What Solid State Transformers Mean For Data Centers

TITLE: What Solid State Transformers Mean for Data Centers CHANNEL: Data Center Richness DATE: Unknown ---TRANSCRIPT--- The transformer is a key component of power delivery, but its core design dates to the 19th century. A number of startups are working to build a better transformer using software-defined power electronics that are built for the AI era. One of these companies is DG Matrix. It’s focused on solid-state transformers, which are emerging as a critical piece of infrastructure for next-generation AI data centers. Today my guest is DG Matrix CEO Haroon Inam, who joins us to discuss solid-state transformers and how a multi-port approach can provide new options for savings and efficiency in the data center power chain. Let’s get started. Welcome to Data Center Richness. I’m Rich Miller, and I’ve spent 25 years telling the story of data centers, [music] cloud computing, and AI infrastructure. In this podcast, I’m [music] sharing conversations with the innovators building our digital future. Now, here’s [music] our show. Haroon Inam, welcome to Data Center Richness.

Thank you very much. Thank you very much for hosting this. So, you’re the CEO and the founder of DG Matrix, which is involved in solid-state transformers, which is a technology that there’s tremendous interest in right now. We always like to start give our listeners a sense of your vantage point, your perspective on the industry. So, if you could maybe give us a little quick overview of your background and and how you you came to the work you’re doing at DG Matrix. Sure. Thank you for for asking. So, I’m I’m one of the two co-founders of DG Matrix, and I started in a power electronics a long time ago. When it was unpopular to go into what was called a totally dead profession. The the the all the cool people were going into digital electronics and into chip design and into software and power electronics didn’t attract a lot of people. However, as I finished my undergraduate degree, I had the benefit of running into a a professor who was at the top of his game in power electronics globally. And and I had the chance to learn a lot from him. Not only about power electronics, but how to really value the work that you do and the thoroughness that you bring in and the hard questions that you ask and the fundamentals that you grasp. So, that’s how I got into power electronics and I sort of devoted my my uh uh my life to my career. And uh and for the last 40 years, I have worked in anything from UPSs, computer room power, data center power, solar inverters. I’ve worked on uh transmission power flow control also through power electronics. There’s Dreamliner jet engine starters, they’re all power electronics, the air compression systems, the Joint Strike Fighter uh jet engine starters and and that came out of my group as well. So, I’ve had a chance to to do a lot of different things and um and I started in SSTs back in 2011 when when we were developing uh SSTs that shouldn’t have been developed, but we learned a lot. Well, the folks in the data center industry are well familiar with uh traditional transformers. Uh it it in a larger fashion now because uh they’re part of the challenge acquiring them and getting them. There’s there’s backlogs, Uh but they’re key pieces for pretty much the whole data center power chain. The with solid state transformers are a change from that and create a lot of new capabilities. Maybe give us an overview of solid state transformers themselves and the problems they solve and how they’re they’re different and can do a few more things than the traditional transformers. Traditional transformers basically a hunk of iron with with copper wound around it. It’s a it’s a massive amount of transformers installed globally. But just like transportation transformers, right? You could have a little scooter you could have a bicycle, a scooter, you could have a rickshaw, you could have a car, you could have a Mac truck, you could have a ship and you could have an airliner. So just like that transformers come in multiple grades. You’ve got transformers that convert hundreds of thousands of volts down to tens of thousands of volts. Those are the large grid transformers, the ones you see in major substations near transmission lines. You also have transformers that convert medium voltage to low voltage and then you’ve got transformers that convert one form of low voltage to another form. Heck, you’ve even got transformers in in possibly your iPod charger or your iPhone charger to isolate from from from from the grid for safety reasons and whatnot. So, the ones that are in short supply, the ones that have a lead time issue are the very large transformers that convert the hundreds of thousands of volts to the tens of thousands of volts. There are also some supply chain limitations on on transformers that convert medium voltage to low voltage, but that’s far less because there’s many many more players that do that. And and and so so which transformer is in short supply? It’s It’s the ones that are more complex. And why are those transformers in short supply? Well, electrification is growing everywhere. And it’s driven was originally driven by EVs and and and well, the population as population grows, you need more electricity. And then and then as you transportation started converting to uh EVs and and and fleet electrification started coming on, there was a increased demand for that. But, the AI data center demand is so concentrated. And and and a gigawatt class, even a 100 megawatt class data center creates such a strain on the grid that that the demand for for those transformers has gone way up and there’s not enough people who know how to do the very large complex transformers. So, that’s where the constraints come from. In addition, the grid has never had this much demand before go up in in this faster manner. So, now you’ve got the transmission line conductors, the insulators, the towers, all that has become longer in in procurement lead times. And and you’ve also got to do permitting when you when you do a large transmission line upgrade. So, I think yeah, the supply chain problems are are not just transformers of that grade, but but the whole infrastructure of upgrading the grid. Within that, I uh as I gather you you guys have been focused on pretty much the the medium voltage range and in where you’re looking at. Is that correct? That’s right. We’re We’re actually doing both medium voltage solid state transformers to low voltage, but we’re also doing um a a multi-port solid state transformer that can that can use a legacy transformer, a a medium voltage to low voltage transformer, but bring together uh the advantage, the superb advantage of multi-port at at low voltage, which is shipping today. And and and so we we have a team of 400 plus engineers with 24 doctors, PhDs in different professions to roll out every one of those versions of medium voltage to low voltage and the low voltage multi-port as well. So, what are the advantages of of the multi-port approach? What are the some examples of ways that’s that’s useful? So, a a transformer, Rick, is is basically a chunk of iron, laminations, iron laminations with copper or aluminum wound around it. And it works because of Faraday’s law. And and and you know, the the rate of change of flux in a magnetic material induces power in in the other other windings and that’s workhorse for the industry with just aluminum or copper and iron. And and a solid-state transformer is very different. That iron is replaced by a special magnetic material that that is coupled with with electronic circuitry, hence the solid-state portion, where the transformer becomes much much smaller, much smaller, because the frequencies that interact with the transformer are much higher frequencies. And so that reduces the burden on on the magnetic material and it comes way down in size. And and so that’s where the solid-state transformer terminology came from. Uh that you’re merging solid-state electronics with that transformer. And what we have found out that if you try to take a solid-state transformer and replace the AC to AC function, it’s a dumb dumb dumb way to go after it. You’ve added complexity, you’ve added a lot of cost, you’ve added less reliability, and I don’t think there’s much market for for an SST where you’re doing AC to AC replacement. There is, however, if you look at doing AC to DC conversion and dropping the voltage or adding more functionality, that’s where the balance of plant cost comes down, the reliability can go up, and and there’s an economic benefit. So so the the additional power conversion adds value. And then taking that further, the more more functionality you add by aggregating multiple sources into a solid state transformer, you can’t do that with a regular transformer, but you can do it with a solid state transformer and specialized controls. Then the value goes way up because now you’ve got what is called a multi-port solid state transformer with multiple DC and AC ports, and you can combine solar energy, fuel cells, which are which are predominantly DC, you can combine natural gas turbine outputs, AC, you can combine grid power, and you can do it almost at any ratio and sequence you want in fractions of a second, and you can feed multiple loads simultaneously, AC or DC. And that’s the magic of of multi-port SSTs, and the economic value is huge because you’ve collapsed everything from a statcom, a UPS, a rectifier, a a the spike elimination from from the servers, the GenAI servers, you have collapsed the energy management system, and you’ve also able to supply both AC and DC power simultaneously, de-risking future data center 800 volt or AC adoption, you don’t have to guess. You can do both and and you’ve also collapsed the energy management system into a very fast software-driven power fabric that can route, aggregate, and route power in any way you want. And and I think that’s that’s that’s the advantage that a multi-port SST brings to the market. And you mentioned the the the software component uh we’re going from the the old hardware that you discussed to you know using semiconductors and software and effectively programmable option here where you didn’t really have that before correct? Correct. I think a lot of folks have have been doing firmware and controls using digital signal processors DSPs microprocessors CPUs but they seem to run down to a limit of a few microseconds. It’s tough for these devices to go below below a few microseconds of computational time and lag and and so if you really want to get down in into a having a true I think firmware and and software controlled converter you’ve got to get into the nanosecond range in programmability. So we offer that capability that from tens of nanoseconds to microseconds to milliseconds to seconds we do the whole stack tightly integrated and and and and that allows you to make it extremely programmable. The same machine can be a car charger or it can be 800 volt DC architecture to to to power your data center and and the advantage of that is that you make one machine you keep duplicating it over and over again you make the same spare parts you make it with a consolidated supply chain you make it with a trained workforce you make it with the same quality principles and you deploy in multiple markets. So we’re big believers in in that software programmability down to the billions of seconds. Now you mentioned that multiport creates the opportunity to use a number of different power sources in the the data center sector there’s been a lot of interest in microgrids uh, that can do that type of thing. Uh, are you guys an alternative to to microgrids or you or your equipment could be used in microgrids? How do how do you uh, uh, reside in sort of that part of the world? Sure, excellent question. I think that the simple answer is we make microgrids far more economical. Um, and and and the reason is people have been combining multiple sources in microgrids for years. The problem is not the the the combination of sources. The problem is, which which by the way yields the lowest levelized cost of energy. The problem is all the electronics, the inverters, the rectifiers, the isolation transformers, the protection gear that it takes to do that job. And and what a multiport does is it collapses all the electronics in a microgrid into a unified block and it makes it more economically viable and it increases the speed. It It really the speed to compute or the speed to get more power goes way up. So, there’s a number of ways uh, you know, as I understand that solid state transformers can be helpful to the data center sector. As you mentioned, there are availability issues with uh, traditional transformers. That uh, that might be one place. Uh, there’s also power distribution that changes coming to the data center particularly with the 800 VDC transmit uh, transition. Want to talk more about that in a second. But then there’s also uh, the power oscillation, the fact that that AI workloads themselves are pretty spiky and that creates problems uh, as well. Uh, uh, can you maybe uh, provide an overview of of the the various ways that SSTs can be uh, can be useful to data centers? So, what one of the first problems is, right? In a in a in a classic data center, um you’re getting the power from the grid. And and those are the data centers that house your email servers. They allow you to watch videos of of cats slapping each other or or whatever, right? And and very or or I love cats, by the way. So So So those those are the standard standard uh data centers, right? That that are very complex machines. AI data centers are different. AI data centers have unprecedented density increases in the amount of power that’s required in a rack. And and so now you you’ve got a rack that was doing 10 kilowatts, you’re headed to a rack that’s doing a megawatt. That’s a 100x increase in in in a rack that’s uh this 19-20 in uh wide. And and so a typical rack would use 10 kilowatts of power, enough for six houses. Now you’re going to a rack, just one rack that’s going to use the same power as 600 houses. So So how do you cope with that density? Number two, when you combine multiple racks, you get hundreds of megawatts of demand for AI. Which you need that computational power. How do you get that power from the utility? And in most places, the utility doesn’t have the ability to deliver that in a handful of months or in under a year or two. And that’s where the need for behind the meter power goes way up. And and and so that’s that’s the second thing that’s driving the need for new architectures. And then when the when the when the servers come, the GPUs come in in in server form, when those servers kick in in learning modes or other modes, they have to operate in synchronicity, a large number of these servers. And the way they work is they have a power profile that can go up to double the power four, five times a second, for example. And boy, that is a real problem. You got a megawatt of power going into that dense little rack, and now you’re pulsating that power to 2 MW, 1 MW, 2 MW. That is very, very nasty for the grid to take. You’re introducing harmonics, you’re introducing oscillations, you’re introducing an instability. And if you try to power those spikes from generators, you’re introducing a resonance that’ll likely to tear off the shaft of the generator, just wear it out or tear it off. And if you present it to batteries that are not designed to take a very high cycle charge discharge cycles, it’s going to wear them out prematurely. So, those are all the problems that have to be solved for a modern data center. So, tell me a little bit about what your your offering for the data centers looks like. What are the components? How’s it delivered? What Where does it fit in the the larger uh uh data center infrastructure? So, we’re we’re working with with our multiport solid state transformer, as as I was saying, can combine solar, fuel cells, any AC or DC source, combine it in milliseconds, and deliver both AC and DC power out. But those transformers, those solid state transformers, have to be put in containerized versions, where where you can take multiple ones of these, you can parallel these into a 1.6 MVA block or a or a 3. 2 MVA block or a 4.8 MVA block, put them on backs of trucks combined with switchgear, and that becomes your deployable skid. And then these skids are placed outside data centers where you’re taking work that would have been normally been done on site with expensive labor, you’re shifting that work to high quality controlled work in a protected factory, and and you’re making the skids there more economically with higher quality, and then you’re deploying them in a fraction of the time that you would have had to construct this this sort of a pod-like arrangement with legacy tools. That’s what we’re doing. The both the SSTs, and then we’re working with partners to skid and containerize these. The data center industry has a a pretty long history of trying to look at the efficiency between the grid and the rack. And partly because there’s UPS systems, backup batteries, uh I think it’s fair to say there’s always been more, you know, conversions and step-downs than uh most folks like because uh there’s always some some energy losses involved with that. That’s been a real uh target area for energy efficiency efforts. Um SSTs offer some real opportunities there, correct? Absolutely. When you reduce the number of power conversion stages, your efficiency goes up. And and and and so, if if the more you take out in that value delivery chain, whether it’s a transformer 95 not 98 99% efficient or more, or it’s a conversion 95 97%, you combine multiple ones, you’re down to you know, low 80s to low 90s conversion efficiency end to end. And with an SST, you’re likely to get mid-90s to high 90s conversion efficiency. So, you’ve gained 10% 15% there versus legacy architectures. More than that, I think I think what what a multiport solid state transformer can also do is it can help you eliminate stranded power. For example, if you’ve got a chiller, and the chiller is 20% of the data center, you’ve got to devote a feeder going exclusively to that chiller. And then you’ve got to devote a feeder going to your DC power. So, if the chiller is 20% rated, you’ve now you you can only deliver 80% of the power to DC or you’re going to trip your breaker upstream. But, that chiller doesn’t kick on all the time. So, imagine a solution in which you’re aggregating that power through a multiport solid-state transformer, it’s supplying the chiller power, and when the chiller power is not going out, in that same device, it can provide all the power to DC racks. Stranded power has a far better economics than than efficiency. Efficiency is great. You might save 5%, 10%. With stranded power elimination, you could save 20, 30, 40% of power and deliver far more compute. And a rule of thumb is it’s about $10, $12 per watt of of AI revenue that you can get per year. Right? So, 100 MW data center, 20% stranded power, 20 MW, quarter billion dollars of revenue a year by eliminating stranded power. That’s a hell of a lot better than saving one percentage point, which at 50% loading is 700K of of savings. So, I think I think the real savings is in killing stranded power, but hey, we’ll take the efficiency improvement as well, too. Well, well, stranded power is also something that uh uh is an industry problem and and uh folks are looking at as well. But, the thing you mentioned about the AI factories, AI infrastructure, is the scale at which it operates, not just in density, but in the size of of these facilities. So, even small gains can ripple across a large footprint uh and and uh add up pretty quickly. So, one of the the big player in in AI infrastructure has been Nvidia, whose hardware and designs tend to drive a a pretty strong share of the Um and one of the initiatives they have uh which is very much looking at the density issue and uh and the the most efficient uh power distribution is the plan transition to 800 VDC. Um and I know that that uh uh this is a a a focal point for for a lot of folks who are working with SSTs. Tell me what sort of a an opportunity uh that presents and what are the ways in which uh uh SST technology can uh play a role in the uh 800 VDC transition. So, yes, uh excellent question. So, I think I think the right when the more systems you combine in an SST, the better the economic value. And one of the values is being able to take AC in, combine it with other sources, or not if you don’t have that tech, and and convert it to 800 V DC. And and the drive to 800 V DC is coming from a need to utilize copper in the data center in a far more effective manner. If you’re supplying power to a data center with 415 V AC, which is what, 240 V line to neutral? And and and and and you’ve got what whatever current you’re you’re you’re passing through, there’s a limit on that current. But if you raise that voltage from 240 V AC to 800 V DC, you can push a lot more power down that same conductor. And and I think that coupled with the power density of the rack going up, you’re at a limit with 240 V or 277 V line to neutral voltages and and the copper delivery system, but the 800 V allows you to put far more compute in a denser fashion and deliver it in a in a more economical fashion. Uh not only the resistive losses are going to be better, your currents may be more manageable, your protection system, and so forth. Um hopefully we’ll also uh uh be able to um uh