Pure Accelerate 2026 | Omar Lari, Crusoe

Clips
More from Pure Accelerate 2026

Omar Lari

Senior Product Leader

Crusoe

play_circle_outline Crusoe AI's Energy-First Model: Spark GPUs Powered by Solar and Recycled EV Batteries with Redwood Materials

play_circle_outline Iceland data center powered by geothermal and hydroelectric energy.

play_circle_outline Importance of reliability, performance, and operational excellence for large GPU clusters.

play_circle_outline Challenges of scaling GPUs: electrical grid stress and high-bandwidth memory performance needs.

Info
Transcript

Omar Lari, Crusoe

Omar Lari

Senior Product Leader Crusoe

In this interview from Pure Accelerate 2026, Omar Lari, senior director of product management at Crusoe Energy Systems, joins theCUBE's Christophe Bertrand and Alison Kosik to discuss how an energy-first approach to AI infrastructure is reshaping what it means to build and operate AI factories at scale. Lari explains that Crusoe's fully vertically integrated model — sourcing energy, building data centers near that energy and deploying infrastructure and managed AI services on top — enables deployments in unconventional locations like Abilene, Texas using wind... Read more

explore Keep Exploring

What is Crusoe's mission, and how does its energy‑first, vertically integrated approach differentiate the company and enable products like Spark? add

How is AI infrastructure being powered with renewable or recycled energy sources, and what examples show that vertical integration in practice? add

What technology and capabilities does your company/platform provide to support large-scale AI training, inference, and the operational needs of AI-native customers? add

- To what extent should enterprises trust AI-native companies with their data so those companies can embed intelligence back into the organization? - How much of the future of AI will be determined by access to power (energy and infrastructure) rather than by technology alone? add

bolt Powered by CUBE AI

Omar Lari, Crusoe

search

Alison Kosik

>> Welcome to Pure Accelerate 2026. We are streaming live right here in Las Vegas. I'm Alison Kosick, joined alongside by Christophe Bertrand. And we're about to bring in a guest who represents a partnership with Everpure.

Christophe Bertrand

>> Absolutely. And I'm very, very curious to see what Omar has to tell us, especially around green data centers, around AI infrastructure, and around the partnership with Pure.

Alison Kosik

>> Yeah. There's also a lot in the headlines about what you just talked about. Let's bring in Omar Lari. He is the senior director of infrastructure as a service at Crusoe AI. Welcome to theCUBE.

Omar Lari

>> Thanks for having me.

Alison Kosik

>> Talk me through what Crusoe does and what makes your business and your business model different from other AI cloud providers.

Omar Lari

>> Yeah. Crusoe's mission is to accelerate the abundance of energy and intelligence. We really think that those things have a symbiotic relationship with each other. Energy is going to drive the next breakthroughs in AI. AI will eventually help us make the next breakthroughs in energy. What differentiates us is we take an energy first approach, and we're fully vertically integrated. What that means is we source our energy, we build the data centers close to that energy, and then deploy our infrastructure as a service and manage AI services on top of that. It's a really unique way for us to be able to take advantage of power that may be in secluded regions. If you look at some of the things that we've done in like Abilene, Texas, using wind and natural gas. Not typical places that you would see a data center getting built, right? Not a lot of fiber, a lot of interconnectivity. But AI allows us to do different types of use cases that really allows us to take advantage of that energy. I give you one good example of this vertical integration is our Spark product. And what we've done is we've partnered with Redwood Materials, and we have a deployment of thousands of Blackwell GPUs that are all powered by recycled EV batteries, which are charged by solar energy. It's a really unique way to see that full integration end to end, and allows us to meet customers where they are in any part of the stack where they want to consume our services.

Alison Kosik

>> It really sounds like you're blazing the trail of building infrastructure using renewable energy sources, something that I'm sure many others are struggling with.

Omar Lari

>> Yeah. I think that's part of our core DNA is always being able to use sustainable or renewable energy sources. Another example of that is our Iceland data center, it's powered purely by geothermal and hydroelectric. We have tens of thousands of GPUs operating there, dozens of customers in a very large GPU footprint. And so, we're always exploring different ways. We have a partnership with GE Vernova, Boom Hypersonic, and a lot of different partnerships with different unique types of energy sources, and being able to really deploy AI infrastructure where we can secure that energy.

Christophe Bertrand

>> Yeah. And as Alison was mentioning, it's a pretty timely conversation to be having. There's a lot of controversy around those big data centers being built for AI and what the power constraints will be, what the impact will be in populations, et cetera. And in your case, you're really solving for two problems. You're solving for the energy component of it and the green part of it, recycling batteries, and using solar power, wind, et cetera, geothermal. But let's switch gears to now. Okay, we're doing this for AI, right? Operationally, if you think about your customers, what makes the difference between somebody who's going to be successful turning essentially a lab project into an AI at scale type of operation, leveraging your technology and your capabilities?

