Enterprises in APAC embracing cloud-first strategies for AI and high-performance computing (HPC) risk substantial cost overruns if workloads are not matched to the right infrastructure, according to Joseph Yang, General Manager of HPC and AI, APAC & India at HPE. Yang cautions that cloud-based HPC can be up to 10 times more expensive than on-premises deployments, urging CIOs to adopt a data-driven total cost of ownership (TCO) approach before scaling.
Key considerations for AI and HPC deployment include:
- Business alignment: Evaluate whether HPC or AI directly addresses specific operational challenges. AI, built on HPC architecture, now extends beyond R&D into productivity, automation, and customer-facing applications.
- Cost structure: In AI and HPC environments, infrastructure can exceed 50% of costs versus 15–20% in traditional workloads, requiring different budgeting and ROI planning.
- System architecture: Purpose-built designs with high compute density, optimised cabling, and advanced cooling can enhance performance and energy efficiency.
HPC’s value is evident in industries such as manufacturing and simulation. For example, Toyo Tire’s adoption of an HPE Cray XD system via HPE GreenLake cut simulation times by half, accelerating design cycles and improving deep learning tools for tire development.
Yang identified three frequent missteps: starting with underpowered systems that degrade AI Agent performance, over-prioritising customisation before establishing baseline use cases, and failing to engage HPC/AI specialists early.
Generative AI workloads, especially large language models, require responsive infrastructure to meet user expectations. Initial deployments can focus on generic productivity gains before custom data integration.
With skilled talent in short supply and on-prem capabilities eroded by cloud-first moves, Yang emphasises early strategic planning and expert collaboration as critical for cost efficiency, scalability, and sustained performance in AI and HPC initiatives.
Source:
