Overview: The Paradigm Shift in AI Data Center Power

The power supply architecture for AI data centers is undergoing a fundamental transformation, shifting from traditional multi-stage Uninterruptible Power Supply (UPS) systems to Solid State Transformers (SST) paired with 800V DC direct conversion. As AI workloads demand unprecedented power densities, companies like Sungrow are pioneering comprehensive solutions that integrate generation, storage, distribution, and consumption. This evolution is critical for sustaining the continuous operation and efficiency required by next-generation high-density computing environments.

Market Context: The Compute Explosion vs. Power Bottlenecks

The urgency for advanced power solutions is driven by explosive market demand. In the first quarter of 2026, domestic AI compute demand surged by approximately 417% year-over-year, while global top-tier cloud providers increased capital expenditures by roughly 79%. Consequently, compute expansion is vastly outpacing grid infrastructure upgrades. Furthermore, as new AI cabinets scale from kilowatts to hundreds of kilowatts and eventually megawatts, legacy 415V/480V AC distribution has hit a ceiling in current capacity, loss management, and response speed, making power supply a primary bottleneck for AI infrastructure delivery.

Sungrow has introduced the philosophy that electricity supports compute and compute revitalizes electricity, treating power supply capacity as the underlying variable of compute competitiveness, enabling bidirectional interaction between power systems and compute loads.

Technical Details: The EnerNeo SST and 800V Architecture

On July 9, 2026, Sungrow unveiled the EnerNeo SST, positioning it as the next-generation power hub for AI data centers. Utilizing third-generation Silicon Carbide (SiC) semiconductors and an optimized H-bridge series two-stage topology, the system directly converts 10–13.8kV medium-voltage AC to 800V DC. It offers scalable power from 1.5 MW up to 4.5 MW, achieving a remarkable 98.5% system efficiency and 99.999% availability. With a power density of 312 kW/㎡, it significantly outperforms traditional setups. The system ensures physical isolation, integrated full insulation, and high-frequency transformer vacuum insulation, alongside an AI sentinel inspection system for online monitoring and intelligent lifespan prediction.

Aligning with Nvidia’s October 2025 whitepaper on 800VDC architecture for next-generation AI infrastructure, this 800V DC approach drastically reduces copper cable usage and transmission losses. The EnerNeo SST features 6+3 medium-voltage elastic redundancy, tolerating up to three module failures while maintaining 75% derated operation, and supports millisecond-level dynamic voltage regulation.

Industry Impact: Redefining Efficiency and Footprint

The transition to an SST and 800V DC architecture yields profound industry impacts. Traditional multi-stage transformer and UPS systems typically achieve 90%–92% end-to-end efficiency with substantial copper consumption and large footprints. In contrast, Sungrow’s integrated approach minimizes spatial requirements and material costs while enabling seamless, phased scalability. To support this vision, Sungrow established its dedicated AIDC subsidiary in May 2025 and formed a specialized AIDC business division in August 2025, backed by a planned 300-person R&D team focused on migrating green power and high-voltage DC safety technologies to the computing sector.

Implications for AI Data Centers: Strategic Deployments

Ecosystem integration is accelerating. Sungrow has secured strategic framework procurements, including a 30 MW deployment with Dongyueguang for 2026–2027, and a 100 MW project with China United Data Group, featuring a late-2026 demonstration. Additionally, a joint innovation center was established with Alibaba Cloud. As of August 28, 2026, the company disclosed approximately 2GWh in AIDC energy storage orders, with several SST units slated for Q4 deployment. It is crucial to distinguish these signed partners from industry witnesses like Microsoft, AWS, Tencent, ByteDance, Baidu, VNET, and Chindata, who attended the July launch but are not necessarily bound by procurement agreements.

Future Outlook: The Roadmap to Megawatt-Scale Cabinets

Looking ahead, the industry is adopting a three-step deployment roadmap: pilot demonstrations by late 2026 to early 2027, grayscale rollouts in 2027, and large-scale deployment of 400kW–1MW cabinets with synchronized SST commercialization by 2028. For data center operators, the strategic imperative is clear: power capability is the foundational variable of compute competitiveness. By adopting modular 1.5MW SST deployments, facilities can mitigate initial risks while future-proofing their infrastructure for the megawatt-scale AI cabinets of tomorrow.