Solving AI Data Center Power at the Point of Load

Hey, everyone. " I've got a new company for our audience and for us here on "Techstrong TV" to introduce you to, as well as their CEO. I want you all to say hello to Hans Hasselbalch Andersen.

Hans is the CEO for Lotus Microsystems. And I got to tell you the truth, Hans, when I first saw it, I was instantly transported back to Lotus, and Notes. Because I got started back in the dot com days, I sold my first company to a roll-up that became what we call an ASP, application service provider.

And our claim to fame is we were, besides IBM themselves, the largest host for Lotus Notes in the world. So Hans, this is before cloud, before virtualization, before bandwidth was cheap. Notes was great, but we were making it up as we go there.

And it was a chore, to say the least. But I'm glad- I remember that ... you're old enough to remember.

And Lotus 1-2-3, and so on. Absolutely. Look, it was still the best spreadsheet they ever made, right?

Yeah. You know what they said, "Windows ain't done till Lotus won't run," right? That was the old saying back then.

But I'm glad to see the Lotus name out here again, and of course, it's Lotus Microsystems. Hans, before we jump in, I want to do a deep dive on Lotus, but I like to give people a sense, so they know who they're talking to. Right.

So if you wouldn't mind, if you give them a little background on how you came to be CEO here. Sure. No problem.

So I've spent more than 25 years in the semiconductor industry. Entered into that industry through an acquisition, actually, all the way back in around the year 2000. I was in a startup that got acquired by TI at that time.

Mm-hmm. So that's my way into semiconductors, and I stayed on with TI for a number of years. Then I jumped out and started my own company, doing audio amplifiers, integrated audio amplifiers actually, for consumer applications.

That we built up as a classical fabless semiconductor company, and it was acquired by Infineon back in 2018. Yeah. Really?

Well, yeah. And shortly after that, that's actually where I started my first, you could say, started to engage with the guys that would become the founders of Lotus Microsystems. At that time, they were doing their PhDs at the Technical University here in Copenhagen, and they knew that they were going to spin out a company at some point.

So they wanted to establish an advisory board already then. So I joined that and helped them discuss how to go about that. And then we didn't really see each other for a while until they reached out, what?

Three years ago, actually, and were looking for a CEO at that time. So to make a long story short, I decided to jump on board after I heard what exciting things they were working on. And I thought that's good enough for me to both join as CEO, but also actually as an investor.

So I also invested in the company. That's great. Yeah.

That's fantastic. Great story. So it sounds like Lotus, the underlying technology research was kind of germinated, if you will, at the university where the two co-founders were doing PhD.

And when you talk about highly, highly technical kind of thing, businesses, it's a pretty common fact pattern, right? Yeah. Because the universities do great research, and then you throw off commercial entities.

Yeah. What is the special sauce? Well, tell us the Lotus story.

Beyond what you told us, from a tech point of view, what is the Lotus story? Yeah. So we, as a company, we develop fully integrated power modules, right?

And we focus on the AI and data center applications. But what we do is fully integrated power modules, meaning that we put our own power management ICs in there. We combine it with the passive components that are required for the operation of the power converter.

And then you need a substrate. And this is what you can call our secret sauce. That's what we call our power interposer technology.

And what that is actually is, it's a silicon-based interposer like, you already know silicon interposers from the industry. They're used to combine XPUs with memory modules on the same silicon substrate. What we do, to put it in plain words, we take that concept into the power domain.

So instead of using the interposer for, you could say, transporting a huge amount of data, logic level signals through the interposer between memory and the XPU and so on, we take that concept into the power domain, meaning that we can place a power converter on the same silicon substrate as, let's say, inductors or capacitors. And then we can move large currents through the interposer. But that requires special manufacturing, a special manufacturing process for doing these types of interposers, especially through silicon Vias and RDL structures.

Those are the critical elements to, you could say, scale for power. Excellent. In some ways, Lotus is hitting head on two of the biggest friction points, if you will, in this whole data center debate, and that is power and heating.

Yeah. Exactly. And so I'm wondering, well, first of all, thank you for coming on here, because let's get the word out about that because there's a lot of people who think these problems have no solutions.

Mm. But I got to ask another question. One of the things we're seeing a lot of, whether it's Nvidia or AMD or any of them, they're not just satisfied with the GPU or the CPU, but they're actually kind of selling pre-configured racks, if you will.

Yeah. That have power and heat. Mm.

So is Lotus kind of built into those racks? Are your markets the people who aren't using these prefabbed racks, if you will? Mm.

Yeah. Well, if you want to place us in the power chain, you could call it that, right? You have power all the way from the grid and into the delivering the last stage, the power directly to the GPU, or the CPU, or whatever.

And that takes place through a number of conversion steps, basically, all the way from the grid and down to the core. And we are the last stage before the core, basically. So working at the point of load currently, conversion.

So right before- So before the rack, if you will. No, on the motherboard. On the motherboard, okay.

Sitting right next to the GPU or CPU or XPU or whatever. The last stage of power conversion. That's what we are working on and developing right now.

