1. Ultra-High Efficiency: System efficiency reaches 98.5-99%. By eliminating multiple AC/DC stages and using high-frequency switching, losses are minimized.
  2. Compact Footprint: Operating frequency increases from 50/60Hz to >20kHz. This reduces transformer volume by 70-80%. A 1MW SST power cabinet occupies only 1m², a >50% reduction vs. traditional schemes.
  3. Fast Response: Millisecond-level dynamic response (<5ms) to GPU load spikes, ensuring stability without oversized buffers.
  4. Power Quality: Integrated Active Power Filter (APF) and Static Var Generator (SVG) functions. Harmonics on the 10kV input side are <1%. No external harmonic filters needed.
  5. Renewable Integration: Native DC ports allow direct connection of PV, Storage, and Fuel Cells. It acts as an "Energy Router," facilitating >50% green power penetration.
  6. "Silicon Replaces Copper": Significant reduction in copper usage. A 2.5MW SST system uses far less copper than equivalent AC/HVDC systems due to higher voltage transmission (800V DC) and lack of heavy copper-wound transformers.

4.5 The NVIDIA Catalyst: 800V HVDC Standardization

A critical driver for SST and HVDC adoption is NVIDIA’s strategic shift.
* Announcement: At COMPUTEX 2025 and OCP Summit, NVIDIA endorsed 800V HVDC for next-gen AI infrastructure (Rubin Ultra platform).
* Rationale:
* Space: 54V DC distribution for 1MW racks would require 64U of rack space just for power shelves, leaving no room for compute. 800V drastically reduces this.
* Weight/Cost: 54V requires massive copper busbars. 800V reduces current by ~15x, slashing copper weight and cost.
* Efficiency: Fewer conversion stages mean less heat and higher overall system efficiency.
* Implication: This validates the move away from traditional AC/UPS and creates a standardized ecosystem for 800V components, benefiting SST manufacturers who can natively output 800V DC.

5. Market Sizing and Forecast

5.1 Global and Domestic AIDC Capacity

• Global: IEA baseline scenarios project global data center capacity to grow from 100GW (2024) to 225GW (2030), a CAGR of 14.5%.
• China’s Share: As the second-largest AIDC market, China’s share of global incremental capacity is expected to rise from 35% (2024) to 62% (2030), driven by investments from Alibaba, Tencent, ByteDance, and state-owned clouds.
• Domestic Incremental Capacity: We forecast China’s new AIDC installed capacity to reach 17.7 GW in 2030.

5.2 SST Market Projection

We model the SST market based on penetration rates and price declines.

Assumptions:
1. Global AIDC CAGR of 14.5%.
2. China’s share of incremental capacity rises steadily to 62%.
3. SST penetration starts low (3%) as it is a new technology but accelerates as costs drop and reliability is proven.
4. SST unit price declines from RMB 5.0/W to RMB 2.5/W due to scale and semiconductor cost reductions.

Result: The domestic SST market will grow from ~RMB 1 billion in 2025 to RMB 13.27 billion in 2030, with a CAGR of 64.9%. This represents a high-growth opportunity within the broader power equipment sector.

6. Competitive Landscape and Ecosystem

The transition to SST and 800V HVDC is fostering new ecosystems.

6.1 International Collaboration

• NVIDIA’s Alliance: NVIDIA is leading an 800V HVDC alliance, partnering with:
  • Navitas Semiconductor: For GaN/SiC power ICs.
  • Infineon: For SiC power modules.
  • Delta Electronics: For power systems and SST prototypes.
  • Vertiv: For DC busway distribution.
• Standardization: These collaborations are crucial for setting safety and interoperability standards for 800V DC in data centers, which currently lack universal norms compared to AC.

6.2 Domestic Dynamics

• Technology Transfer: China’s strong EV sector (which widely adopts 400V/800V architectures) provides a mature supply chain for high-voltage DC components. This expertise is transferring to the data center sector.
• Policy Drivers: "East Data, West Computing" and green power mandates are forcing local providers to adopt advanced solutions like SST to meet PUE and green energy ratios.
• Key Players:
  • State-Owned Enterprises: China XD Electric, TBEA are leveraging their grid expertise to develop SST for large-scale hubs.
  • Private Tech Leaders: Sifang Shares, Jinpan Technology are innovating in modular SST designs.
  • Power Supply Specialists: Megmeet, Eltek (via local partners), Kehua Data are adapting server PSUs and rack-level power for 800V inputs.

