AI Infrastructure and Architecture - Infinia was conceived due to the need for a different architecture for AI, one that scales efficiently for training, has low latency, is distributed on-premise and multi-cloud, and minimizes data movement [1] - The industry is shifting towards Data Intelligence, reframing storage of raw data into informational form, which is a new opportunity for DDN to provide data intelligence for enterprises running AI [1] - Metadata and tagging are essential for multimodal AI, enabling the movement of metadata and making the economics viable due to the compression ratio [1] AI Application and Adoption - Enterprises need to adopt AI at an accelerated pace, requiring the application layer to be supercharged and the infrastructure to be efficient [1] - The industry is moving from high-performance computing to Enterprise, and then to digital twins of Enterprise, enabled by technologies like Omniverse [2] - AI is enabling companies to create digital twins, allowing them to run thousands of experiments simultaneously and optimize outcomes, applicable to enterprises, governments, and individuals [2] AI Model and Ecosystem - Post-training, which involves problem-solving and reasoning, is a crucial and compute-intensive part of intelligence, following pre-training [3] - The release of open-source reasoning models like DeepSeek's R1 is accelerating AI adoption by highlighting opportunities for more efficient models [3] - The CUDA ecosystem is enabling the application of AI in specific industries like Life Sciences, Financial Services, and autonomous driving [3] Strategic Partnership and Future Vision - The partnership between Nvidia and DDN is expanding from supercomputing to Enterprise and Omniverse, with Infinia playing a key role [4] - Companies should both use public cloud AI and build their own specialized AI, curating AI agents from various sources to solve large problems [3] - Differentiation for organizations comes from specialized application of AI, enabled by technologies like Nvidia's Nims and DDN's Infinia [4]

Product Overview - Infinia is a software-defined, next-generation data intelligence platform designed to support both structured and unstructured data, integrating diverse data sources into a unified platform [1] - Infinia aims to reduce complexity in AI deployments and can be deployed on various hardware configurations, including cloud, data centers, reference architectures, and OEM solutions [1] - Infinia operates on the premise of SLAs for capacity, performance, and resilience, abstracting away traditional storage management complexities like volumes and IOPS capping [1] Key Features and Capabilities - Infinia focuses on simplifying data ingestion with low-latency access, emphasizing data tagging for categorization and retrieval in complex environments with billions of objects [1] - Infinia addresses limitations in existing storage solutions by offering unlimited metadata scalability, allowing for thousands of tags per object to refine data definition [1] - Infinia includes capabilities for filtering data at the edge, enabling efficient transfer of relevant data to the core data center for analytics [2] - Infinia offers a software development kit (SDK) to enable customers and service providers to accelerate their applications by using a high-performance interface into Infinia [2] Performance and Efficiency - Infinia achieved a 22x speedup in an indexing and ingest workload on AWS by integrating Nvidia GPUs and utilizing Infinia's S3 interface, resulting in a 70% reduction in GPU resource consumption [1][2] - The company demonstrated a factor of 600 speedup in certain workloads related to identifying data for training AI models [1] Architecture and Integration - Infinia implements data services on top of a key-value store, providing different views onto the data through protocols like S3, file interfaces, and SQL interfaces [2] - Infinia includes a massively distributed SQL server, allowing native access to data and metadata stored in the key-value store through SQL queries [2]