In this GTC 2026 interview, Jordan Nanos of SemiAnalysis talks with Patrick Oenschlegel, Head of Infrastructure at Radiant, about the company’s newly announced NeoCloud platform. Patrick outlines Radiant’s four‑pillar strategy—software platform, compute hardware, land access, and capital—and describes a global infrastructure roadmap that will fill several hundred megawatts of site capacity with the latest NVIDIA GPU architectures, including upcoming 400 kW GPU racks. He explains how the recent merger with Brookfield resolves the traditional bottlenecks of land acquisition, permitting, and power supply, enabling greenfield data‑center builds in roughly 12 months. The conversation covers power‑supply strategies (grid‑connected versus behind‑the‑meter), the Radiant CloudOS stack that abstracts bare‑metal, VM, and Kubernetes services, the talent model across the software and application layers, and the observability/monitoring framework that underpins high‑SLA uptime. Throughout, Patrick emphasizes flexibility for both large enterprise customers and multiple smaller tenants, and highlights vertical integration as a competitive advantage.
Chapters
Hi everyone, this is Jordan Nanos from SemiAnalysis. I'm here with Patrick Oenschlegel from Radiant, the newly launched NeoCloud. Yeah, welcome.
Well, thank you. Very happy to be here.
Awesome. So why don't we start with a little intro? Just tell everybody where you're from and Yeah, what's new with Radiant? It's brand new, right?
Yeah, it is brand new. It's been announced only a couple of weeks ago. So yeah, I'm Patrick. I'm an engineer at heart. I've been doing high-performance computing for the past 15 years. And yeah, I joined the Radiant adventure about 2 years ago before it got announced as looking after the infrastructure. So basically deploying supercomputers, bringing them to production and operating them. And today I'm at Radiant and I'm the head of infrastructure. So I'm looking after the infrastructure roadmap and procurement and things like that.
So yeah. Awesome. So talk me through the infrastructure roadmap a little bit. I know you guys have big plans. There's existing sites, but you're expanding globally.
Yeah, no, absolutely. I mean, yes, we obviously, we basically Radiant is a new venture, right? It's been, it's built around 4 key pillars. The software platform is one, the compute is very important because this is the underlying infrastructure that we're bringing to market. We've got access to land, we've got access to a lot of capital in order to execute. And our infrastructure roadmap is really global, right? So we've got multiple sites available to us that represent several hundreds of megawatts, and we're looking to fill these sites with the latest NVIDIA architecture. So being here at GTC is really good because we get to see the technology very very closely, right? We're going to see what's coming up, the Vera Rubin architecture and these kinds of things. And also, you know, looking at all these providers doing all the ancillary equipment and CDUs. So combining all these technologies is really nice and very much looking forward to deploying these new clusters globally.
Awesome.
Yeah.
So what do you think is, I guess, the key constraint? You know, there's, you said you've got 4 key pillars. Is there Things that you think you'll have to be working on over the course of this year, next year, as you guys try to expand?
Yeah, no, I mean, we're working on all angles, right? Our software platform is, you know, we've got an engineering team that keeps developing and bringing new capabilities in that. The infrastructure, you know, building the new designs, getting ready for what's coming is also a very important angle. And I think that one of the key constraints until the region announcement, right, was access to land, access to power. and the capital to deploy this. So as the ability for us to merge with Brookfield is incredibly exciting because that resolves that problem, right? And we now can basically take greenfield sites with planning permission. We've got the power plants in order to supply the energy, which is, you know, one of the key bottlenecks in the industry. And we can now, you know, build these data centers and bring this infrastructure to market, you know, in record times. In, you know, about 12 months, we can go from a field to a fully productive AI environment delivering services. So it's incredibly exciting. And yeah.
Can you talk me through maybe the Brookfield portfolio across how that helps? I know, you know, going from greenfield or dirt to data center requires many, many different organizations to contribute something. Equipment, like you said, permitting. Are there special things that you're doing with the Brookfield portfolio there?
Yeah, no, absolutely. I mean, this is one of the key aspects that Brookfield is bringing in this merger, right? So we are part of the Brookfield AI Fund, which has quite a lot of companies and I couldn't name them all, right? But we're obviously looking to partner with these companies. We are also looking to partner with a number of data centers that are also owned and operating with Brookfield, which means that we can start talking to them now about the future capacity that they're going to bring online in the next 24, 36 months and basically design these environments for the upcoming infrastructure that is being rolled out by NVIDIA. I mean, we've had discussions about GPU racks of 400 kilowatts as soon as January next year, right? And All the data centers need to be ready to be able to accept this new infrastructure. So being able to collaborate with these players that are part of the Brookfield portfolio puts us in a really unique position in order to execute on that.
