The Reason Behind Stalled AI Projects

As enterprises race to adopt AI, weak data foundations are preventing more than half (58%) of organizations in the United States and Canada from realizing value and contributing to an estimated $108 billion in wasted global AI investment each year, according to a report from Hitachi Vantara. The reason is rarely bad models or lack of ambition. 

AI models depend on connected and continuously available data. But many organizations still operate across siloed systems, inconsistent metadata standards, and legacy infrastructure environments. While organizations continue investing heavily in generative AI, machine learning, and advanced analytics, many still struggle to turn AI pilots into scalable business outcomes because the data foundation beneath those initiatives is not ready to support them at enterprise scale. 

For storage administrators and cloud architects, this feels familiar: teams are being asked to operationalize AI on fragmented environments that aren’t designed for AI-scale data movement, governance, or workload placement.

Research continues to reinforce this challenge. According to McKinsey, only a small percentage of organizations have successfully scaled AI across the enterprise, with data quality, governance, and infrastructure remaining major barriers to adoption. Research from Bain & Company reinforce the same point: AI outcomes depend on trusted, accessible data and a strong data strategy. Organizations that do not follow this guidance often face stalled deployments, unreliable AI outputs, rising operational complexity, and wasted investment. 

Where critical applications depending on block storage for performance-sensitive transactions, and generated data that feeds AI pipelines (which rely on file and object storage) intersect, application modernization and AI readiness can make a difference. 

The opportunity:  create a data foundation where block, file, and object work together more effectively.

Six Data Modernization Challenges Holding AI Back

1. Data silos across the enterprise limit access

AI depends on broad access to trusted data, yet most organizations still operate fragmented environments across on-premises, cloud, and edge. The modernization challenge is connecting siloed block, file, and object environments without forcing disruptive migrations or creating another layer of silos.

Connected data enables AI applications to use information from across legacy platforms, cloud environments, applications, and business units to:

• Break down data silos. AI-ready organizations connect data across ERP, CRM, operational systems, analytics platforms, cloud repositories, and edge environments. 
• Modernize data pipelines. ETL, ELT, streaming, and real-time data pipelines help move, transform, and prepare data for AI and machine learning workflows. 
• Use APIs across hybrid environments. API-driven integration helps connect data across on-premises systems, public clouds, SaaS platforms, and distributed applications. 
• Create a unified data layer. A data fabric, data platform, or unified data layer helps teams access governed data consistently without manually stitching together disconnected sources. 

2. Performance requirements have hanged

Scalable infrastructure gives AI workloads the storage, compute, performance, and flexibility needed to process enterprise data at scale. Legacy infrastructure can limit AI readiness when it cannot support large data volumes, high-throughput pipelines, hybrid cloud operations, or GPU-accelerated workloads. 

AI workloads require higher throughput, lower latency and the ability to process growing volumes of unstructured data.

Unlike traditional enterprise applications, AI pipelines often combine training data, inferencing, analytics, and retrieval across multiple environments. A workload optimized for block-based transactional databases may be essential to the application, but AI pipelines often depend on file and object storage for unstructured data, training sets, model checkpoints, retrieval workflows and analytics at scale.

The challenge is not simply adding more storage. It is determining where each workload and data type should run based on performance, economics and operational requirements. Organizations should ensure that data foundations:

3. Governance and sovereignty slow deployment    

Data governance is the set of policies, ownership standards, access controls, and compliance practices that determine how enterprise data is managed and used. As organizations scale AI, governance becomes harder and more important. 

Questions around lineage, sovereignty, security, access controls and compliance become increasingly difficult in fragmented environments. For many organizations, the question is no longer “Can we access the data?” but “Should this data move at all?” Strong governance makes enterprise data usable for AI by allowing organizations to scale secure access to trusted data across systems, teams, and hybrid environments. Additionally, modernization helps organizations improve consistency around data provenance, residency, access controls and regulatory obligations so teams can confidently move faster.

To avoid deployment blockades, organizations should:

• Classify and catalog enterprise data. Data catalogs, metadata management, and classification policies help users understand what data exists, where it lives, and how it can be used. 
• Enforce policies with auditability. AI-ready governance requires policy enforcement, lineage tracking, and audit trails so organizations can prove how data was accessed and applied. 

