The AI Power Mandate: Why SolidState Transformers Are No Longer Optional

Introduction

For two years, solidstate transformers (SSTs) have been a technology to watch. A promising alternative to conventional copperandiron units, but still a lab project for most engineers.

That is no longer true.

2026 is the year SST went from “nice to have” to “must have” for AI data centers, driven by new requirements from the industry’s most influential player: NVIDIA.

Part One: The Mandate – NVIDIA Sets the Rule

In October 2025, NVIDIA published a white paper on 800 V DC power architectures, explicitly stating that all AI data centers equipped with its GB300series computing clusters must adopt SSTs to convert 10 kV AC directly to 800 V DC [10†L2-L5]. Without SSTs, systems cannot receive official technical support or adapter certification [8†L3-L6]. This has made SSTs a mandatory component for highend AI infrastructure, not a “premium upgrade”.

From NVIDIA’s perspective, the logic is clear. AI clusters are moving from megawatt to gigawatt scale; singlerack power density already exceeds 100 kW [15†L3-L5]. Conventional multistage ACDC conversion incurs excessive energy loss and generates unacceptable heat. With SSTs, the entire power chain is shortened to one stage, slashing losses and increasing usable rack power [8†L8-L15].

Part Two: The Tech – How SST Outperforms Traditional Transformers

Compared with conventional transformers, SSTs deliver drastic improvements:

• Efficiency:Endtoend efficiency is at least 98.5%; pure DCDC conversion exceeds 99% – far higher than conventional solutions.
• Power density:Volume and weight are 60–80% lower, freeing up precious floor space for more GPUs [9†L10-L12; 14†L14-L17].
• Response speed:SSTs react in microseconds to load swings, an absolute requirement for AI workloads that ramp from idle to full load multiple times per second [8†L16-L20].
• DC integration:Directly outputs 800 V DC or ±400 VDC, eliminating the need for separate rectifiers and UPS systems [5†L19-L24].

Part Three: The Players – Who Is Delivering SSTs Today

The technology is moving from test labs to production environments at an accelerating pace, with deployments already visible in both Chinese and global data centers.

Global Leaders

In April 2026, Enphase Energy announced its IQ SolidState Transformer (IQ SST) platform, specifically designed for AI and hyperscale data centers. Each 1.25 MW SST rack is built from 342 GaNbased power modules and achieves a maximum conversion efficiency of 98.5% with 99.999% system availability [5†L5-L32][7†L4-L24].

Hitachi Energy has also been heavily investing in grid infrastructure for AI data centers, recently committing $10 million to a new North Carolina grid technology center and signing an MoU with OpenAI to codevelop “speedtopower” solutions for AI facilities [4†L9-L42].

Chinese Progress

China’s manufacturers are moving at an equally rapid pace.

China XD Electric (Xidian Power Electronics) has supplied a 10 kV/2.42 MVA/800 V SST to the Guian National “EasttoWest Computing” data center project, with the unit already in operation [11†L5-L8].

Sifang Co. released its “Digital Intelligence SST 1.0” in April 2026 – an energy platform that directly converts mediumvoltage AC to 800 V DC and is already in mass production [12†L2-L10].

Delta Electronics implemented SST technology in a Meituan data center in late 2025, using silicon carbide as the core switching device – one of the first realworld, loadbearing deployments of SST infrastructure [10†L7-L13].

Part Four: The Market – Size, Scale, and What’s to Come

Market forecasts underline that this is not a passing trend.

The Enphase announcement projects that the annual US addressable opportunity for SSTs in AI data centers will exceed 11 GW by 2031 [5†L35-L36][7†L4-L24]. A MarketsandMarkets report released in May 2026 notes that efficiency gains of 68–72% from SSTs are driving adoption of mediumvoltage Power Distribution in hyperscale AI facilities [1†L9-L14].

Globally, the Data Center SolidState Transformer market was valued at US35.5millionin2025andisprojectedtoreach**US35.5millionin2025andisprojectedtoreach**US 704 million by 2032**, growing at a CAGR of 31.2% [1†L37-L40].

At the same time, cost remains the primary barrier. SSTs are currently about four times more expensive than conventional transformers, with SiC devices and highfrequency magnetic materials driving the price premium [14†L23-L24]. Industry consensus expects SST costs to drop significantly as manufacturing scales and the technology matures, with wide deployment across power distribution networks anticipated by 2028–2030.

Conclusion

The question for data center owners is no longer “Should I consider SST technology?” but rather “Can my next project move forward without it?”

With NVIDIA mandating SSTs for its GB300 AI clusters, with lead times for conventional transformers stretching to two years or more, and with energy efficiency and rack density now defining AI economics, the industry has reached an inflection point. The solidstate transformer is not a future concept anymore – it is the enabling technology for the next generation of AI infrastructure.