Has the Solid State Transformer’s Time Finally Come?

0:02 How do you replace a technology that’s been battle tested for literally over a hundred years? 1968 saw a new type of transformer, enhanced with power semiconductors: the solid state transformer, or SST. The SST is significantly smaller and adds some special functionality. But despite being "almost ready" for fifty years, hype cycle after hype cycle has passed without much success. Can AI data centers finally be that killer app? In this video, we profile the solid-state transformer.

0:38 ## Beginnings Transformers transform voltage and current from one level to another. Modern transformers operate on the same physics as the original device shown off by the sexy Michael Faraday when he first demonstrated induction over a hundred years ago. It is made from two coils of copper wires wound around a shared iron core. If you pass an alternating current through the first of two copper coils, its changing magnetic field will induce voltage in that second copper coil.

1:11 But while the transformer's core operating principle has not changed, the thing's designs have diversified. Kind of like how the internal combustion engines of a Ford Model T and a Lambo use the same principles but look radically different today. And you can find all sorts of different transformers in the modern power grid. Most are hidden inside unassuming but ubiquitous containers that people don't look at twice. There is a transformer mounted on your telephone pole - it’s a gray cylinder thing. Or one on the street, maybe with a nice paint job.

1:44 And there are many smaller transformers scattered in your home. Inside your phone charger for instance. Or your microwave oven. The transformer is the anonymous lunch pail worker of the grid. So despite all the world's technological progress, the transformer has not changed all that much. And for good reason! They are simple, 98% to 99% efficient, and have a history of good reliability. ## Going Solid State But for all of its upsides, the transformer has limitations. Most noticeably, the traditional transformer is large and bulky. There is an inverse relationship between a current's frequency and the size of its transformer.

2:22 Since the power grid runs at a relatively low frequency of either 50 or 60 hertz, that means that the transformer has to be physically large. Another major issue is that the transformer is fundamentally passive. It cannot perform fast electronic regulation or clean up waveform defects without additional equipment. So any problems in the waveform like voltage dips or spikes simply pass through. Fair enough. But the emergence of power semiconductors like the thyristor in the 1960s brought forth new possibilities. Why not combine transistors and transformers to get something new?

2:59 ## McMurray's Electronic Transformer In 1968, a research engineer at General Electric named William McMurray filed a US patent that combined solid state power semiconductors with a high-frequency transformer. It takes in a low-frequency current, then uses power semiconductors to "chop up" that current to raise its frequency to maybe 5,000 or even several tens of thousands hertz. Now that the current is of a higher frequency, we can use a physically smaller transformer to step the voltage up or down.

3:31 After that, another set of power semiconductors can lower the frequency back down to 50, 60 or whatever hertz for actual use. This is the core concept behind the solid-state transformer: Power electronics coupled with high frequency transformers to shrink the whole device's footprint by a significant percentage - in some cases, 70-80% - while also providing some form of controllability. I was not able to dig up whether McMurray and GE ever built a commercial version of the thing he patented - which he called at the time the "electronic transformer" - though a few sources did say he built a prototype.

4:09 The patent also does not explain where the idea came from, but I wonder if it has to do with GE's work with European trains. Electric trains in the German-speaking countries and Scandinavia - so Germany, Austria, Switzerland, Norway and Sweden - run at a standard frequency of 16.67 hertz. So that is even lower than the 50 or 60 hertz in the world power grid. The reasons for such a low frequency are historical - and to me, kind of amusing - but the consequences are literally large.

4:39 Since a transformer's size is inversely proportional to the power frequency, a 16.67 hertz transformer ends up being substantially larger than a 50 hertz one. Which is not ideal in a constrained environment like a train. More on this later. ## The Navy There were a few reasons why the US Navy started studying solid-state transformers in the 1970s. One reason had to do with the transformer's passive nature. It lets waveform defects pass right on through - which can pose a problem for sensitive military systems like radars.