cover the the the different cases. But, 800 volts, that’s what’s driving it, we think. There’s a lot of interest in what the timeline for that is going to look like. Obviously, uh a lot of it is tied to Nvidia’s roadmap for the kind of chips and density that they’re they’re bringing forward. Uh there’s a lot of focus on maybe about a year or two from now, we’ll start seeing Nvidia equipment that’s uh uh uh will support the 800 VDC. Um what’s uh uh from uh your perspective, what do the timelines look like for for DG Matrix and other folks who make uh SSTs who are trying to participate in this, and and how do you see all that evolving? So, we have been working in um in in those reference architectures um of of Nvidia and and coming up with a multi-port SST version for quite a while. And by the time the Vera Rubin’s hit the market with their density in the 800 volt architecture, it’ll be too late to get in. You’ve got to build ahead, you’ve got to design ahead. And and and so, we see this coming not because we have a solution that’s better. We we we see this coming because we’re ardent users of AI ourselves. We use it to develop our software. We use it to design our mechanical parts automatically. We use it for diagnostics. We use it to write our white papers in seconds. We use it to generate presentations. There’s a long list of how DG Matrix has gone AI native and and and and we’re big believers. And the value that we keep getting out of AI is going up faster than we could have ever imagined. So, we think that the the AI value is is just going to continue to grow. And as it does, the demand for AI factories is going to grow exponentially. And and and and so what enables that today is that 800 V architecture with that higher density. The timing of that might be mid-27, maybe late 27, maybe early 2028. But if you’re not ramped up to deliver multiple gigawatts by that time, uh you’re not going to be able to take part of the game. So, you’ve got to be in the game by the time the demand hits. Well, that’s one of the interesting questions for uh end users and folks who who build data centers is uh for them this represents a lot of change. Uh and uh the kind of density you mentioned, how how much denser uh these products are. For many years, the data center industry has been talking about density, talking about the need for liquid cooling, and that density never really arrived by and large. Now, all of a sudden, it’s here. It’s moving really quickly to much higher densities where sort of entirely new architectures are required uh to make it work. So, there’s a lot of people in the data center uh you know, side who are trying to figure out what the next steps are in front of them. And a lot of the the some of the large vendors that work across the power and cooling ecosystem, you know, see a transition that involves uh uh an intermediate step with like side cars that can provide uh local power architecture to help with the 800 uh uh volt transition to the rack while the the rack itself can then be packed with the GPUs and not any power equipment. And then after that a uh perhaps a stage where the SSTs play an even larger role and then you can take a direct 800 VDC rack-level power you know, right out up to the the 34 uh uh KV transformers. Um what’s your what’s your take on uh on those options and and what that will look like? You’re saying you got to get ready now. What does that look like for for folks who are sitting trying to sort out the the 800 VDC transition and how DDM matrix can help them and and where things fit in the uh in the road map? There’s a lot of flux in the industry. There’s a lot of questions on when that DC hits. There’s a lot of questions on the percentage of data centers that’ll be DC versus AC before entire factories go to 800 volts DC. And I think there may be another wave coming in X years where where you go from 800 volts to maybe a higher voltage like 1,500 driven by what’s happening in the in the solar world. So with that uncertainty I think I think that in in you don’t want to have financial uh stranded assets where where you’re losing out on an investment cuz you chose the wrong architecture. And one of the things we’ve developed to help uh a data center developers, neo clouds, and hyperscalers mitigate that is we simultaneously provide AC and DC power in any ratio you want. And and that allows you to have a far more flexible architecture. Or you can buy one of our flexible ports that can be software programmed to go from AC output to DC output. And and I think you you you also have to think about how do you aggregate all that power for for the data center as you supply that combination of AC and DC power, where do you source it from, what behind the meter sources do you use, but all of that is enabled by a power fabric that I think the AI data centers need. And that’s where multi-port SSTs happen to have a very compelling value proposition. That’s the good part. The hard part is, realistically looking at it, you’ve got both early adopters and you’ve got the conservative people in the industry. The early adopters are gung-ho to to deploy the new value. The more conservative ones would like to see X gigawatts deployed or X hundreds of megawatts deployed for 6 months or a year, and and then they will feel more comfortable potentially about the reliability of the unit and whatnot. And and so you’ve got to cater to both. You