Risks / Headwinds

While the outlook for SST and advanced power architectures is positive, investors must consider the following risks:

1. Technology Iteration and Maturity Risk

• Unproven Long-Term Reliability: SST is a relatively new technology in the data center context. While promising in labs and pilots, its long-term reliability under continuous 24/7 high-load operation needs further validation.
• Semiconductor Dependence: SST relies heavily on SiC and GaN devices. Any supply chain disruptions or yield issues in these semiconductors could delay deployment.
• Standardization Lag: Lack of unified industry standards for 800V DC and SST interfaces may lead to fragmentation, increasing integration costs and slowing adoption.

2. Policy and Regulatory Risk

• Subsidy Changes: The current growth is partly driven by government subsidies for green data centers and "New Infrastructure." If these subsidies are reduced or eligibility criteria tighten, project ROI could deteriorate.
• Grid Connection Rules: Regulations regarding how data centers interact with the grid (e.g., virtual power plant participation, reverse power flow from onsite PV) are still evolving. Restrictive policies could limit the economic benefits of SST’s renewable integration capabilities.

3. Market Competition and Cost Pressure

• Price Wars: As more players enter the SST and HVDC market, competition could intensify, leading to margin compression.
• Incumbent Resistance: Traditional UPS manufacturers (e.g., Eaton, Schneider, Vertiv) have strong market positions and may innovate to extend the life of AC architectures, slowing the transition to DC/SST.
• CapEx Barrier: SST has a higher initial CapEx than traditional UPS. In a downturn, customers may prioritize lower upfront costs over long-term OPEX savings, delaying adoption.

4. Macro Economic and AI Demand Risk

• AI Bubble Concerns: If the commercialization of AI applications slows down, the demand for new AIDC capacity could plateau, reducing the need for new power infrastructure.
• Energy Prices: Fluctuations in electricity prices affect the payback period for efficiency upgrades. If electricity prices drop, the financial incentive for high-efficiency SST diminishes.

Rating / Sector Outlook

Sector Outlook: Overweight (Leading the Market)

We maintain an Overweight rating on the AIDC Power Infrastructure sector, specifically highlighting the Solid-State Transformer (SST) and 800V HVDC sub-segments.

Rationale:
1. Structural Growth: The transition to AI-driven computing is irreversible, and power is the primary bottleneck. This guarantees sustained demand for power infrastructure upgrades.
2. Technological Inflection Point: We are at the cusp of a generational shift from AC/UPS to DC/SST. Early movers in SST will capture disproportionate value as the market scales.
3. Policy Tailwinds: China’s dual-carbon goals and green data center mandates provide a regulatory floor and incentive for adopting high-efficiency, renewable-friendly technologies like SST.
4. High Barrier to Entry: SST requires expertise in both high-voltage power engineering and advanced power electronics (SiC/GaN), creating a moat for established players and tech leaders.

Investment Horizon: 6-12 Months.

Investment View

We recommend a focused investment strategy targeting companies with clear technological leadership in SST, HVDC, and related power components. We categorize recommendations into four tiers.

Tier 1: SST Technology Leaders

These companies are at the forefront of developing and commercializing Solid-State Transformer solutions. They benefit directly from the projected 64.9% CAGR in the SST market.

• Sifang Shares (601126.SH):

  • Logic: A leader in power automation and protection. Sifang has been actively researching and piloting SST technology for distribution networks and data centers. Its strong R&D in power electronics positions it well for the SST transition.
  • Catalyst: Potential large-scale orders from national hub data centers.

• China XD Electric (601179.SH):

  • Logic: A state-owned enterprise with dominant position in high-voltage transmission equipment. Leveraging its grid expertise, China XD is developing medium-voltage DC and SST solutions for large-scale energy bases and data centers.
  • Catalyst: Participation in "East Data, West Computing" flagship projects.

