TAIPEI, TAIWAN – August 6, 2026 – H3 Platform Inc., a pioneer in software-defined PCIe fabric technology, today officially introduced Falcon 6048, its next-generation AI memory/storage server for large-scale AI infrastructure, specializing in Graph Neural Networks (GNNs), vector database indexing, vector search, and Agentic AI applications. Driven by H3's proprietary fabric orchestration, the 3U platform works with NVIDIA® SCADA™ to use NVMe SSDs as a second-tier memory pool for massive AI workloads.

As AI models and datasets continue to scale, the industry faces an inevitable infrastructure challenge: the memory wall. While modern GPU clusters deliver unprecedented compute performance, AI workloads are increasingly constrained by limited HBM capacity, high memory expansion costs, and inefficient data movement across traditional CPU-centric architectures. Addressing this challenge requires GPUs to access data directly, without routing through the CPU.

NVIDIA® SCADA™ (Scaled Accelerated Data Access) is the GPU-initiated storage technology designed to meet this need. It changes how GPUs access data: instead of routing storage traffic through the CPU, it enables GPUs to communicate directly with storage, reducing latency and overhead. It is designed to handle the large volume of small, random data requests generated by GPU workloads — a pattern common to GNNs, vector database indexing, vector search, and large-scale retrieval, where performance depends on quickly accessing small pieces of data from massive datasets. Falcon 6048 works with NVIDIA® SCADA™, providing the system-level foundation for H3's software-defined PCIe fabric orchestration.

H3's software-defined fabric orchestration enables flexible connectivity between GPUs and NVMe SSDs, supporting scalable many-to-many data paths, deep queue distribution, and efficient resource sharing across the system. Falcon 6048 delivers up to approximately 200TB of NVMe capacity with over 200 million IOPS at fine-grained 512B random access — a profile that aligns closely with the data access patterns required by GNNs, vector search, and large-scale retrieval workloads.

Beyond established AI workloads, Falcon 6048 is also well-suited for the emerging demands of Agentic AI infrastructure. As AI systems evolve from single-model inference toward multi-agent pipelines — where autonomous agents continuously query knowledge bases, retrieve tool call histories, and maintain contextual memory across extended reasoning chains — the underlying memory/storage infrastructure must support rapid, concurrent access to large, distributed datasets. The fine-grained, high-throughput I/O profile of Falcon 6048, combined with its scalable PCIe fabric architecture, is designed to meet these access patterns.

"We are entering an era where legacy memory and storage architectures can no longer keep pace with the data demands of modern GPU clusters," said Brian Pan, CEO of H3 Platform. "While High-Bandwidth Memory (HBM) delivers exceptional performance, its capacity and cost limitations create challenges for enterprises scaling large AI workloads. With Falcon 6048, H3 Platform provides a new system-level approach that uses cost-effective, high-density NVMe storage as an extended second-tier memory for AI infrastructure, helping customers improve scalability, GPU utilization, and total cost of ownership."

For Cloud Service Providers (CSPs), hyperscale operators, and enterprise data centers, Falcon 6048 provides a flexible blueprint for future-ready AI infrastructure. By enabling high-capacity storage resources to support memory-intensive workloads, the platform helps operators reduce dependence on costly system memory expansion while keeping valuable GPU resources better utilized.

For more technical specifications, architecture diagrams, or to schedule a deep-dive technical inquiry, please visit the official product portal at https://www.h3platform.com/product-detail/overview/37 or contact sales@h3platform.com.

About H3 Platform

Founded in 2016 and headquartered in New Taipei, Taiwan, H3 Platform is a pioneer in PCIe and composable infrastructure solutions. The company helps data centers improve efficiency, scalability, and flexibility through advanced PCIe fabric orchestration, memory expansion, and resource management technologies.