When GPUs sing together: Solving the AI data centre power spike

Ruth: Nyobolt spun out of the University of Cambridge in 2019, I think. And it was a material science lab and turned into a commercial company.

Can you walk me through that journey? Where did it start and where are you now?

Ramesh: Basically, Nyobolt's mission from day one never changed, right? It was always about delivering, instant power, eliminating downtime, in operations where, you know, every second actually counts. So that was always the vision and mission for the company.

The first step to that was the materials. So we started focusing on the materials, which is a key ingredient. And as we were developing and then as we were interacting with the customers, you know, what we realised was that customers, basically, they wanted this problem to be solved rather than just one part of the solution.

Mm-hmm. Uh, so we actually evolved from a materials company to what we are today, and I will go through that story, but basically what we learned was, you know, don't confuse your product with purpose. Right. Um, that was really an important lesson for us as we evolved as a company. Purpose was always delivering instant power and eliminating downtime or reducing downtime and how we approach that, whether it's through materials or through the end product, you know, it didn't matter.

Ruth: And what forced that shift? Was it the customer, the market, or something you discovered internally?

Ramesh: Yeah, I mean, it was certainly, you know, market dictated that to a large extent. We believed at that time that advanced battery materials, would be, would be, enough, to have those discussions.

But customers were not buying materials, right? They were buying a, a full solution. So they wanted a, like a complete system. They wanted guaranteed, you know, performance integration that is much faster. And they certainly were not an expert in this area, so they wanted to reduce their risk, so they were looking for a complete solution, which then made us evolve.

But. The first demonstration that we did was with, actually with a vacuum cleaner. Uh mm-hmm. You know, those, iRobot like things, right? The one that you can automate. We swapped the battery, in one of them as a prototype to demonstrate what, our technology can do. I think at that time, the robot could run for an hour and needed about three hours of charging.

Ruth: No.

Ramesh: Um, whereas with hours, it ran for 30 minutes, but with three minutes of charging, so, and it was half the pack, so you can increase it to, to six minute and it would've run for an hour, as opposed to a three hour. That's how we started, because we needed to showcase what the underlying technology can do, and it was all about addressing that downtime and delivering that power.

That was always the, the foundation of the company.

Ruth: And before we get to AI data centres, what does Nyobolt's battery actually do? What does it do differently than conventional lithium ion batteries? What's your secret sauce?

Ramesh: Yes. We do have a, a secret sauce and that, basically enables very high fast charging capability.

There are others who do fast charging capability, but it does not affect the life cycle of the product. And that was really important because we always focused on the total cost of ownership. So delivering instant power, eliminating downtime at a cost that is economically competitive over the lifecycle.

That was really important to us. And then as we evolved, you know, we got into the products, which, you know, as, as I said earlier, customers don't buy technology, right? They buy outcomes. And that's what we started focusing on is, you know, where are the industries where this is critical and what is the outcome that we can actually deliver?

And combining that low downtime or zero downtime with a long lifecycle, was, was really critical. And that's the differentiation.

Ruth: And you mentioned something about how fast you, your battery recharges, I think I read it's 80% charge in just under five minutes. And you even mentioned under three minutes.

Those are striking numbers. How do you achieve that?

Ramesh: Yeah, so different applications have slightly different charge rates, and what I was referring to was the first product that we actually developed, which was the vacuum cleaner as a demo. But we, we then of course, as any, any, company with, you know, where we have a lot of belief in the outcome, we said, you know, we need to demonstrate this in a proof point where people can actually understand it even better.

So we actually built our own Nyobolt car as a demonstrator, and that's where we showcase that you could actually charge the car within less than five minutes. That was 4.37 minutes to 80% of the battery capacity. And it was actually limited by the fact that the charger wasn't, you know, big enough, was a 350 kilowatt charger.

Ours can take much more power than that. So it, it was again, to showcase when, when you say you can charge your car under five minutes, where today, you know, it's, it's, I mean it is of course evolving. It is changing and, and, and I think others are, you know, coming up with the, with the slightly, faster capacity, but it is still 30 to 40 minutes, you know, charge time.

Mm-hmm. Uh, whereas, you know, under five minutes is essentially like filling up petrol Now, you know, go grab a coffee while the car is charging and you can continue driving.

Ruth: Yeah.

Ramesh: And what it actually means is you don't need to put these huge heavy batteries in the vehicle when you don't need it. I mean, if you, when people say you have a thousand miles.

It actually means you are carrying a lot of weight in the car. Mm-hmm. And nobody drives a thousand miles without a break. And if you can do that in a five minutes, that suits most of the driving requirement.

Ruth: Yeah.

Ramesh: It reduces the cost, reduces the weight, reduces the wear and tear, and more importantly, less impact on the planet as well.

So that, that's what, the aim was to demonstrate how actually we need to think about it differently.

