Alluxio's Chief Executive, Haoyuan Li, has shared his insights on the anticipated technological advancements expected by 2025, covering areas such as artificial intelligence, data and analytics, cloud computing, modern data centres and DevOps.
In the realm of AI and machine learning (AI/ML), Li predicts that multi-modal training will become more mainstream. This method of integrating various types of data - text, images, audio, and video - is expected to be a more dominant approach in model training. The shift is driven by the necessity for AI systems to process the complex nature of real-world data, enabling richer, context-aware applications. "The rise of these models will also spur demand for more advanced hardware and storage solutions, as the complexity of training environments continues to grow," said Li.
Li also forecasts that pre-training will become a key differentiator for organisations developing large language models (LLMs). With the AI landscape evolving, access to vast quantities of high-quality data, especially industry-specific data, will offer a competitive edge. "Preparing and curating the right data for pre-training is increasingly complex, and companies without robust big data infrastructure will struggle to keep up," Li noted, highlighting the importance of data handling expertise.
In the data and analytics sector, Li underlines the critical need to overcome data access challenges for AI success by 2025. "The explosion of data across multiple clouds, regions, and storage systems has created significant bottlenecks in data availability and movement," he pointed out. Organisations will need to efficiently manage data access across distributed environments while reducing data movement and duplication to remain competitive in scaling AI initiatives.
Regarding cloud computing, Li believes that AI-driven cloud economics will reshape infrastructure decisions. "The focus will shift from traditional cloud cost optimisation to AI-specific ROI optimisation," he explained, as organisations develop enhanced modelling capabilities to understand and forecast AI workload costs. This strategic shift is expected to influence how hybrid deployment strategies are formed, balancing cost and performance between cloud providers and on-premises resources.
Turning to modern data centres, Li foresees that by 2025, maximising GPU utilisation will become a standard practice. As AI model training datasets grow exponentially, the optimisation of expensive GPU infrastructure investments will be paramount. "Success will be measured by how effectively data centres can keep their GPU resources busy while managing larger model checkpoints and growing data requirements," he stated.
In the area of DevOps, the evolution from MLOps to comprehensive AIOps platforms is expected. "These platforms will integrate sophisticated monitoring and automation capabilities for both models and infrastructure," Li anticipates, foreseeing a shift towards treating AI models as dynamic systems with built-in continuous learning and adaptation capabilities.