I love it. Yeah. And when you get to that- Modern AI processors are manufactured in the most advanced processes.

They're two or three nanometers and so on. And that means that the supply voltage is also very, very small. 7 volts or something like that.

But if these beasts require two kilowatts of power, then with a fixed low supply voltage, that converts into a massive amount of current. And this is really where the challenges start to amount up, because any type of a resistive loss in the traces on the circuit board will just generate a huge amount of heat. And that's what's happening, actually, and that's where you have the heat issue, right?

Mm-hmm. Generated right next to the XPU. That's where you want it.

Yeah, that's what we are looking at. That's a huge challenge there to deliver, I would say, enough power, but also do it in a way where you minimize the thermal issues related to that operation. Where is Lotus in terms-- Do you have actual product out there?

Still finishing? Yeah. So just a couple of months ago, we announced the upcoming sampling of our first point-of-load converter specifically for these applications.

So you could say the platform, which is going to consist of several products, is called Beast Rider, and the first product on that platform is called the LSC0580. That's the point-of-load converter that I'm talking about here. And we are just preparing right now, we're in the final stages of manufacturing the first batch of that module, and we expect to sample it within the next few months or so.

So that's obviously a major milestone. Exciting. Yeah.

Definitely a major milestone for us to get that out and start sampling it to customers. Now, from an alliances, partnership point of view, what systems would this run on? So, right now, we're working with XPU providers and hyperscalers.

Most or if not even all the hyperscalers, they do their own processors for AI- Mm-hmm ... processing. So we are talking to most of them, and fundamentally, it's their design, right?

The motherboard, the tray design is on their side. Obviously, we need to understand where our device will, how it'll fit into those next generations of server tray designs. So obviously we have a close communication where we exchange, you could say, we get requirements, we get what are they looking for, what kind of performance are they looking for, and then we convert that into our product specifications and develop our solutions from that point on.

But, eventually our modules will sit on the motherboard with the CPU, GPU, and feed it with power. Excellent. Hans, we were talking off camera before we started today.

This whole notion of these giant AI data centers- Yeah ... have gone from the darlings of government and politicians because they saw dollar signs on how much investment is required, and taxes, not a lot of jobs being thrown off, but still massive amounts of money. We're talking worldwide, $8 trillion or something like this.

I think that 8 trillion doesn't include China, actually, which may be the biggest of them all. Yeah. But now, in the last couple of months, I know here in the US as well as in Europe, we're starting to see pushback from local groups.

Not in my backyard, right? Yeah. The NIMBYs.

Yeah. And two of the biggest objections is that of power usage. We can't run this on the grid, and if you're going to run it off grid, it's got to be clean, it's got to be efficient, everything.

Yeah. And then secondly, water cooling. Yeah.

There's all of these things that here in the US, we use more water watering golf courses than we do data centers, right? Mm-hmm. But water is a critical element for life, and it's also still a critical element for data centers.

Yeah. Is that something Lotus can seize on, right? " Yeah, it definitely is.

Even though those considerations are, it might seem far away from what we do here, but it actually trickles all the way back to what we do. And that's simply because if you look at issues that are becoming sensitive, like in terms of a grid allocation of who's first in line for more power out of the grid? Is it a data center, or is it charging stations for electrical vehicles?

How do you address that? The bottom line is that those topics are being discussed right now. So for us, it means that what we need to focus on is, I would say there are primarily two things that we need to optimize for.

One is efficiency, right? Meaning, how do we maximize the compute power per watt, basically, right? How do we get as much computing power for every single watt that we put into the system?

And the way it is today, if you look at an average data center, 35% to 40% of the power that goes into a data center is used for cooling, which is obviously, nobody wants that level of, you could say, spending energy on something that is an unwanted side effect of running a data center, right? So, getting that down, in the industry today, you already see that there's a whole range of companies focusing on cooling solutions. And that's necessary, of course.

The way we look at that is that we see that as treating the symptoms and not the cure. Instead of the disease. Yeah, not curing the illness, right?

Because if you work with cooling, it's because the problem has already shown itself, right? So we work on creating high-efficiency power conversion, basically. And that's one of our trademarks about what we do here.

So your computational power per watt, but also computational power per square foot, basically. We talk about compute density, basically. How can you squeeze more compute power into a rack which has a certain layout and a certain size?

So those are, I think, things that convert all the way from demands from society to optimize the spending and the use of scarce resources like energy and water. That trickles all the way back to what we do here, actually. Excellent.

Hans, we're almost out of time, but we didn't even mention the website. For people who want to maybe dig in and get some more information, where can they go? com.

We are quite active on LinkedIn as well, so I really recommend anybody who want to follow what we do to follow us there. That's where we post all of our news. So LinkedIn is a good place to follow Lotus Microsystems.

Love it. I wish you a lot of success with Lotus Microsystems, Hans. Thank you.

It sounds like a great story. Continue to follow it. " Thanks for having me.

My pleasure. " We're going to take a break. We've got more coming at you.