NVIDIA DGX Station Brings 20-Petaflop AI to 7 PC Makers
Key insights
- DGX Station packs 20 petaflops and 784GB unified system memory via the GB300 Grace Blackwell Ultra Desktop Superchip.
- Seven major PC makers including Acer, ASUS, Dell Technologies, GIGABYTE, HP, Lenovo, and MSI will produce both DGX systems.
- DGX Spark, the compact 1-petaflop variant with 128GB unified memory, is taking reservations now for July 2025 availability.
Why this matters
NVIDIA's COMPUTEX announcement marks the first time 20-petaflop AI compute has been packaged as a personal desktop product across seven major PC manufacturers, compressing the gap between data-center infrastructure and individual workstations. For AI practitioners and founders with locally sensitive workloads, DGX Station's 784GB unified memory and Multi-Instance GPU support make on-premise inference and fine-tuning viable without data leaving the building. The simultaneous entry of seven global PC makers into the DGX supply chain signals that deskside AI supercomputing is transitioning from niche HPC instrument to a standard enterprise procurement category.
Summary
NVIDIA unveiled DGX Spark and DGX Station at COMPUTEX in Taiwan, moving data-center-class AI compute onto personal desktops.
DGX Station runs the GB300 Grace Blackwell Ultra Desktop Superchip with 20 petaflops of AI performance and 784GB unified system memory. Its ConnectX-8 SuperNIC reaches 800Gb/s networking, and Multi-Instance GPU partitions the system into up to seven independent instances. DGX Spark, the compact sibling built on the GB10 Grace Blackwell Superchip, delivers 1 petaflop with 128GB unified memory; reservations are open now with availability from July 2025.
Essentially: (NVIDIA, Acer, ASUS, Dell Technologies, GIGABYTE, HP, Lenovo, MSI) are building a new personal AI platform aimed at developers, researchers, enterprises, and government agencies.
- DGX Spark ships from July 2025; DGX Station is expected later in 2025.
- Both systems let users run AI workloads locally without cloud dependency, serving privacy-sensitive use cases.
With seven global PC makers in the supply chain, deskside AI supercomputing is moving from niche research instrument to mainstream enterprise hardware.
Potential risks and opportunities
Risks
- If DGX Station's 'later in 2025' timeline slips, all seven manufacturing partners face channel commitments and inventory planning they cannot easily unwind.
- Enterprises deploying 20-petaflop deskside hardware face physical security and insider-threat exposures that current IT endpoint policies were not designed to address.
- DGX Spark's July 2025 availability window gives competitors several months to announce rival desktop AI systems that could soften pre-reservation momentum before first shipments.
Opportunities
- Enterprise AI software vendors (Hugging Face, Weights and Biases, Scale AI) can target DGX Spark and DGX Station as a new certified on-premise deployment tier across seven established PC distribution networks.
- Managed IT service providers can build DGX installation, tuning, and support offerings as seven global PC makers bring these systems into standard enterprise procurement channels.
- Organizations with data-residency requirements in healthcare, finance, and government gain a certified alternative to cloud inference, accelerating budgets for on-premise AI hardware deployments.
What we don't know yet
- DGX Station pricing is not disclosed; whether it falls within enterprise workstation budgets or requires data-center-level capital spend remains unknown.
- The announcement does not specify which AI frameworks, model repositories, or software stacks are pre-validated or certified for either system.
- Whether all seven PC partners will offer both DGX Spark and DGX Station, or each will specialize by product line, is unspecified as of the COMPUTEX launch.
Originally reported by nvidianews.nvidia.com
Read the original article →Original headline: NVIDIA DGX Station for Windows Launches at GTC Taipei — GB300 Superchip, 20 PFLOPS FP4, Runs 1-Trillion-Parameter Models Locally, Partner Shipments Begin This Month