AI day spotlights sovereignty, storage & data woes

Technology leaders are using AI Appreciation Day to draw attention to the less visible infrastructure and data issues behind the rapid adoption of artificial intelligence across enterprises and the public sector.

Executives from Zadara, Spectra Logic and CTERA say the conversation now extends well beyond models and algorithms to questions of sovereignty, storage economics and data quality.

Yoram Novick, Chief Executive Officer of Zadara, said recent geopolitical tensions and regulatory pressures have led many organisations to reassess their dependence on foreign-controlled platforms when deploying AI at scale. He pointed to a shift from experimental projects to production workloads involving sensitive and strategically important information.

"In recognition of AI Appreciation Day, it is important to look not only at what AI can do, but at the infrastructure choices that will determine who can use it securely, sustainably and on their own terms. AI is quickly becoming a critical capability for enterprises, governments, service providers and entire economies. But as adoption moves from experimentation to production, organizations are realizing that access to AI is not just a technology question. It is also a question of sovereignty, resilience and operational control. Recent geopolitical developments have made clear that reliance on foreign-controlled AI platforms can create real business continuity risks, especially for organizations managing sensitive, regulated or strategically important data. The goal is not isolation or moving away from global innovation. The goal is strategic optionality. Organisations need the ability to run AI workloads locally, keep data within required jurisdictions, control who governs access to critical services, and scale infrastructure in a way that is economically sustainable. This is especially important as AI workloads become more distributed, latency-sensitive and unpredictable. Centralized public cloud environments will remain important, but they cannot support every use case. The next phase of AI will require flexible, localized and multi-tenant infrastructure models that combine cloud-like simplicity and consumption economics with data locality, strong tenant isolation and greater control. What makes AI so powerful is its ability to help organizations act faster and make better decisions. To fully realize that potential, AI must be built on infrastructure that gives organizations choice, protects sovereignty and allows innovation to scale without creating new dependencies."

His comments underline an emerging divide between centralised public cloud services and more distributed approaches that keep compute and data closer to users and within regulated jurisdictions. The emphasis on "strategic optionality" aligns with a broader industry focus on multi-cloud and hybrid deployments for AI training and inference.

Storage architecture is also under scrutiny as AI workloads generate unprecedented data volumes. David Feller, Chief Technology Officer of Spectra Logic, said many organisations remain unprepared for the long-term retention demands created by training data, model outputs and machine-generated content.

"AI Appreciation Day is a good moment to recognize not only what AI can do, but what it demands from the infrastructure behind it. AI is accelerating data growth at a pace many organizations are not prepared for, creating massive volumes of training data, model outputs, video, analytics and machine-generated information that may need to be retained for years or even decades. The challenge is that not all of this data belongs on always-on, high-cost flash or disk infrastructure, and power and supply constraints can make that approach unviable. As AI pipelines mature, organizations need to think more strategically about where data lives, how often it needs to be accessed, and how it can be preserved securely, sustainably and cost-effectively over time without sacrificing the ability to restore and retrain models. This is where tape and modern archival architectures become central to AI success. When integrated into object-based and hybrid storage environments, tape can provide a durable, energy-efficient and cost-predictable foundation for long-term retention, while also supporting offline protection against ransomware and other cyber threats, with restore rates that can easily keep up with demanding retrain workflows. The organizations that will get the most value from AI are those that heavily invest in GPU farms and keep them busy 24/7 by properly balancing surrounding storage across appropriate tiers that include tape and take advantage of new technologies to optimize their AI investment. It is likely that one of the oldest storage technologies will have the biggest impact in advancing the AI revolution."

Feller's emphasis on tape-based archives reflects concern about the power, cost and supply constraints affecting always-on flash and disk systems. The argument ties AI investment returns to the ability to tier data intelligently and keep expensive compute resources fully utilised.

Alongside sovereignty and storage, the state of enterprise file data is emerging as a decisive factor in AI outcomes. Aron Brand, Chief Technology Officer of CTERA, said many projects fail not because of model performance, but because of fragmented and poorly governed data estates.

"AI Appreciation Day feels like something AI probably created so we'd all remember to say thank you. Fair enough. The machines have been very busy this year. Maybe we do need a day to make sure AI feels seen, valued and emotionally supported before it optimizes us out of the workplace. But I want to move past the hype and talk about what actually makes enterprise AI work. Or fail. Every struggling AI initiative I see shares the same root cause: the underlying data layer. Enterprise file data is messy. It lives across edge locations, branch offices, remote sites and cloud environments simultaneously. It is unstructured, rarely tagged and historically siloed. Building AI-ready infrastructure means addressing issues that most vendors do not discuss publicly: How do you index petabytes of file data without disrupting production workloads? How do you enforce access policies for AI retrievals? How do you ensure AI tools see one coherent data environment, regardless of where files physically live? These are hard engineering problems. Solving them is what separates organizations that run successful AI projects from those that run expensive pilots that never scale. So yes, happy AI Appreciation Day. But maybe the best way to appreciate AI is to give it the one thing it really needs: clean, governed, secure access to the right data."

Brand's comments highlight the operational challenges beneath AI pilots and proofs of concept. Indexing, access control and data coherence across edge and cloud environments are now central engineering issues for enterprises seeking to move AI systems into production.