5:14 Their existing transformers used a PCB-based chemical called Askarel for cooling and electrical insulation. Unfortunately, Askarel is toxic and accumulates in nature, causing health problems. But since removing it is difficult, they were looking for an alternative system. Another practical reason involved power savings. This was during the energy crises of the 1970s, and “efficiency” was the word of the day. When ships berth at port, they take power from the shore. But their power loads were variable and unpredictable depending on task and ship type.

5:50 While a traditional transformer's nameplate efficiency can reach 98%, that efficiency varies with loading. When loads are variable, you get losses. The Navy estimated that tens of millions of dollars are lost to distribution inefficiencies - which wasn’t ideal. After seeing how certain sequences of power semiconductors can replicate voltage step-down behavior, the Navy hired a contractor to run computer simulations and breadboard testing for a feasibility study. That study concluded that a possible solid-state transformer can compensate for its lower overall efficiency with consistent performance under variable loads.

6:30 I would note that few would consider what the Navy explored to be a true solid-state transformer today. They were circuits without a high-frequency transformer inside - meaning that it did not isolate parts of that circuit to prevent current flow. They call this "galvanic isolation", and normal transformers have it. Anyway. So it is slightly ironic to me that this 1980 paper coined the name "solid state transformer", which of course stuck. The paper concludes that a tremendous amount of R&D and testing work laid ahead. But predicted that the the solid-state transformer could become the "building block" of electrical conservation schemes, and a practical one laid maybe 10-15 years away.

7:17 ## During the 1990s While this did not come to pass, occasional development efforts continued. In 1995, the American-based Electrical Power Research Institute (EPRI) sponsored some of their first research on SST prototypes. The work would slowly continue into the early 2000s, eventually producing what they called the Intelligent Universal Transformer. In 1996, a team in Japan led by Koosuke Harada of Kyushu University proposed the "Intelligent Transformer", which is largely a newer version of McMurray's transformer but with more modern silicon-based IGBT power semiconductor switches.

7:56 While it worked - and offered benefits that traditional transformers cannot provide like controllability and smaller size - its peak efficiency was about 80% overall. This was not enough to compete against traditional transformers. And in the 1990s, power semiconductors had not yet reached the performance levels needed to make the concept practical. ## ABB's PETTs In 2007, the Swedish-Swiss company ABB presented a prototype of the Power Electronic Traction Transformer or PETT.

8:28 They had developed this solid-state transformer variant for trains. yes, we are back to the world of trains again. ABB's work targeted one type of transformer inside the train: The traction transformer, which steps down voltage of the line power so to feed an electric train's traction motor. Unlike a standard transformer, the traction transformer must fit inside a constrained space beneath the train’s floor. For that reason - as well as the fact that the car's abrupt stops and starts causes damaging shocks - designers optimize this traction transformer for size and resiliency rather than pure efficiency.

9:07 So it tends to be more like 90% efficient than 98%. Despite that, it is still one of the heaviest bits on the train car - which adds energy consumption as well as wear and tear. So it seemed like a unique opportunity for the solid state transformer. The PETT that ABB presented was a three-stage device. The first stage takes in AC input from the line and raises the frequency using silicon IGBTs - which stand for insulated gate bipolar transistors and became commercially available in the 1980s.

9:40 The second stage is a medium-frequency transformer that steps down the voltage. Then finally the third stage is a converter that can produce power that feeds into the train's traction drive. ABB continued to work on the PETT until at least 2012, when they presented it at the INNOTRANS 2012 show. This most advanced PETT boasted unprecedented efficiencies of 95-96%. It was half the size of a regular transformer - making it far easier to stick under the floor.

10:13 There were over twenty prototype PETT-based systems introduced - including from big train makers like Alstom, Siemens, and Bombardier. However, the PETT never reached the commercial stage despite delivering on various size/weight and efficiency promises. They were untested and seen as too complex and expensive - costing over 50% more than their traditional counterparts. Ultimately, interest moved away from railways to the power grid. ## Smart Grid SSTs One of the most significant obstacles in solid-state transformer progress was the insufficiency of silicon-based power semiconductors.

