Insight · Industry Watch

SST's Q4 2026 Deployment Wave: Two Product Species and One Number That Is Not Comparable

Published by MVSST Power · November 2026

Read time: 8 min Topic: Deployment Tags: #商业化进展 #AI数据中心 #功率密度 #SiC器件
Home/Resources/White Papers/SST Deployment Wave Q4 2026

Three solid-state transformer deployments were reported within eleven days in October 2026: Delta's prefabricated, containerised SST-based DC power system, launched at Energy Taiwan 2026 in Taipei on 14 October; Guangdong Power Grid's 1 MW ±10 kV medium-voltage DC SST, commissioned at Hengqin in Zhuhai and reported on 19 October; and TBEA's Roy Sync series at 2.5 MW, declared in mass production on 24 October in Tianjin.

Different markets, different voltage classes, different applications — so it is tempting to file these as a general "commercialisation year" story. That framing misses the useful part. Read together, the three show the product splitting into two species with opposite architectural roles, and they show that the efficiency figure every vendor leads with has stopped being a differentiator. All three report 98.5%, and all three mean something different by it.

1. Two species: the grid-side DC hub and the load-side DC entry

The Hengqin unit and the TBEA unit belong to the same component family and are not the same product. Guangdong Power Grid's device is a bipolar ±10 kV medium-voltage DC SST rated 1 MW, developed in-house by the Zhuhai Power Supply Bureau as the core demonstration output of a Southern Power Grid key science and technology programme [2].

Its job is architectural: establish a DC "common bus" so that rooftop PV, storage, EV chargers and data centres connect locally and share one dispatch system, and — via a multi-port flexible interconnection device — lash independent feeders into a mesh so power can route around a faulted line. The voltage class is the specification here; DC bus topology is the value.

TBEA's Roy Sync is the mirror image. Rated 2.5 MW, it takes 10 kV AC in and delivers 800 V DC out, deliberately removing the line-frequency transformer and the separate multi-stage rectification stage in between [3]. Here the 800 V DC output is the specification, because that is the interface the AI load side is standardising on, and the value is holding that bus stable against a load that swings in milliseconds. Its first field application is not a data centre at all but heavy-truck ultra-fast charging, at the Zhiyun Green Logistics station in Guangyuan, Sichuan [3].

2. The footprint gain came from module count, not switching frequency

The Hengqin team's account of how they cut size and cost is worth close reading, because it names the lever most SST discussion skips: raising the voltage each power module can withstand halved the module count, and footprint and cost both fell by more than 30% on that basis [2].

In a cascaded MV SST the module count drives the bill of materials, the enclosure volume, and the number of gate drivers, sensors and bypass paths that must all be qualified. Raising switching frequency shrinks the transformer but adds loss and EMI; raising per-module blocking voltage deletes whole modules and everything attached to them. That is why device voltage class, not frequency, tends to set achievable power density.

TBEA quotes gains on the same axis: roughly 45% less copper and about 60% less floor area than a conventional scheme, plus an internal estimate that a 1 MW rack saves about 280 000 kWh per year [3]. Those last three figures are the company's own internal calculations rather than third-party measurements, and should be weighted accordingly.

3. Three different 98.5% numbers

Delta states SST conversion efficiency of up to 98.5%, and separately claims a 4–5% gain in grid-to-chip energy efficiency in AI data centres [1]. TBEA states maximum whole-machine efficiency of 98.5%, beside input THDi below 3% and output regulation within 1% [3]. Guangdong Power Grid states maximum whole-machine efficiency of 98.5%, beside operating noise below 65 dB [2].

Three vendors, three markets, one identical headline. When a number converges this tightly across competitors it stops carrying information — especially because the boundaries differ. Delta's 98.5% is the SST alone, with the 4–5% figure applied to the wider chain. TBEA's is a whole-machine number published beside power-quality limits. Guangdong's is a whole-machine number published beside an acoustic limit. None of the three defines its measurement boundary precisely enough for cross-vendor comparison, and none should be read as one.

4. Protection and redundancy have become the product

What the three announcements actually compete on is everything around the converter. The Hengqin SST suppresses fault current within 0.1 ms [2] — a grid-support function a line-frequency transformer does not perform at all, and one that matters the moment DC is on the bus, because a DC fault gives no zero crossing to wait for.