4. Operational complexity creates friction

Too many management tools, disconnected workflows, and manual processes slow teams down. Infrastructure teams supporting AI need time to enable innovation, not simply maintain complexity. As AI initiatives expand across hybrid environments, operational complexity can grow just as quickly as the data itself. IT teams then spend valuable time coordinating across siloed tools, troubleshooting issues and managing routine tasks instead of driving innovation projects and growing revenue. Simplifying operations and improving visibility across environments can help reduce friction and free resources for higher-value work.

5. Cost models need more flexibility

Modernization for AI does not necessarily mean moving more data to the cloud. In many environments, success comes from selectively choosing where data lives and where workloads execute based on cost, governance, and performance needs.

High-performance AI training may justify premium infrastructure, while secondary datasets, backup repositories, or lower-priority analytics may be better suited elsewhere. The goal is flexibility — not a one-size-fits-all model.

6. A workload placement challenge

Many modernization discussions focus on moving data. But AI-ready modernization is not about moving everything. It is about placing workloads intentionally.

Storage and cloud teams increasingly face difficult tradeoffs:

• Which workloads belong on-premises for performance or sovereignty reasons?
• Which data can move to cloud environments?
• Where should inferencing happen?
• How should teams balance cost, governance and resilience?

Modernization for AI requires balancing performance, governance, sovereignty, economics and operational simplicity, often across environments never designed to work together.

AI Modernization Risk Matrix for Infrastructure Teams

If modernization sounds risky, that’s because it can be.    

Storage and cloud teams are responsible for mission-critical systems where downtime, governance gaps or performance issues carry real consequences. But avoiding modernization creates risk, too, especially as AI workloads place new demands on infrastructure.

The most successful organizations approach modernization pragmatically by identifying risks early and reducing friction before migration begins.

The goal of modernization is not perfection on day one: it’s reducing friction, and creating a data environment capable of supporting future workloads, including AI.

Building an AI-Ready Roadmap and Data Foundation

An AI-ready data foundation isn't built by checking off a sequence of steps — it's built by strengthening a set of capabilities that reinforce each other. Instead of trying to solve every challenge at once, organizations can focus on the foundational capabilities that enable AI to scale reliably across the enterprise:

• Foundational stability: A stable data foundation depends on data quality, governance, and operational visibility across systems. Strengthening this capability means reducing inconsistency, establishing trusted ownership standards, standardizing metadata, and building the governance framework that scalable enterprise AI requires. Organizations that invest in these foundational data practices are better positioned to avoid unreliable AI outputs, compliance risks, and fragmented adoption as AI initiatives grow.
• Infrastructure flexibility: AI workloads need an infrastructure layer that can flex with them. This capability centers on integration across hybrid cloud environments, modernizing legacy storage systems, enabling scalable data access, and adopting more flexible infrastructure consumption models. Modern AI workloads often require high-performance storage, accelerated compute, and hybrid cloud architectures that can support distributed data processing at scale. Technology partnerships, including collaborations between Hitachi Vantara and NVIDIA, help organizations build infrastructure environments designed to support GPU-accelerated AI workloads, analytics, and enterprise-scale data operations.
• Enterprise-scale operations: Operationalizing AI across business functions means continuously improving performance, governance, resilience, and efficiency — not just supporting isolated pilots. This capability shows up as ongoing monitoring, optimization, automation, and lifecycle management for production AI workloads. Sustaining this at scale requires continuous investment in data quality, governance, infrastructure modernization, and operational consistency as business requirements and data environments evolve.

Innovation's Most Powerful Data Engine for AI

AI success requires more than better models. It depends on resilient, governed infrastructure that can support evolving workloads, performance demands, and compliance requirements. Consequently, AI modernization will be defined less by how much data organizations move and more by how intentionally workloads are placed across environments.

In other words, neither are possible without a strong data foundation.

With the Hitachi Vantara platform — including VSP One capabilities across block, file, object and software-defined storage — organizations can modernize their data foundation. All while improving flexibility, resilience, governance, and operational simplicity.

Learn more about the Hitachi Vantara platform — powering your AI future on a foundation of trusted, modernized data.