For Crusoe AI, a sustainable AI infrastructure has always been the founding premise, a welcome alternative amid a surge of antipathy for data centers in local communities.

The company takes a vertically integrated, energy-first approach to building AI infrastructure, sourcing power and deploying managed AI services on top of it, according to Omar Lari (pictured), senior director of product management at Crusoe Energy Systems LLC. This method has enabled deployments in locations far from traditional sources (such as wind and natural gas in Abilene, Texas, or geothermal and hydroelectric in Iceland), as well as a partnership with Redwood Materials that powers thousands of Blackwell graphics processing units with recycled electric vehicle batteries.

“Crusoe’s mission is to accelerate the abundance of energy and intelligence,” Lari said. “Energy is going to drive the next breakthroughs in AI. AI will eventually help us make the next breakthroughs in energy.”

Lari spoke with theCUBE’s Christophe Bertrand and Alison Kosik at Pure Accelerate 2026, during an exclusive broadcast on theCUBE, SiliconANGLE Media’s livestreaming studio. They discussed how Crusoe’s energy-first strategy differentiates its sustainable AI infrastructure offering and what enterprises need to prioritize as they scale AI workloads. (* Disclosure below.)

Sustainable AI infrastructure requires reliability at scale

Crusoe’s partnership with Everpure reflects three priorities that matter most to AI infrastructure customers: reliability, empathy for the service-provider model and supply chain expertise, according to Lari. These qualities are of enormous importance because AI-native customers running massive GPU training clusters have no tolerance for downtime. Roughly 800 Blackwell GPUs represent a quarter petabyte of high-bandwidth memory, and idle time burns capital at scale.

“Performance is a really important piece,” Lari said. “Think about if you have 800 Blackwell GPUs; that represents about a quarter petabyte of high-bandwidth memory. 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.”

Looking ahead, connecting AI-native model builders with enterprise data owners is the most important thing, according to Lari. AI natives have the models, and enterprises have decades of accumulated data and domain expertise. Securely combining those two worlds, along with proper governance, is where the next wave of industry-specific AI value will be created. AI infrastructure demands being treated as a step-function increase in complexity rather than an incremental upgrade, and building that expertise now is critical, Lari noted.

“The intelligence that you deploy is only going to be as good as the expertise and the data that you feed it,” he said. “AI adoption across the enterprise is going to accelerate massively.”

(* Disclosure: TheCUBE is a paid media partner for the Pure Accelerate event. Neither Everpure, the sponsor of theCUBE’s event coverage, nor other sponsors have editorial control over content on theCUBE or SiliconANGLE.)

Photo: SiliconANGLE