According to Nutanix’s Enterprise Cloud Index study, more than half of global organisations say their IT systems require substantial upgrades before generative AI can be deployed at scale, while a similar proportion cite gaps in internal skills.
In a conversation with Frontier Enterprise, Nutanix executives Manosiz Bhattacharyya and Daryush Ashjari discussed where AI initiatives break down in practice, and what prevents organisations from scaling beyond early experimentation.
Infrastructure setbacks
According to Bhattacharyya, Chief Technology Officer at Nutanix, legacy three-tier architectures are a common constraint because they lack the agility, scalability, and data mobility that AI workloads require.
“Without a flexible foundation, it becomes challenging to move data and models securely across data centres, the edge, and the cloud,” he said.
A lack of readiness for containerisation further complicates the situation. Even as nearly 90% of global enterprises begin containerising applications, many still lack the tools and operational expertise needed to manage Kubernetes environments effectively. This limits the portability and elasticity that AI workloads depend on, Bhattacharyya added.
Security and compliance remain another pressure point. Around 95% of global organisations believe they could be doing more to secure generative AI models and applications.
“At the same time, another critical gap lies in AI lifecycle management. AI models are dynamic, requiring constant monitoring, retraining, and governance to prevent drift and maintain accuracy,” he said.
APAC orgs encounter more complexity
In the Asia-Pacific region, AI scaling introduces additional layers of complexity. While enterprises globally face challenges around data integration, infrastructure, and governance at scale, APAC organisations must also contend with fragmented regulations and highly diverse markets, spanning languages, cultures, levels of digital maturity, and data sovereignty requirements.
“Data quality and availability vary widely across the region, with some markets still in early stages of digitisation, making it harder to train and deploy robust AI models. Localisation adds another layer of complexity, requiring models to understand multiple scripts, dialects, and cultural nuances to earn user trust,” said Ashjari, APJ Chief Technology Officer and Vice President, Solution Engineering.
In India, for example, there are 22 officially recognised languages, but the total number of languages spoken is estimated to exceed 120.
“Imagine the challenges faced in training language models for AI projects,” he said.
Infrastructure availability adds further constraints. Cloud access and compute capacity differ sharply between advanced and emerging markets, limiting consistency and scalability across the region.
Hybrid multi-cloud
For Bhattacharyya, operational consistency is central to running AI workloads across data centres, cloud environments, and the edge.
“The most effective approach is a consistent hybrid multi-cloud operating model that allows organisations to run workloads wherever it makes the most sense, without refactoring or adding complexity. A unified platform that delivers identical operations and data services across environments simplifies lifecycle management, improves visibility, and contains costs,” he said.
Further, modern AI applications rely on containers for agility and rapid iteration. Infrastructure that integrates Kubernetes with built-in data services and security makes it easier to move models and data across environments, he explained.
To address performance and governance challenges, AI infrastructure must keep data close to compute while maintaining consistent policies for encryption, access control, and compliance across locations.
“A cloud-smart architecture, one that keeps data close to compute, enforces consistent governance, and scales securely, is essential for running AI anywhere with confidence,” Bhattacharyya said.
From an APJ perspective, Ashjari pointed to the importance of open, extendable AI platforms that can be adapted to local conditions while remaining cost-conscious.
“This must include investing in global, regional, and local partnerships, talent development, and regulatory agility. It is critical to look at scaling not by replicating western models, but by embracing the region’s diversity and unique constraints,” he said.
Additional safeguards
Bhattacharyya also urged CIOs to rethink governance and observability for generative and agentic AI, positioning them as enablers rather than barriers.
“The key is to embed governance, observability, and security within the infrastructure and operational fabric, rather than bolt them on later,” he said.
Avoiding bottlenecks requires unified data and model management across environments, with consistent access, lineage tracking, and compliance from core systems to cloud and edge deployments.
“Observability should go beyond system metrics to include model performance, data drift, and ethical compliance, then consolidated into a single, cloud-smart operating model that delivers transparency without friction,” Bhattacharyya added.
To prevent hybrid multi-cloud flexibility from becoming another source of operational complexity, Ashjari said organisations need a common operating model that unifies operations, governance, and data services across data centre, edge, and cloud environments, for both virtualised and containerised workloads.
“Over the next decade, enterprises will continue to operate across these two continuums, data centre to cloud and VM to container. Success will depend on managing them under a single operational fabric,” he remarked.
Unified data strategy
For CIOs, the path beyond isolated pilots starts with a unified data strategy. That means ensuring visibility, accessibility, and security across data silos so AI models can learn from comprehensive and reliable datasets, Ashjari said.
Establishing robust machine learning operations and lifecycle management frameworks is equally important to support continuous monitoring, retraining, and compliance checks.
“This helps prevent model drift and ensures accountability. At the same time, security and governance should be embedded from the outset so AI can scale safely within regulatory frameworks and organisational risk thresholds,” he said.
Bhattacharyya echoed the sentiment, stressing that effective governance depends on a consistent foundation that integrates control, visibility, and security across environments.
“The goal is to operationalise trust at scale, applying the same discipline to AI as to any critical workload, while retaining the ability to innovate responsibly across hybrid and multi-cloud environments,” he said.