Roy Sync uses N+1 redundancy in which a failed power module is bypassed automatically within milliseconds with supply uninterrupted, and provides millisecond-level load response for compute load swings [3]. Delta's contribution is delivery-side: the SST, DC/DC converters, control equipment and DC protection are integrated in the factory into a containerised unit, cutting on-site deployment time by up to 50%, with an optional high C-rate battery for millisecond-level power support [1].

None of these are converter-topology claims. Bypass behaviour, fault ride-through, load buffering and factory pre-integration are system claims, and they are now the part of an SST that is hard to copy.

5. Delivery maturity differs sharply, and that is the real discriminator

The three projects sit at visibly different stages. TBEA has completed type testing with reports from Dongdian Testing Technology (Xi'an) and TÜV Rheinland, runs batch production at a Tianjin base with capacity stated at 200 units a year, and offers indoor and outdoor variants [3]. Delta's contribution is a prefabrication and integration model rather than a capacity figure; Guangdong Power Grid's is a single demonstration unit, with grid-wide scale-up stated as intent rather than as an order book [2].

Only one of the three has published an annual capacity figure, and only one has published third-party type-test reports. That gap — not the efficiency number — is what currently separates a product from a project.

The three deployments side by side

ProjectRating and interfaceDeclared performanceStage
Guangdong Power Grid, Hengqin [2]1 MW, ±10 kV DC bus98.5% max whole-machine; noise <65 dB; fault-current response <0.1 ms; module count halved, footprint and cost −30%+Single demonstration, commissioned
TBEA Roy Sync [3]2.5 MW rated, 1–5 MW scalable; 10 kV AC → 800 V DC98.5% max whole-machine; THDi <3%; regulation <1%; N+1 with ms bypassType-tested, mass production, 200 units/yr
Delta prefabricated SST system [1]Containerised SST + DC/DC + controlSST efficiency up to 98.5%; grid-to-chip +4–5%; deployment time −50%Launched and exhibited

If you specify or build MV SSTs

  1. Ask for the measurement boundary before the efficiency. A whole-machine figure, an SST-only figure and a grid-to-chip figure are not the same claim, even when all three read 98.5%.
  2. Treat module count as the cost and power-density lever. Ask how many cascaded modules the design needs at your voltage class and why — it sets BOM, volume and qualification effort.
  3. Compare protection response times, not just topology. Sub-millisecond fault-current suppression and millisecond module bypass are functions you will need the first time a DC bus faults.
  4. Treat factory pre-integration as a delivery metric. Ask for on-site wiring and commissioning hours, not just equipment lead time.
  5. Weight published capacity and third-party type test heavily. They are the clearest signals of whether a supplier can deliver the second unit, not just the first.
The takeaway: The October 2026 deployments do not close the technology question — they re-site it. Efficiency has converged into a marketing constant, while the engineering value has moved to two distinct product architectures, to module count as the cost driver, and to protection, redundancy and factory integration as the parts that are genuinely hard to reproduce.

Related: MV SST in H2 2026: Pilots, Products and the Move to Procurement · 2026: SST Commercialization Hits Its Inflection Point · Medium-Voltage SST Topologies, Control and the Engineering Gap

Sources

  1. Delta Electronics. "Delta Unveils Microgrid Solutions and the Newly Launched SST-based DC Power System for the AI Era at Energy Taiwan 2026." Press release, 14 October 2026. https://www.deltaww.com/en-US/press/41167
  2. China News Service, Guangdong. "固态变压器示范项目在广东横琴投运" [Solid-state transformer demonstration project commissioned in Hengqin, Guangdong]. 19 October 2026, published 23 October 2026. https://m.chinanews.com.cn/wap/detail/chs/zw/460246.shtml
  3. Xinhua Finance. "中压直降、800V直达算力:特变电工睿驭系列固态变压器发布即量产" [Medium-voltage step-down, 800 V direct to compute: TBEA Roy Sync solid-state transformer launched into mass production]. 26 October 2026. https://www.cnfin.com/cmjj-lb/detail/20261026/4486320_1.html

Disclaimer: This article is compiled from the publicly available sources listed above for industry information purposes only. Performance figures are as published by each project's own organisation, are measured on different system boundaries, and are not independently verified or directly comparable. Figures described by the manufacturer as internal estimates are identified as such. Design decisions on a specific project require project-specific engineering analysis.

Building a Medium-Voltage SST Platform?

Tell us your voltage class, bus architecture and load profile — our engineers respond within 24 hours with a device, module-count and protection assessment.

Request Technical Consultation