Ruth: Yeah. We did have a guest on the show from, from a battery company as well, and, we had discussed the topic of reach that people are obsessed with reach when it comes to electric vehicles and that it actually doesn't really matter because no one drives that long, as you mentioned, and especially not without a break.

And, yeah, the need for it is just like a mental thing that, people would like to, I need my car to be able to drive a thousand miles without stopping, which is actually not very useful at all. Yeah.

Ramesh: Yeah. I think the key is gonna be having that infrastructure, charging infrastructure. So the focus really is if we get the infrastructure right, we can be much more efficient with all of these technology going in that enables fast charging.

Ruth: Let us move from vacuum robots to, electric vehicles now to AI data centres. AI data centres are a very polarising topic right now. The power demand is just enormous and it keeps on growing. Paint us a picture of the actual problem. What is happening inside a data centre that existing power infrastructure just cannot handle.

Ramesh: Yeah, it's not data centre, it's the AI data centre. And what happens is when the GPU is all synchronised and start singing together, the power demand spikes, exponentially. And that puts huge pressure on the infrastructure, on the grid. And that, you know, ultimately affects the communities that is, near these, AI data centres.

And, and that's a real problem. What are data centres doing is if it's a hundred megawatt, they oversize the grid. They are utilising GPUs at a much lower capacity through artificial constant utilisation, stuff like that. So, so you're not having high productivity from a GPU perspective, which is, you know, quite expensive.

You're oversizing, you know, the grid facilities, which means it's more expensive and you're also incurring a higher operating cost for it. Right. So the whole ecosystem is under pressure because of those spikes. And basically our solution is to actually address that very spike. Um, what we have done is, as I said, the the vision hasn't changed.

It's the instant power, right? So our technology suits perfectly well within this. Um, it addresses that spike, it absorbs the spike and provides the power, and it is required, to support the spike and then gets charged when the spike goes down and it responds within milliseconds. And it can provide power for 60 to 90 seconds, which is all that the spike lasts four seconds, you know, very rare it goes to 30 seconds.

But yeah, we can provide that up to, you know, 60 to 90 seconds. And it's, it's a rack based solution. So it's with, with the service. Mm-hmm. Um, what that means is the latency of power delivery towards the spike is minimal. So what that does is, one, it absorbs the spike. It takes away oversizing, the, the grid facilities.

It has a better GPU leads to a better GPU utilisation and also lowering, you know, the both capital cost and the operating expenses as well. So it addresses the whole ecosystem to make it much more efficient, overall and more importantly from, you know, general community purposes takes that pressure off the grid so you don't have blackouts, certainly not due to the spikes anymore.

Ruth: Mm-hmm. And that, if I understand it correctly, is also the d differentiator between a conventional UPS and your solution, right?

Ramesh: Yeah. Um, as just mentioned, the UPS is is, you know, is slightly further away. So there is latency and mm-hmm. Normally they do not react. I mean, ours, our solution reacts in milliseconds.

You know, UPS typically does not, react that fast and the spikes happen so quickly that you do need that, you know, instant reaction. So one, it addresses that. Of course you have super caps, but super caps don't have, cannot provide power for 30 seconds or 60 seconds. Uh, you know, so we have got super cap type of response time, but with a battery, like, you know, backup time.

Mm-hmm. So it combines both. It's a total power solution.

Ruth: We will take a short break, stay with us, and we will be hearing from our guests very shortly. This podcast is brought to you by Avnet Silica, the engineers of Evolution. Subscribe to our Avnet Silica newsletter or connect with us on LinkedIn. If you want to learn more about us, we have put information and links in this episode.

Show notes.

You recently signed, a memorandum of understanding with the state of Rajasthan for more than 100 megawatts of off grid AI data centres. What does off grid actually mean in practise?

Ramesh: We are looking at expanding our manufacturing capacity

Ramesh: To support the data centre solution. So we call our product DRS, our Dynamic Response System.

We are seeing a lot of demand for this product and we want to be ready. And so we were looking at setting up the facility and we were looking at solar, powering it so that we are actually sustainable throughout our manufacturing process as well. So that's what it meant by off-grid.

Ruth: Mm-hmm. Yeah. For a startup, you do have a lot of interesting use cases.

Can you share some, some other stories with us? Um, or some other

Ramesh: Yeah.

Ruth: Other interesting projects you have been, working on at the moment?

Ramesh: Yeah, no, sure. I mean, one is, already known in the public, in physical AI space. This is a, you know, we play a critical solution. Uh, we actually announced our partnership with Symbotic, you know, where we are charging it again in less than 10 seconds.

The bot runs for a certain period of time, and then it charges again in 10 seconds. So basically the charging happens while the bot is loading and unloading. Mm-hmm. So basically there's zero downtime because it is still doing other things while the charging is happening. 'cause it is that quick.