Yeah. How about the sites themselves? So let's talk about maybe power for a second. The big discussions this week at GTC is the difference between grid-connected versus behind-the-meter power generation. Do you guys have preferences there or are there ways in which you can move faster versus, you know, have more durable sites over time that you're thinking about?
Yeah, that's a very good question. I think that connection to the grid is definitely, you know, one aspect, but I think that it can be slower, right? And Brookfield brings that ability to get behind the meter power generation. And when we are looking at these new sites, we're definitely looking about, you know, how we are going to bring that infrastructure, how we're going to get that energy. So yeah, I think that we've got all the cards, right, at the moment. So yeah, it's exciting.
Okay, cool. Maybe can you talk a little bit about talent? You know, you guys are obviously bringing a lot of talent at the software layer. You know, what does that look like maybe across the stack when it comes to actually providing a service to an end user that's going to come in, be a tenant at one of these sites? be that off-taker that people keep talking about?
Whew, broad question. So yeah, I mean, our software stack is very comprehensive, right? I mean, we've got a team dedicated on the lower-level platform, which is basically the ability to take this bare metal infrastructure and turn it into compute, be it bare metal services or virtual machine or Kubernetes environment. And we've got sort of a team that is doing this, which is what we call the Radiant Cloud OS, which is the lower-level platform. And we are looking to bolster that team manyfold. And then there is, you know, all the applications that come at a higher level around data science and model registry and inference services, et cetera, which really augment that infrastructure for our end users. So we also have already quite a significant team doing these things. And yeah, we will bring certainly more of them over the coming months, that's for sure.
Makes sense. Yeah. What do you think that breakdown is between like a large-scale bare metal service that you might provide to, uh, you know, a very large customer versus the paradigm where you've got many small customers who want to consume smaller services?
It's, it's an interesting one. Um, I don't think that there is so much of a difference between the two because even when you're talking to one very large customer, you may still have many teams operating within the organization who would have different needs and different expectations. And basically the idea with our software platform is the ability to reprovision the infrastructure and reshape it based on the needs of the customer, right? Based on, you know, we've been engaged with a lot of enterprise customers and, you know, one day they want to do lots of training. So they will be provisioning the infrastructure as a big monolithic kind of slurp bare metal environments and just run these batch nodes and these batch jobs. And maybe a few months later, a few weeks later, they want to run inference, right? So you want to decommission this and reprovision it in a different fashion. And our software stack is built to handle that level of flexibility. So whether it's one big customer that has multiple little teams or whether it's multiple customers that are coming from a public cloud angle, for instance, From our perspective, it's the same software stack that will do this.
Awesome. Yeah. And earlier you talked a little bit about, you know, monitoring and reliability and how that helps. I think that's a big discussion, uh, for many of these big systems. They're very expensive. They're hard to run. If they're down, it can be even more expensive because now you're paying for something you can't use. Um, maybe can you talk me through anything interesting that you guys might be doing in terms of like reliability and uptime and How vertical integration to these sites that are all under one roof might be different than others that have to bring a bunch of different partners to the table and, and have contracts instead of you guys are just, you know, it's just all employees under the same umbrella, if that makes sense.
Yeah, no, absolutely. So in the Radiant CloudOS, we've got multiple plugins. I mentioned the lower level platform, the AI application services. There's also, as you rightly point out, this whole other angle, which is kind of more vertical about observability, monitoring, logging, triggering alerts, et cetera. So yes, in our software, we have this capability, right? Which basically will feed on data. We will capture all the data from the servers, the switches, the equipments, the racks, even the data center, and compile this into actionable alerts that we can actually operate. And that allows us to effectively ensure that we meet very, very high SLAs in terms of uptime and availability of the systems, which is what enterprise customers rightly ask and deserve to get when Right. So this is critical for us because at the end of the day, right, for secure success, you want to make sure that your customers remain happy. And the way that you make them happy is make sure that, you know, the service remains available at all times. Awesome.
Well, this has been a great interview. Thanks for taking the time. Again, this is Jordan and Patrick. I'm signing off for SemiAnalysis and Sail from GTC 2026. Thanks for watching. Thank you.
