Experience up to 10 petaFLOP  of AI performance at FP4 precision with the NVIDIA Grace Blackwell architecture. 

Run AI development and testing workloads with AI models up to 200 billion parameters at your desktop with a large, unified system memory. 

High-performance NVIDIA ConnectX™ networking enables the connection of up to four NVIDIA DGX Spark systems to work with AI models of up to 700 billion parameters. 

Utilize a full-stack solution for generative AI workloads, encompassing NVIDIA tools, frameworks, libraries, and pre-trained models including NVIDIA NIM™. 

DGX Spark provides students, AI researchers, data scientists, and staff access to the latest powerful NVIDIA Grace™ Blackwell platform in the world's smallest AI supercomputer that is capable of running 200B-parameter models. 

Your students (and others) will have access to software for experimentation and prototyping, including software development kits, orchestration tools, frameworks, and models, available in the NVIDIA NGC catalog and on the NVIDIA Developer portal. Developers can now fine-tune models with the NVIDIA NeMo™ framework, accelerate data science with NVIDIA RAPIDS libraries, and run common frameworks such as PyTorch, Python, and Jupyter notebooks 

DGX Spark gives utility organizations a practical starting point for applying AI to the grid, operations and engineering workflows without requiring large-scale infrastructure commitments upfront. By putting local AI compute directly in the hands of developers, data scientists, power systems engineers and operations teams, utilities can prototype and validate AI use cases close to the data, systems and teams that understand them best — from outage response and load forecasting to maintenance planning, engineering documentation and field operations support. 

For utilities, balancing modernization goals with regulatory, budget and reliability constraints, DGX Spark creates a secure and predictable path to AI adoption. Teams can experiment with sensitive operational data inside their own environment, reduce dependency on variable cloud usage, and build applications that can scale from a local desktop system to larger NVIDIA-powered infrastructure as needs mature. The result is a lower-friction way to move from AI readiness to real operational impact — helping utilities improve resilience, accelerate engineering productivity and support more intelligent grid operations. 

 DGX Spark gives life sciences organizations a secure, cost-effective way to bring AI directly to researchers, bioinformaticians, computational chemists and data scientists, without waiting on shared HPC queues or moving sensitive data to the cloud. With 128 GB of unified memory and up to 1 petaFLOP of AI performance, teams can prototype, fine-tune and run models of up to 200 billion parameters at the desktop, then scale the same workflows to larger NVIDIA-powered infrastructure as programs mature. 

Using the NVIDIA AI software stack, including NVIDIA BioNeMo™, NIM microservices, RAPIDS™ and PyTorch, teams can explore AI-driven drug discovery and development use cases close to the data and the scientists who know it best. Examples include protein structure and function prediction, generative molecular design, single-cell and genomics analysis, medical imaging, and agentic literature mining and regulatory document drafting. 

For organizations balancing innovation with GxP, privacy and IP-protection requirements, DGX Spark enables experimentation on proprietary and patient-derived data within a controlled local environment, reducing cloud dependency and variable token costs. The result is a lower-friction path from AI curiosity to validated, production-ready impact, helping life sciences teams shorten discovery cycles, improve research productivity and accelerate the delivery of new therapies. 

Physical AI is reshaping how products are designed, built and moved, and  DGX Spark puts a development platform for it on the engineer's desk. Powered by the NVIDIA GB10 Grace Blackwell Superchip, DGX Spark lets manufacturing teams develop and test the AI behind vision AI, robotics, humanoids and digital twins locally, then scale to larger NVIDIA-powered infrastructure when they're ready for production. 

Manufacturers generate enormous volumes of data every day, including machine telemetry, sensor streams, camera feeds, quality records and CAD and simulation files. This data is often scattered across plants, systems and OT/IT boundaries, and much of it is too large, too sensitive or too time-critical to move to the cloud. DGX Spark brings AI to where the data already lives. Teams can develop and validate models against plant data on site, avoiding costly data movement, egress fees and added latency, and without exposing proprietary designs or process IP. 

Manufacturers can use it to: 

• Vision AI: Build and fine-tune computer vision models for inspection, defect detection, safety monitoring and process analytics with NVIDIA Metropolis, using high-volume production imagery that stays on site.
• Robotics and humanoids: Develop and test robotics applications, including AI-driven robots and humanoid workflows, with NVIDIA Isaac, bringing development closer to the engineers building them.
• Digital twins and simulation: Prototype the AI models and workflows behind digital twins and simulation locally, then scale to larger NVIDIA-powered infrastructure for full-fidelity simulation of lines, cells and facilities.
• Agentic assistants for the enterprise: Run always-on local agents that surface work instructions, maintenance history and engineering documentation for operators, technicians and planners, without cloud dependency.

By working across siloed plant data in place, teams can move faster, keep costs predictable and reduce reliance on variable cloud usage. The same workflows then carry over to larger NVIDIA-powered infrastructure, helping manufacturers move from pilot to production as their Physical AI programs mature.