Michael Tso, (Cloudian) – Revolutionizing AI Storage: Cloudian’s Partnership with NVIDIA GPUDirect
Michael Tso, Co-Founder of Cloudian, shares how the company is advancing AI-ready data platforms through its integration with NVIDIA GPUDirect. By combining unparalleled performance and scalability, Cloudian is redefining storage technology to meet the demands of AI-driven enterprises. Meet the Cloudian team and explore these groundbreaking innovations at the Confare CIOSUMMIT, Austria’s premier IT management event.
What led to the decision to develop an integration with NVIDIA GPUDirect, and what strategic role does this partnership play for Cloudian?
Our decision to integrate with GPUDirect addresses the growing complexity of AI storage infrastructure. Until now, AI storage solutions often required separate storage layers: file storage for performance and object storage for scale and cost. This multi-layer architecture added cost, management overhead, and ongoing data migration challenges. Cloudian eliminates this complexity by consolidating data to a unified data lake.
NVIDIA GPUDirect integration now lets users access this data lake at speeds of 200GB/s from a single system, three times faster than without GPUDirect, delivering both performance and scale in a unified platform that can serve all stages of the AI workflow.
By integrating GPUDirect, Cloudian evolves from a traditional object storage provider to an AI-ready data platform capable of handling the next generation of training data – from multi-modal content to massive synthetic datasets. This aligns perfectly with the changing needs of AI.
AI enables faster and more accurate decisions. Which specific industries or use cases benefit the most from this integration?
By simplifying and cost-reducing the AI data storage infrastructure, this integration delivers significant benefits to industries facing exponential growth in the volume of their training data sets.
- Generative AI companies can accelerate model training with faster access to massive datasets.
- Healthcare firms can effectively manage volumes of synthetic data employed to enhance data privacy, and augment datasets for predictive analytics.
- Genomics researchers can analyze vast amounts of genetic data more efficiently, accelerating precision medicine discoveries.
- Autonomous vehicle developers can process both real-world sensor data and synthetic training data at scale.
- Media companies benefit from faster video processing for applications like anomaly detection and object recognition.
In each case, the combination of GPU-class performance with scalable object storage transforms how organizations manage and process their data.
Interested in all things AI? Then we have the ideal spot for you! At Confare #CIOSUMMIT you will see the cream of the crop when it comes to IT-decision-makers. Sign up for March the 26th & 27th 2025 here.
How does Cloudian ensure that the HyperStore platform remains scalable and performs well as the volume of AI data grows exponentially?
Cloudian’s foundational architecture ensures true scalability in both capacity and performance. While traditional file systems hit scaling limits due to their hierarchical structure, HyperStore’s flat address space enables virtually unlimited growth. The modular design allows organizations to start small and scale seamlessly to exabyte capacity, with performance scaling linearly as nodes are added. Now, with GPUDirect integration, a single system can easily deliver 200GB/second throughput, scaling to TB/s performance levels as capacity expands.
The competition in the field of S3-compatible storage systems is fierce. What fundamentally sets HyperStore apart from other solutions on the market?
HyperStore stands out by eliminating the traditional compromise between performance and scale. As the first object storage platform to integrate NVIDIA GPUDirect, it enables direct GPU-to-object transfer without intermediate storage layers, dramatically simplifying AI infrastructure. This consolidation reduces costs through fewer storage tiers, 42% lower CPU usage, and improved power efficiency.
Beyond GPU integration, HyperStore brings unique enterprise capabilities. With its native S3 API compatibility – ensuring seamless integration with AI tools — it’s the only on-premises object storage platform offered by AWS. Military-grade security includes Secure Shell for intrusion prevention, FIPS validation and SEC compliance, and S3 Object Lock for ransomware protection.
True multi-tenancy enables secure infrastructure sharing, and the geo-distributed architecture allows deployment anywhere with single-screen management through HyperIQ observability. Built on industry-standard hardware, HyperStore delivers this enterprise functionality while maintaining cost-effectiveness at scale.
Be a part of the Confare Female IT Community! Got a female colleague or do you yourself want to be under the guidance of a well-seasoned IT-powerwoman? The Confare Female IT-Mentoring makes it easy for high-potentials to gather valuable experience. Sign up here.
How do you see the future of deep learning in connection with object storage? What challenges and opportunities do you foresee for Cloudian in this context?
The relationship between deep learning and object storage grows closer as AI models increase in size and complexity. We’re seeing the emergence of multi-modal AI requiring diverse data types, greater adoption of synthetic data for training, and an increased focus on data consistency across training pipelines. Cloudian is well-positioned to address these trends through its unified data lake approach and continued innovation in high-performance data access. As the industry evolves, the ability to efficiently manage and access vast amounts of training data will become even more critical.
The volume of unstructured data is growing rapidly. How is Cloudian addressing challenges related to data management, data protection, and sustainability?
Cloudian takes a comprehensive approach to managing unstructured data growth, addressing management, protection, and sustainability. From a management perspective, the platform provides a single namespace across locations, from edge to core to cloud, and rich metadata capabilities, complemented by direct GPU access for performance. Integrated cloud tiering enables seamless data management between cloud and on-prem.
Data protection is ensured through immutable storage, ransomware protection, fine-grained access controls, and comprehensive encryption.
On the sustainability front, the solution reduces environmental impact through storage consolidation, lower power consumption via efficient data access, and optimized CPU utilization through GPUDirect. This holistic approach enables organizations to confidently scale their unstructured data while maintaining control over costs and complexity.