cannot bring out a system, even for early adopters, that’s going to have a lot of failures. You’ve got to design the reliability into the product. You cannot test the reliability afterwards. So so you really have to think about somebody putting in, you know, $1 billion to $100 billion into an AI factory. And and say, hey, am I going to give them a system that’s really thought out from from the reliability standpoint all the way up the supply chain to my delivery and deployment system. It’s it’s a very very core need, and that’s what’s going to set the the inflection point of how new architectures get adopted. And one of the things I’m you know, sort of tying into that is what are the things that still need to be sorted out uh to to make all these technologies well work together? In terms of the road map for for SST fitting into this 800 VDC architecture. Tell me where you guys are at in terms of timeline and what the key things are that need to be thought about and and solved cuz I know that there’s there’s some of this another element to that ecosystem. What are the challenges here? The challenges are being able to make sure you are working with multiplicity of energy sources. Turbines, natural gas turbines, fuel cells, batteries, multiple manufacturers in each one. You’ve got to have that ecosystem in place. You’ve got to have an ecosystem in place where you have deals with companies that have thousands to tens of thousands of field service agents deployed around the world. And and and you cannot possibly deliver a gigawatt of power to one customer and think your two field service engineers or whatever you have 10 are going to be able to do the job. Right? You’ve got to have a trained workforce that’s used to working on on on a very very safety critical electrical infrastructure. And and and so you’ve got to have deals with those companies especially if you’re rolling out a new tech. Then you’ve got to have a deal with the integrators who can roll out gigantic skids in in their massive factories where they can integrate your your multi-port SSTs along with the with with the other energy sources and they can integrate the switchgear and they can provide one monolithic package that boom can be installed literally in in hours once it leaves their factories on those trucks. And then And got to think about all the commissioning. How do you take months of commissioning today and testing of all these disparate systems and crunch it way down? So, those are some of the things in addition to reliability and certification that must be handled by all the SSD companies. One of the topics that always comes up when we talk about DC power and particularly our medium voltage DC is is safety in the data centers. Uh Uh you know, the old hands always are are you know, cautious about uh uh the the safety implications of this. Um I know you must have have thoughts on this and and both in like the DC power and and particularly the transition that 800 volts and uh and what the you know, the the safety issues are and how to address them. Yeah, I think I think uh uh you know, 60 volts DC and under is thought to be safer, right? That’s where telecom rectifiers come in from 50 years ago and they were all 48 volts, right? Plus 48 or minus 48 and and so you could touch the bus bar and not be worried. Um that’s not the case at above 60 volts. That’s the that’s what UL would regard as lethal voltage or dangerous voltage. 800 volts? You don’t want to go anywhere close to that bus bar. So, you’ve got to think about the safety aspect of of that higher voltage. Number two. You’ve got an issue where if you have a short on a DC bus at a higher voltage you may have a tremendous amount of current that’s going to flow through and cause an arc flash. And the temperatures of arc flashes are very high. They can they can melt copper and you don’t want a a situation where molten copper is spewing all over and and hurting people. So, you’ve got to be able to deal with protection very well. Just like you have AC protection issues. But, in AC protection, you have a zero crossing 60 times a second in in America, 50 times a second in Europe. So, the thought is it’s easier to interrupt. Well, you better have those that DC protection also thought out really well and and thought out to contain the arc flashes and contain the dangers of that while shielding with insulation and and protection equipment for for the people that are in the data center. All that has to be thought out as well as the architectural layout. You’ve announced a number of of partnerships with folks that are you know, familiar to to us on the the data center side. And I wanted to ask about a couple of them. Um one company you’re working with is Exowatt, which is developing solar energy solutions that they’re some of which they’re targeting for the the data center sector. Tell me a little bit about your uh your partnership with them and and what kind of things you’re working on with them on. Sure. So, there there’s some things I can share, right? And and some things I might not be able to share. Hanan, the the the founder and CEO of Exowatt has a brilliant vision to combine solar power with long duration battery. And and so, however they’re doing it in their large monolithic block that he wants to make gigawatts of these blocks and allow his special generators to work off stored heat and and give you give you steady power