With the adoption of AI in its early stages, organisations are identifying the right use cases that align with their goals.
The deployment of AI also varies often and demands a strong IT infrastructure.
There are traditional AI approaches, including analytics and statistics that industries have relied on for years. On the other, there are implementations that include deploying advanced models for inference, retrieval-augmented generation (RAG) architectures, or even building custom AI models from scratch.
Each use case has unique infrastructure requirements, for example, running basic analytics differs from supporting complex AI models.
Speaking to iTnews Asia exclusively on getting AI integration right, Pure Storage’s chief technology officer, Robert Lee, said organisations must first define their objectives as to what they aim to achieve with AI.
This clarity will help determine the appropriate infrastructure, whether it's leveraging AI-powered services or designing systems to support complex model development and deployment, said Lee.
AI adoption needs a flexible infrastructure to work
Many businesses are still relying on outdated technologies, including legacy systems with spinning hard drives that are slow and unreliable, hence the need for modernisation.
“Legacy systems often operate in silos, as they were traditionally designed to support specific functions like finance or marketing without sharing data effectively.
It’s crucial to integrate systems and ensure data flows, making data accessible and actionable,” Lee said.
He added that the pace of technological advancements in AI is fast unlike traditional IT environments. AI innovation brings new tools, models, and capabilities almost weekly.
Planning for this change requires investing in flexible infrastructure that can adapt as needs evolve.
The past two years had highlighted the need for organisations to reassess the balance between cloud-based solutions and on-premises infrastructure, Lee explained.
While cloud adoption was once seen as the future, companies are now grappling with rising costs and cloud management challenges.
Lee said cloud solutions offer scalability, agility, and reduced operational overhead but can lead to cost overruns, data security concerns, and reliance on external providers.
On-premises infrastructure, on the other hand, provides control, cost predictability, and customisation, but it requires an upfront investment, has limitations in scalability and maintenance overheads.
“Ultimately, the decision between cloud and on-premises solutions depends on factors including company size, budget, and needs,” said Lee.
“Smaller businesses may benefit from the cloud’s flexibility, while larger enterprises might prefer a hybrid or on-premises approach to balance cost and control. Additionally, as AI systems rely on vast and diverse datasets, ensuring these datasets are accurate, unbiased, and secure is critical to achieving trustworthy AI outcomes.”
- Robert Lee, CTO, Pure Storage
A data governance framework is also necessary for responsible AI
Equally important – an organisation needs a strong data governance framework for responsible AI. This framework must include tracking data provenance, managing its lifecycle, addressing bias, and ensuring transparency.
Lee said provenance tracking ensures AI models use high-quality and reliable data, while lifecycle management organises and catalogues data securely.
“By addressing biases in training data, we can prevent AI systems from perpetuating inequalities, and clear governance policies ensure accountability through regular audits and defined data stewardship roles,’ he said.
As organisations adopt governance principles, the growing volume of unstructured data presents additional challenges.
Concluding, Lee emphasised that to manage unstructured data while ensuring security and compliance, organisations must focus on three key practices.
“First, organisations must have the infrastructure in place to support the performance and network capabilities needed to handle and exchange data efficiently. Second, investing in open data formats is essential. Open formats enable integration with new technologies and applications as they evolve.”
He observed that object storage, often paired with open formats, is becoming a standard for managing unstructured data effectively.
“Thirdly, organisations must implement strong processes and technologies to track the security and provenance of data. This includes knowing who has access to specific data, understanding how it can be used, and maintaining an audit trail of changes made since its origin,” Lee said.