Ruth: And SymBots are warehouse robots.

Yes.

Ramesh: Yeah. So Symbotic is a, is a company that does warehouse robots and they are a, a major supplier to, to Walmart, in, in US and, and, and, south Americas. Wow.

Ruth: So,

Ramesh: so, so yeah, that's one solution. We are, I can't name because we haven't announced it, but we are also working with a partner in terms of humanoid, which is an enabling them to increase the uptime of the robot by about, 70%.

So productive time of a humanoid, you know, by 70% higher makes TCO the total cost of ownership. You know, much more interesting to adopt that kind of technology. So, and that's why I said earlier on, it's all about TCO. So the uptime delivers, addressing the economic problem of moving towards a sustainable solution.

Ruth: Mm-hmm. Yeah. That's really, those are really impressive numbers. Yeah. It's not only faster, it's lighter, it's cheaper, and it's more reliable. Yeah. I must ask, are there any trade-offs to, to your solution?

Ramesh: You know, it depends on the solution. Uh, you know, what the solution is for where there is an instant power, where there is downtime that needs to be addressed.

Ramesh: No, but there are instances when our solution is not required, right. Where high energy density is, what is the solution. So it, that has its place and know we have slightly different focus, and they both do different things. So we, we are not in the space of high energy density and high energy density cannot do what we are trying to, to solve and, and address.

Right. So it just depends on what is the customer need, what is the solution that they're looking for, and how can we help them, with that solution.

Ruth: And are there any constraints regarding the scaling.

Ramesh: Uh, no, I mean, we have a very diversified supply chain from Asia, in and out of China. We have enough, capacity and, supply chain capability to scale very, very rapidly.

Uh, and when I say rapidly, you know, between, depending on what we are talking about, between three to nine months, essentially we could scale up quite, quite quickly.

Ruth: Data centres, warehouse, robotics, heavy duty, automotive grid storage. That's a wide front. Where do you focus first and why?

Ramesh: So, our focus is always, as I said, is to eliminate that downtime.

The application. While it is different segments, the underlying technology that we are, utilising is our strength. You know, we work very, very closely with customers to provide customised solutions. So our priority is to deliver what the customer needs rather than worrying about what segment they are in.

Ruth: And, looking ahead into the future, where do you think, the road will lead for Nyobolt?

Ramesh: I, I think, the key challenge is for people to even think about the operations differently, to understand how a technology like ours can eliminate and address their downtime. And in a lot of times it takes few discussions to.

For them to open up to, to that thinking of their operations in a completely different way. And I think that's the, the key is the more we roll out our technology, the more the industries will start adapting this and understanding how a fast charging battery solutions and power solutions can actually support and help them and make them more efficient.

Mm-hmm. So I'm hoping that's what, you know, that's where we are heading towards.

Ruth: Hmm. Interesting. You mentioned that, the customers sometimes are a little bit hesitant to open up. Why do you think that is?

Ramesh: I, I, I didn't say hesitant. I think it's that they just haven't understood because they haven't been exposed to the technology.

So how does the fast charging can actually change their operations? So when you think about approaching the solution differently, it opens up a completely different way of going about, doing what you're doing today. So that will happen. It will take time, but that's what we are heading towards.

Ruth: Terrific. Is there anything we haven't talked about that you wish we had talked about?

Ramesh: I mean, of course, you know, one thing that, while, while we talk about Nyobolt and the interesting technologies, we have any and all companies are made by either the talents that you have. And, you know, being in Cambridge, having the ecosystem around Cambridge has been extremely helpful and, and, and useful.

And tapping into the talent here, especially in the material science and, and into the engineering side, has been very, very instrumental in us, you know, growing and evolving as a company.

Ruth: Yeah, it all has started with a big research project. Yes,

Ramesh: yes. Yeah. But, but the key is access to talent and then, you know, and, and the talents that we have with the Nyobolt has been instrumental in growing.

Uh, as, as a company.

Ruth: If you had to put together a soundtrack for this episode, what song would you put on it?

Ramesh: It's a, it was in a Hindi movie, but it, it means, you know, be victorious. Uh, and it's a song called Jai Ho, which means, you know, victory to us. So that's what I would, you know,

Ruth: that's a terrific addition.

Thank you very much.

Ramesh: Thank you.

Ruth: It was really interesting to hear how customer demand shaped Nyobolt's evolution from material science into a full infrastructure provider. The AI data centre power problem is real, and it is really good to hear that the solution is on the way. Thank you for being on the show, Ramesh, and thank you for sharing your insights and your expertise with us.

Ramesh: Thank you. Pleasure being here.

Ruth: Thank you for listening to We Talk IoT, stay curious and keep innovating. This was Avnet Silica's We Talk IoT. If you enjoyed this episode, please subscribe and leave a rating. Talk to you soon.