Riding the AI Wave: Thailand at the Storage Crossroads

Generative AI: It's poised to revolutionize industries across Thailand, from bustling Bangkok to the rice paddies of the Northeast. But as companies tentatively wade into these uncharted waters, a critical question looms: can their storage infrastructure keep up with the tsunami of data AI demands?

The stakes are high. Companies are beginning to see the dangers of outdated storage killing their AI dreams. Meanwhile, others with foresight are building robust data foundations, ready to ride the AI wave. As a result, IDC predicts a surge in investment in dedicated compute and storage and public cloud infrastructure services over the next 18 months. The demand for performance-oriented storage is about to explode, it says.

“GenAI models operate on a vast ocean of unstructured data for training. This includes documents, audio, images, videos, and other objects,” explains Yang How Tung, storage technical specialist at IBM ASEAN.

But simply throwing more bytes at the problem won't cut it. Companies are waking up to the sheer volume and velocity of AI data, which goes far beyond mere data ingestion for model training. Model development demands storage for checkpoints, hyperparameters, configurations, and evaluation metrics. Many models also crave space for inference and storing vector embeddings as companies explore retrieval augmented generation (RAG). As GenAI morphs into multimodal and real-time systems, the strain on storage subsystems will be immense, especially considering companies' ongoing storage needs for daily operations.

Tung believes it’s time for a fundamental rethink of storage strategy, particularly as companies transition from experimental AI sandboxes to production-scale deployments.

Taking a byte out of the storage challenge

The harsh truth is that many existing storage systems weren't designed for the unique demands of AI workloads. Much of the data still sits on subsystems and architectures engineered before GenAI hit the markets.

AI training and inference require lightning-fast data access, massive scalability, and the ability to handle diverse data types simultaneously. Traditional storage architectures often buckle under these pressures, creating bottlenecks that can derail AI initiatives.

This is where modern solutions like IBM FlashSystem offer a way out. Built with our industry-leading IBM FlashSystem NVMe technology and FlashCore Module (FCM) drives, these systems deliver the micro-latency performance essential for AI workloads. But raw performance is just the beginning.

“IBM Storage Scale System provides scalable and flexible storage compute options that can grow with the needs of AI workloads, accommodating large volumes of data without compromising performance,” notes Tung. This hybrid elasticity is crucial for organizations that need to expand their AI capabilities over time, creating a sustainable IT architecture.

Data security and privacy: the unseen threats

Here’s what many companies overlook when trying to address the GenAI storage problem:  It’s not just about space and speed; data security and privacy are equally critical. This requires a more holistic approach that will need data, compliance, and security teams to work together. 

Why? For one, many companies feed sensitive data into AI models for specific company use cases. This inadvertently widens the threat surface that a CISO manages. An unintentional single mistake, an intentional injection of bad data, or data encryption as part of a ransomware attack can result in a rogue or inaccurate model. Ransomware has been ranked as the most common form of cyberattack globally, costing victims an average of USD4 million per breach. 

Thailand's data protection regulations add another layer of complexity to AI implementations, making data storage a compliance matter as well.

IBM’s approach addresses this through comprehensive security features, including encryption, access controls, and advanced authentication already baked into the solutions.

“IBM FlashCore Module (FCM) drive is our patented computational storage flash drives built with Ransomware Threat Detection that brings detection as close to the data as possible, further reducing time to detection and enabling quick decision making,” says Tung. This AI and ML-driven feature uses predictive analytics to identify anomalies in data access patterns, stopping attacks before they wreak havoc. 

So what if the worst happens? “IBM FlashSystem Safeguarded Copy supports immutable snapshots, which automatically create read-only copies of data that cannot be altered or deleted according to the user scheduling,” Tung explains.

This provides a crucial safety net for organizations deploying AI systems, ensuring they can quickly recover from data corruption or security incidents. While the latter can lead to fines and reputational damage, data corruption and model inaccuracy can derail months of planning and investments.

Building for the hybrid cloud pivot

Thai companies aren't just planning for AI — they're also navigating the transition to a hybrid cloud infrastructure as they balance data security and privacy with model training and inference needs.

IBM's storage solutions are designed from the ground up for this hybrid reality, supporting seamless data movement between on-premises systems and cloud environments. This flexibility is essential for companies that leverage cloud-based AI services while maintaining control over sensitive data.

The IBM Storage Fusion HCI System, a self-service hybrid cloud platform for Red Hat OpenShift solution with container-native data orchestration, exemplifies this approach, integrating storage, compute, and data management into a unified system optimized for AI workloads. Its integration with IBM's watsonx platform provides a clear path for organizations looking to deploy AI applications in hybrid cloud environments.

Understanding the cost of transformation

Investing in new storage building blocks does cost. But while it might seem significant, Tung emphasizes the long-term benefits: “The initial investment to any new implementation or replacement might seem significant, [but] integrating IBM storage solutions with existing or legacy IT infrastructure deliver long-term benefits in terms of operational cost savings.”

These savings come from various features, including zero-performance impact data compression, enterprise-grade reliability with "six nines" (99.9999%) availability, and efficiency gains from advanced NVMe technology. Companies need to consider these factors when evaluating the total cost of ownership for their AI infrastructure.

Join the discussion

So, where do you begin your GenAI storage journey? A good starting point will be an upcoming executive luncheon in Bangkok, “Build a Robust and Resilient IT Infrastructure Optimized for AI Workloads and Hybrid Cloud.”

The luncheon will dive deeper into the above topics and offer practical guidance for business leaders. They will gain practical insights into building robust AI infrastructure, with a focus on real-world implementation challenges and solutions. Perhaps most valuably, they will have the opportunity to engage with experts who understand both the technical demands of AI workloads and learn from other Thai leaders on their storage journeys.

As Thailand's AI landscape continues to evolve, the companies that build the right foundation today will be best positioned to capitalize on tomorrow's opportunities. The storage infrastructure decisions made now will have long-lasting implications for AI success — making this a crucial moment for data leaders to engage with these challenges head-on.

Image credit: iStockphoto/tampatra