10:49 Even the aforementioned 6.5 kilovolt silicon IGBTs start incurring significant losses when the frequency rises above the low kilohertz range. Or in the case of the PETT trains, since the silicon IGBTs are rated only for 6.5 kilovolts, they cannot connect right to the 25 kilovolt power line - forcing additional system complexity and discouraging PETT adoption. So the emergence of practical wide bandgap semiconductors like silicon carbide in the late-2000s caused a great deal of excitement.

11:23 Though not yet cheap, they set off commercialization efforts to make solid state transformers for a new, smarter energy grid. The idea was that as intermittent renewables replace fossil fuels, it gets challenging to store and control the flow of this energy. Distributed renewables like rooftop solar pose special challenges to the grid - voltage swings due to passing clouds, voltage drops during peak demand times, voltage rises during low demand times, and other power quality issues.

11:55 So we need new equipment. One popular concept from the Future Renewable Electric Energy Delivery and Management (FREEDM) center in NC State University promoted the idea of the "internet of energy". In this vision, the solid-state transformer is more than just a tool to step the voltage up or down. The idea is by adding to the SST multiple interface AC/DC ports, we make it into the router of the Energy Internet - transforming electricity and distributing it to different subsystems.

12:27 The solid-state transformer can handle the perceived problems of this intermittence. Researcher Alex Q Huang - who was the main driver behind the whole Energy Internet idea - said: > " would make connecting a solar panel or electric car to the grid as simple as connecting a digital camera or printer to a computer. > That would reduce our reliance on fossil fuels by making it easier for small-scale sources of cleaner energy to contribute to the grid." Excited by the possibilities, the "smart transformer" was chosen as one of MIT's breakthrough technologies of 2011. And a few startups emerged to build them.

13:06 One was Varentec, which raised millions from prestigious investors like Bill Gates and Khosla Ventures. Initial news reports had them planning to build a solid-state transformer. Another was Amantys, which spun out from Cambridge University, with intentions to build SST prototypes targeting wind farms. At the start, the solid-state transformer was seen as this massive market opportunity as utilities adopt renewables and build smart grids. One optimistic estimate in 2012 said that SSTs would grow 82% a year, reaching $5 billion by the year 2020.

13:43 ## The SST's Shortcomings In the end, all of these SST startups pivoted away from SSTs. The product that Varentec eventually debuted were "grid edge" devices - electronics that accompany traditional transformers in delivering control of grid voltage while also providing grid analytics to utilities. Unfortunately, that market did not develop as hoped. And in 2021, the company's grid edge hardware and software assets were sold to Sentient Energy, a subsidiary of the Koch conglomerate.

14:14 Another once-hyped grid electronics startup was GridCo, which raised $54 million but closed its doors in 2018. Many of these startups found that few utilities wanted to buy and roll out an "energy router-type" SST. A recent episode of the SemiAnalysis Weekly podcast hosted Varentec's VP of Engineering and Ops from 2011 to 2013, Haroon Inam. Inam is now the co-founder and CEO of a startup called DG Matrix and I will mention them later. Hold that thought. Inam recalled that the original idea at Varentec was to build an SST for AC-to-AC conversion with some power cleanup and value-add. He reflected: > We realized after two years of work and spending millions and millions of dollars that [building an AC-to-AC Solid State Transformer] was probably one of the dumbest things we could have done...

15:05 > You're taking a hunk of iron and a hunk of copper... that's going to last 40, 50 years... why the hell would anybody in their right mind try to replace that with a bunch of electronics that are going to be more delicate? It also costs way more. Some estimates find that an SST might cost five times more than a traditional transformer - though that gap closes if we look at it on a system-level basis. Interestingly, Varentec's former CEO Deepak Divan went to academia afterwards. And he contributed to a 2022 paper finding that neither SSTs nor "hybrid transformers" - which merge transformers with a converter - to be economically viable without a 60% capital cost reduction.

