As regulatory frameworks tighten and geopolitical tensions intensify, organizations worldwide are realizing that sovereign AI is one of the most consequential themes shaping how enterprises build and govern infrastructure.

By 2030, more than 75% of all enterprises outside of the U.S. are predicted to have a digital sovereignty strategy, underscoring how urgently enterprises need to rethink control across the stack. The challenge extends well beyond data residency to encompass economic competitiveness and cybersecurity risk management, according to Vincent Caldeira (pictured, right), chief technology officer for Asia-Pacific at Red Hat Inc.

“The way we actually define sovereignty is the ability to exert control over your digital destiny,” Caldeira told theCUBE. “It’s taking us beyond just the regulatory compliance and security discussions to other dimensions, such as economic competitiveness. Countries that don’t have enough access to GPUs — they can’t have AI factories, meaning they cannot build their own models, their own AI capability.”

Caldeira and Roberto Carratalá (left), principal AI platform architect at Red Hat, spoke with theCUBE’s Rebecca Knight and Rob Strechay at KubeCon + CloudNativeCon EU, during an exclusive broadcast from theCUBE, SiliconANGLE Media’s livestreaming studio. They discussed how sovereign AI strategies are reshaping enterprise architecture from model provenance to deployment flexibility. (* Disclosure below.)

Sovereign AI demands transparency, flexibility and cost control

While sovereign AI conversations often begin with regulation, enterprises are quickly discovering the challenge is operational. Different countries face vastly different constraints — from nations that can manufacture their own accelerators to those that must partner across the supply chain for hardware and models, Caldeira explained. Defense, in particular, has emerged as one of the most affected verticals, he noted.

“Even for the defense industry and the usual amount of means they have, [it] is still not practical for them to train a generative AI model from scratch,” Caldeira said. “At the same time, we are basically running critical decisions on this technology. We need a huge amount of risk management and guardrails around how they actually use it.”

To address these challenges, Red Hat is anchoring its sovereign AI approach around three architectural principles: interoperability across deployment environments, trust through model provenance and transparency, and cost control across accelerator infrastructure. The company’s validated model catalog and contributions to open-source inference projects, such as llm-d — recently presented to the Cloud Native Computing Foundation — reflect this strategy, Carratalá noted. This comes as enterprises face the rising cost of production AI.

“We already saw that a lot of our customers transitioned from the exploratory phase to now putting things in production [and] that consumes a lot of different tokens,” Carratalá said. “They are skyrocketing on this consumption. You need to be able to have these cost-effective models.”

Stay tuned for the complete video interview, part of SiliconANGLE’s and theCUBE’s coverage of KubeCon + CloudNativeCon EU.

(* Disclosure: Red Hat sponsored this segment of theCUBE. Neither Red Hat nor other sponsors have editorial control over content on theCUBE or SiliconANGLE.)

Photo: SiliconANGLE