long after the sun has gone down. Right? So, the issue with solar power always has been, hey you know, the sun doesn’t always shine when you need it to to get power. So, when it goes down, 6-8 hours later, I mean, what what are you going to do? And and so, the battery cost is high, but with Hanan’s solution, you you have an integrated one. Uh but, the engines that he uses, the the generators that he uses have their characteristics of of uh creating that power from thermal to electricity, however he’s doing it. And and so, you take that you take that that generation source, and then you say, “Well, how are we going to interface with the grid?” They may be very different voltages. How are you going to interface with the GenAI servers with all their high cyclic power needs? How are you going to um uh use this How are you going to use supercapacitors with this? How are you going to mate another form of an engine uh like an AC engine to to to work with this a natural gas turbine in case that happens um uh to also be on spot. So, DG Matrix comes in and helps Exowatt integrate um all these different sources and and yet take care of data center power um in a very consolidated manner. That’s that’s the That’s the integrated system that we’re working with them on. Uh you’ve also uh um been working with uh Infra Partners, uh which is a a company that works uh uh very close to the cutting edge on modular data centers uh and modular deployment uh models. Can you tell me a little bit about uh what that relationship’s like and what the opportunities are there? Sure, that’s uh uh Infra Partners is is of course right looking at um more for uh inference uh engines, deployable data centers that are 5-10 MW uh deployable outside corporate campuses all over. They’ve got deals with JLL and Prologis and and and I think Nvidia and Apri. And they uh have uh have selected DG Matrix as um um as of their uh reference architectures, if not the leading reference architecture, because of how much it can do in a very consolidated footprint. The ability to support the grid functionality, the UPS, the rectification, everything that I mentioned, that’s that’s why they’ve chosen us is we’re bringing a software-driven power fabric to power intelligence. And and that’s what they like. And and we like them because it’s a lot of modular data centers all over the country, which removes the risk of one large project coming or not coming to to fruition. Yeah, the whole importance of modular both in terms of the upgradeability of some of these data centers, but also we’ve been writing a lot about some of these uh initiatives that Infraner Partners and Nvidia and others are undertaking with the the distributed modular data centers for inference. That’s I think it’s going to be a really interesting area over the next couple of years. Um we’ve covered a lot of ground. I just to to kind of maybe wrap a little bit and give our listeners you know, something to think about. How would you sort of think about the impact that SSTs are going to have on data centers, but sort of on the larger ecosystem. It’s a it seems to be a pretty big big disruption. It it is a pretty big disruption. You know, for us Bridge the goal is not uh to to to stick SSTs where SSTs uh don’t belong. Our goal is to help the human race get to superhuman intelligence by designing a power fabric for the intelligence age. A power fabric that can solve the power problem and and bringing it in a more cost-effective and faster manner in a manner that’s highly flexible for these for these dense AI loads. That’s our goal. And um and and so so let the let the better architecture win as customers try it out. And I think um because we did our homework in economics and physics, we’re placing the bet on multi-port solid-state transformers of any variety, low voltage or medium voltage. And if folks are interested in learning more about DG Matrix and and your work, uh where can I go? You could go to our website at dgmatrix.com. You can approach us on LinkedIn. You can send an email to info@dgmatrix.com. And uh one of our uh commercial people or one of our application engineers uh will reach out and um and be happy to take down your requirements and and uh with you. Well, Haroun, I I very much appreciate you uh uh taking the time to uh to speak with me and and kind of give us the the download on what you guys are are doing. Uh these are exciting times in the the world of data center and AI infrastructure and uh a lot changing and it’s very cool to see some of the interesting things you guys are working on. You’ve got an excellent uh reputation for for your podcast. You’ve got a great reputation and it’s an honor for us to contribute. Thank you. Well, I thank you. Listen, I I’m having a blast doing it because there is so much to learn so much to talk about. Uh everybody’s keeping busy and and uh so thanks to you and thanks to our listeners. Uh I know that you guys are are keeping really busy as well. If you appreciate conversations like this one, uh just go in. Uh don’t be afraid to give it a like. If you’ve got questions about anything we’ve talked today, uh share them in the comments and we’d love to take a look at those as well. And as always, please be sure to subscribe to the to the podcast or the YouTube channel so that you will always get our content as soon as [music] it’s available here on Data Center Richness. [music] [music]