Omar Lari

>> Yeah. That's a really good question. I think, being a neocloud, many of our customers are AI natives. That's where we see the largest amount of investment coming from today. I think their first mission is to get large scale clusters for training online as quickly as possible. The things that they value are reliability, performance, and operational excellence, right? Obviously, they're spending a huge amount of CapEx on the GPU infrastructure. They want it to run reliability and they want it to be up. From a performance perspective, the entire AI stack is really important. It's all the way from the application layer, from PyTorch and TensorFlow to the CUDA drivers, all the way down to the metal and to the bare metal hardware, and being able to run that in a performant way. Think about if you have 800 Blackwell GPUs, that represents about a quarter petabyte of high bandwidth memory, and they want that to run really quickly. If you're waiting 20, 30 minutes for that to get hydrated, you're burning through an enormous amount of capital just waiting for bytes to float around. Performance is a really important piece. Operational excellence is another one. Again, the coupling between the application layer and the actual hardware is super important. And all of these things really need to be conducted as an orchestra because they all have to work in unison. If you make one little change, it could have widespread impact. And so, those are the things that our customers value. The thing that they're trying to do is run massive training jobs of these neural networks so that they can find product market fit and eventually start generating revenue. Being able to run that quickly, get their researchers out of the infrastructure management business and really focus on the data science and the AI parts of this, and then transition that into actually inference and where they start generating revenue. At that point, they're not just supporting research, they're supporting a business, and they're supporting a revenue stream. And that's what we've typically seen from most of the AI native customers that we work with.

Alison Kosik

>> Walk us through the decision to include Everpure as part of your strategy.

Omar Lari

>> Yeah. It was a lot more than just a strict technology decision. Obviously, we need features and functionality, and that's a really important part of it. But there's three things that we really value from Everpure. One, I would say, is that they have a really strong reputation for reliability. And they've been running enterprise workloads for several decades really successfully and have really good reputation in the market. Our customers also value that reliability from us, and being able to leverage a partner that has those same characteristics was really important for us. Two is that they really have a lot of empathy for the service provider. I think we're maybe a little bit of a unique customer for Everpure. We're not buying appliances and using it for our own enterprise data workloads, right? We're actually running multi-tenant infrastructure for several companies and several customers that have all of those needs and requirements that I talked about before. And they really understand the unique challenges that service providers have in that space. And then the third thing that I would say is that they really understand the supply chain. And this is not just in terms of capacity. Obviously, we need to get capacity online and we need to get that online quickly and predictably to get the outcomes that we want, but they understand the technology piece of that. And supply chain is so important in putting together these AI factories. And their understanding of that space and being able to really understand not just the capacity and the technology is really beneficial to us, and being able to outsource that to a company who has expertise in that. We have the expertise in data centers and GPUs and compute infrastructure. Everpure has a lot of operational expertise in the storage space, and partnering with that really helps us meet our mission of accelerating the abundance of energy and intelligence.

Christophe Bertrand

>> Right. And this is a global business, and you're a global business, so clearly there's a very good match there, very good fit. I'm curious about your customers in general. Certainly, a lot of adoption of AI going on these days. From your standpoint, what do you see as maybe industries or the type of customers that may be more successful than others? What do they do maybe differently? How do they approach it to be more successful leveraging the partnership?

Omar Lari

>> Yeah. Again, we're in a space where we're serving a lot of just new AI native companies, and it's breakthrough technology for them. They're really about, how do we develop the next frontier lab models? How do we get that into the hands of customers? Their main intellectual property and business value is really centered around model weights and user data, and iterating on that. Enterprise, on the other hand, is very different. Enterprise, they're not out there trying to build the next ChatGPT or Opus model, or frontier model. They have a different set of challenges. Their intellectual property is all in the decades of data that they've accumulated, the processes that they have, the people and the expertise. And I think that's really the opportunity over the next several years is, how do we bring those two worlds together and make sure that the enterprises are able to trust the AI natives with the data that they have so that they can really embed that intelligence back into that organization? The AI natives would love to have that data, they could build verticalized solutions that are specific to different types of industry. But from my perspective, I think a lot of these things are still early and emerging. A lot of the innovation, obviously in the AI native space around the model building and the inference. And then, how do we bridge that and bring that back to the enterprise?

Alison Kosik

>> Crusoe seems to have embraced an energy first strategy, right? How much of the future of AI is going to be determined, not necessarily by the technology, but by access to power?

Omar Lari

>> I think it's the number one thing that the entire industry is going after today is, how can we build out as much capacity as possible? But that's just one part of it. Getting that allocation to the capacity and the energy is the first step. And building specific AI factories based on that from the technical specifications that we see from our partners and the new technologies and the new infrastructure is really what's going to unlock the value of that power. Just having access to the energy and no expertise and actually running that infrastructure does us no good. Our vision is to bring all of those pieces together, provide the energy, the data centers, the infrastructure, and we even have managed AI services on top of that. And being able to provide that end to end expertise is really important for us. And we see really unique things as part of that. Some of the things that we've experienced we don't think other customers see. If you bring up hundreds of thousands of GPUs simultaneously, the stress that puts on the electrical subcomponents and on the grid. There's a lot of innovation that's happening in that space as well, and allowing us to see that full picture end to end lets us optimize that entire stack. And importantly, it lets customers enter at any layer of that stack that they want to, whether that's managed AI or whether that's infrastructure services.

