NVIDIA Announces DGX Spark 64GB: A 1-PetaFLOP Grace Blackwell Desktop for Local AI Agents, Fine-Tuning, and Inference

NVIDIA has announced a new 64GB configuration of its DGX Spark desktop AI system, manufactured by leading hardware partners including Acer, ASUS, Dell, Gigabyte, HP, and MSI. This development marks a significant step in democratizing access to enterprise-grade artificial intelligence infrastructure, enabling developers and organizations to deploy sophisticated AI models locally while maintaining the flexibility to scale operations across multiple systems.

The DGX Spark 64GB is powered by NVIDIA's Grace Blackwell architecture and delivers 1 petaFLOP of computational performance in a desktop form factor. The system provides developers with a standalone platform capable of supporting local AI agents, model fine-tuning, and inference tasks. A key feature of the architecture is its clustering capability—two 64GB units can be connected to create a unified 128GB memory configuration, enabling organizations to scale their AI infrastructure incrementally without requiring significant capital investment in larger, more expensive systems.

• Enables smaller enterprises and individual developers to access powerful AI capabilities previously requiring data center investment
• Reduces barrier to entry for organizations seeking to experiment with and deploy local language models and AI agents
• Supports the growing trend of on-premises AI deployment, addressing privacy and latency concerns associated with cloud-based solutions
• Provides flexibility for workload scaling through modular clustering, allowing companies to grow computational capacity incrementally
• Supports multiple use cases including model fine-tuning, inference optimization, and local agent development
• Positions NVIDIA to capture emerging demand in the mid-market AI infrastructure segment

The introduction of the DGX Spark 64GB represents a critical convergence of computing power, affordability, and accessibility in the AI market. As enterprises increasingly seek to deploy proprietary models locally while maintaining control over sensitive data, systems like this fill an essential gap between consumer-grade GPUs and prohibitively expensive data center infrastructure. The modularity of the design—allowing users to start small and scale horizontally—makes enterprise AI adoption more financially pragmatic, potentially accelerating widespread deployment of sophisticated AI capabilities across organizations of all sizes.