15:49 Anyway. The point is that SSTs will not succeed by doing what a traditional transformer can do. For it be commercially compelling, it has to do something that transformers simply cannot do - and that people are willing to pay for. Just being smaller or being able to handle voltage fluctuation is not enough. ## AI Data Centers Time for us to talk about AI data centers again. Sorry, guys. A major shift in these data centers has to do with their power usage.

16:15 I am not going to spend a lot of time explaining this, there are big SemiAnalysis articles you can read instead. But traditional data centers receive medium voltage AC power from the utility and run that through a hierarchy of systems - including two sets of transformers and an Uninterruptible Power System (UPS) - to get the DC voltages suitable for the chips. Modern AI racks however need way more power, like 1 megawatt. Power is equal to voltage times current. Bigger current necessitates having a thicker copper wire. So either we can have a thicker copper wire at lower voltage, or a thinner copper wire at higher voltage.

16:52 At 1 megawatt with the existing 54 volt rack architecture, you may need up to 200 kilograms of copper busbars - which is expensive and too heavy considering that this is space that is not generating revenue for the data center. So rather than re-engineering the whole rack and buying ultra-expensive heavy copper wires, we redesign the whole architecture for 800 volts. Nvidia says that such a 800 volt DC or VDC architecture drastically reduces complexity. ## The SST's Time?

17:21 With this, the "Energy Internet Router" type concept of SST might work. Power station builders no longer have to worry so much that their AC or DC power infrastructure will have no customers due to 800VDC transition issues and gets "stranded". Not to mention that SSTs take up less space and eliminate a bunch of old hardware. Seems legit enough to me. And Nvidia does include it into their architecture road map. Moreover, wide bandgap semiconductors have gone a long way since the late 2000s and early 2010s. Silicon Carbide power semis are cheaper and more capable today - thanks to the EV boom in China - though by no means are they a panacea.

18:01 Thus we have a new rush for the SST. There is the aforementioned DG Matrix, which has raised over $60 million for a multiport SST product that acts as a universal power router to distribute the right power to the right user. In addition to them, there is Heron Power, which closed a $140 million series B to bring to market an SST to get around manufacturing constraints in certain larger transformers. Over in Asia, there is the Singapore-based Amperesand, which raised $80 million from Temasek and Walden Catalyst Ventures... which I am fairly sure is Lip Bu-tan's firm.

18:37 And of course, the Taiwanese company that I profiled earlier: Delta Electronics. They recently discussed a future solid-state transformer product which takes medium voltage grid power and turns it into 800-volt distribution voltage for the rack. In February 2026, they discussed installing a SST inside a data center campus in Mainland China run by the company Meituan. Side note. This is how this whole video started. I saw a model of an SST in their showroom. ## Conclusion I know, I know. The rule goes that when someone leads off their video with a question, the answer is usually NO.

19:13 The transformer is insanely hard to compete against. It has confidently slew all comers to the throne. Of course, there are reasons to be skeptical. Right now, gains are largely from simulations. We have yet to see actual performance data from 800 VDC AI data centers equipped with solid-state transformers. And can it outperform other 800 VDC implementations based around traditional transformers, i.e. hybrid transformers? But there is reason for optimism that the SST's time has finally come - at least in some small form. The AI data center buildout differs from the prior hype cycles surrounding the smart grid and train industries.

19:54 The cost improvements from greater simplicity - removing extraneous rectifiers and Uninterruptible Power Supplies - are significant. There is a real forcing function with regards to this copper problem. And unlike with the smart grid situation, there is a centralized circle of builders and key suppliers who can drive standards or even self-build fast without legacy baggage. So I reckon the door is open, we shall see what comes through.