• Jinpan Technology (688676.SH):

  • Logic: Specializes in dry-type transformers and has expanded into energy storage and digital power solutions. Jinpan is exploring modular SST designs that integrate with its existing transformer manufacturing capabilities.
  • Catalyst: Export opportunities and domestic green data center contracts.

• TBEA (600089.SH):

  • Logic: A global leader in transformers and new energy. TBEA’s integrated approach (polysilicon, inverters, transformers) allows it to offer holistic "Source-Grid-Load-Storage" solutions featuring SST technology.
  • Catalyst: Synergies with its massive renewable energy portfolio.

Tier 2: 800V HVDC System Providers

These companies provide the broader HVDC infrastructure, including rectifiers, distribution units, and monitoring systems, benefiting from the NVIDIA-led 800V standardization.

• Zhongheng Electric (002364.SZ):

  • Logic: A pioneer in HVDC power systems in China. Zhongheng has extensive experience in telecom and data center HVDC deployments and is well-positioned to upgrade its offerings to 800V standards.
  • Catalyst: Renewed CAPEX from telecom operators and internet giants.

• Kehua Data (002335.SZ):

  • Logic: A leading UPS and data center solution provider. Kehua is transitioning its product mix towards HVDC and modular power systems, aligning with industry trends.
  • Catalyst: Strong brand recognition and existing customer base in financial and internet sectors.

• Hopewind Electric (603063.SH):

  • Logic: Originally focused on wind/solar inverters, Hopewind has strong expertise in high-power power electronics. This expertise is transferable to data center HVDC and SST applications.
  • Catalyst: Cross-sector technology application and cost advantages from renewable supply chains.

Tier 3: AI Server Power Supply Manufacturers

As rack power densities increase, the demand for high-efficiency, high-power-density server PSUs (Power Supply Units) grows.

• Megmeet (002851.SZ):

  • Logic: A leading provider of industrial power supplies. Megmeet has developed high-power PSUs for AI servers and is collaborating with major server OEMs.
  • Catalyst: Direct supply chain entry into major AI server manufacturers.

• Eltek / Local Partners (Note: Report mentions "Oulutong" - likely referring to local assembly or specific entity, but generally Shenzhen Kstar or similar might be implied in broader context, however sticking to report: Oulutong):

  • Logic: Specialized in communication and server power supplies. Benefiting from the volume growth of AI servers.

• Aikesaibo (688711.SH):

  • Logic: Focuses on precision power testing and power supplies. Its technology is relevant for ensuring the quality and efficiency of high-end AI power modules.

Tier 4: Critical Components and Upstream Materials

The shift to SST and HVDC increases demand for wide-bandgap semiconductors and advanced magnetic materials.

• Taiyong Changzheng (002801.SZ) & Liangxin Shares (002706.SZ):

  • Logic: Leaders in circuit breakers. SST and HVDC require specialized DC circuit breakers (solid-state or hybrid) for protection. These companies are developing DC-breaking technologies.

• Cloud Road Shares (603097.SH):

  • Logic: Supplier of amorphous and nanocrystalline magnetic materials. These materials are essential for high-frequency transformers in SSTs, offering lower losses than traditional silicon steel.

• Sanan Optoelectronics (600703.SH) & Innoscience (Unlisted/Pre-IPO context):

  • Logic: Leading manufacturers of SiC and GaN semiconductors. The core enablers of SST efficiency and size reduction. Increased SST adoption directly drives demand for their wafers and devices.

• Other Potential Targets:

  • Xinte Electric (002202.SZ): Potential involvement in integrated energy solutions.
  • Xinfeng Guang (688663.SH): Energy storage and power electronics.
  • Shenghong Shares (300693.SZ): Charging modules and power supplies, technology transferable to DC data centers.
  • Shuangjie Electric (300444.SZ): Switchgear and power distribution equipment.

Conclusion

The AIDC power supply sector is undergoing a profound transformation driven by the physical limits of AI compute density and the regulatory imperative for green energy. Solid-State Transformers (SST) represent the technological apex of this evolution, offering unparalleled efficiency, density, and renewable integration. While the technology is in the early commercialization phase, the projected 64.9% CAGR and RMB 13.27 billion market size by 2030 indicate a robust growth trajectory.