Christophe Bertrand

>> Yeah. I have to say it's very unique. Because what you're describing is, yes, the best of both worlds in terms of the innovators building the next big AI thing versus the maybe more traditional enterprises. Yet I see another angle, which is that leadership through what would be a conscious and a good citizen policy on consumption of energy, optimization of energy, building all of that infrastructure, the expertise you're developing is pretty substantial. And I can see how many enterprises don't want to do that. It's not their job. Their job is to serve the customers with whatever it is that they provide. I think that puts you in a very interesting position. Now, let's think about the keynote. The keynote was not even a couple of hours ago. What have you heard that got your attention today?

Omar Lari

>> Yeah. I think a lot of the things that we've talked about around data governance and security. And how do we actually make these decades of accumulated data and expertise actually relevant to AI systems? And I think that's the hardest part. And I think that's where I'm seeing a lot of the innovation happen today. There was a lot of talk about agents. In my opinion, agents are still a little bit early in their days. They're like glorified workflows and knowledge management systems. But we're getting there, and we're starting to see more and more usage of that. The software development life cycle is probably the best example of where it's been really successful. A lot of these AI native companies, you take Cursor for example, why were they so successful? There was a bunch of developers building for other developers. And they had their own expertise, they know what software engineers want. They know the features and capabilities that are required to build them. I think the same is going to happen in the enterprise, like having something for manufacturing or for healthcare or financial services. As a software developer, I don't have the expertise in financial services or healthcare, or what happens in those businesses. You really have to understand the details to unlock the value of AI and the intelligence that's being provided.

Christophe Bertrand

>> And Omar, I have a quick follow up on that. Because do you see... And again, you don't have to answer necessarily a big strategy question here. But do you see this notion of providing services around context and intelligent data management literally becomes a new set of offerings, or at least guidance you can provide?

Omar Lari

>> Yeah, absolutely. We work really closely with Nvidia, and we appreciate the partnership with them. And as they take things from reference architecture into production, we always evaluate those things based on what our customers are asking for us. What I really like about their ecosystem is that they have a lot of breadth and depth, but these things are loosely coupled. We can pick the modular pieces that our customers actually need from us, whether that's RAG or context or agents, and build those offerings on behalf of our customers. Same thing with the customers don't want to build data centers, they don't want to deploy IAS, they don't want to deploy managed AI service. They may not want to deploy some of these data streaming services and the data management systems themselves because there's a lot of operational overhead that's associated with it. Our job is to pick the right ones and make the right bets because we need to run those reliably for our customers. We can't pick everything out of the ecosystem, we have to see the technologies that actually emerge, the ones that are going to produce the business outcomes that our customers are asking for us. And frankly, at the end of the day, is things that we can build our own business around.

Christophe Bertrand

>> And that's where the Everpure partnership makes so much sense, and with the other ecosystem partners as well.

Omar Lari

>> Yeah, absolutely.

Alison Kosik

>> How do you think enterprises should balance the desire to move quickly on AI with the need to maintain control governance and security around their data?

Omar Lari

>> Yeah. Look, I think we are in an era where everything is moving so fast that speed is a very important factor in deciding this. Now, again, going back to what I said, the intellectual property and the value of a lot of these companies is embedded in their data, and so security and governance is going to be super important and really important. But that's not an excuse to sit around and wait, right? I think we really need to move quickly. Because if the enterprises aren't doing it, their competitors are. And they're going to hire the people, they're going to make the investments that they need to do to maintain competitiveness in the market. It's a tough balance between speed and security, but I think both of those things need to have some significant investment. And leveraging the partner ecosystem to do that can perhaps accelerate some of those ambitions.

Christophe Bertrand

>> And Omar, any closing thoughts for our viewers? What should they do next?

Omar Lari

>> Yeah. As I talked about in the end of the keynote there, I think there's two things that enterprises really need to focus on. One is they've been running mission critical systems for several years. They need to adapt that to AI. This is not an incremental step, it's a significant step function up. The complexity of these AI infrastructure components is exponentially more difficult than what we've seen in the past, so building that expertise now is really important. Second is embrace the data that you have, embrace the intellectual property and the expertise that you have, because the intelligence that you deploy is only going to be as good as the expertise and the data that you feed it. I'm really excited to see what happens in the next couple years. I think AI adoption across the enterprises is going to accelerate massively. And as Crusoe, we're looking forward to helping move that mission forward.

Alison Kosik

>> Wonderful to have you stop by theCUBE today. Thanks so much for your time.

Omar Lari

>> Thank you.

Alison Kosik

>> And you've been watching theCUBE, the leader in live technology coverage and expert technology analysis. We'll be right back.