Investors should prioritize companies with verified SST prototypes, strong partnerships with hyperscalers or grid operators, and vertical integration in key components like SiC/GaN. The transition to 800V HVDC, endorsed by NVIDIA, provides a near-term catalyst for HVDC system providers, while SST offers the long-term structural alpha. We recommend building positions in the identified leaders, monitoring technical milestones and pilot project conversions as key indicators of momentum.

Appendix: Technical Deep Dive

A. Why SST is Superior for AI Workloads

1. Dynamic Response to Load Spikes
AI training jobs are not constant. They involve periods of intense matrix multiplication followed by communication phases. This creates a "square wave" like power demand.
* Traditional UPS: Relies on large capacitors and batteries to smooth these spikes. This is slow and inefficient.
* SST: Uses high-frequency switching (kHz range) to adjust power flow in microseconds. It can instantly match the GPU’s demand, reducing the need for oversized buffering components. This "AI-grade response" (<5ms) ensures that voltage sags do not trigger server resets, which are costly in long training runs.

2. Harmonic Mitigation
Non-linear loads (like server PSUs) generate harmonics that pollute the grid.
* Traditional: Requires separate Active Power Filters (APF).
* SST: The control algorithm of the SST can inherently shape the input current to be sinusoidal, acting as an APF itself. This reduces equipment count and cost.

3. Voltage Transformation Flexibility
* Traditional: Fixed turns ratio transformer. To change voltage, you need a different transformer.
* SST: Software-defined voltage transformation. It can step down 10kV to 800V, 400V, or even variable DC voltages dynamically. This flexibility is crucial for a microgrid that might switch between grid power, battery discharge, and PV input.

B. Economic Model: SST vs. UPS TCO (Total Cost of Ownership)

Assumption: 2.5MW Data Center Module, 10-Year Lifecycle, 90% Load Factor, Electricity Cost RMB 0.8/kWh.

Note: This simplified model illustrates that despite higher CapEx, SST becomes cost-effective within 3-4 years due to energy savings. In high-electricity-cost regions, the payback is even faster.

C. Regulatory Timeline and Impact

• 2024: Data Center Green Low-Carbon Development Special Action Plan released. Sets the tone for PUE < 1.25 and green power > 80%.
• 2025: Implementation phase. National hub nodes begin strict enforcement. "East Data, West Computing" projects require integrated storage and green power.
• 2026-2027: Expected rollout of national standards for 800V DC and SST safety. NVIDIA’s Rubin platform launches, driving global demand for 800V infrastructure.
• 2030: Target for >90% green power in hub nodes. PUE targets tighten further. SST expected to be a standard option for new high-density builds.

Final Remarks

The convergence of AI and Energy is the defining infrastructure theme of the next decade. Investors who recognize the shift from "passive power distribution" to "active, intelligent energy management" will be well-positioned. SST is not just a component upgrade; it is an architectural paradigm shift that enables the sustainable scaling of artificial intelligence. We urge institutional investors to closely monitor the commercial deployment progress of the recommended companies, particularly their order books from major cloud providers and state-owned energy groups.

Disclaimer:
This report is prepared by Huajin Securities Co., Ltd. for institutional clients only. The information contained herein is based on publicly available data and believes to be reliable, but Huajin Securities does not guarantee its accuracy or completeness. The opinions expressed are subject to change without notice. This report does not constitute an offer to sell or a solicitation of an offer to buy any securities. Investors should make their own independent decisions and consult with their financial advisors. Past performance is not indicative of future results.

Analyst Certification:
The analysts named in this report certify that they have accurately represented their personal views about the subject companies and securities. They also certify that no part of their compensation was, is, or will be directly or indirectly related to the specific recommendations or views expressed in this report.

Huajin Securities Research Institute
Address: 30th Floor, Lujiazui Century Financial Plaza, No. 759 Yanggao South Road, Pudong New Area, Shanghai.
Website: www.huajinsc.cn