Nvidia has officially expanded its personal AI supercomputer portfolio by introducing a new 64GB unified memory configuration of the DGX Spark workstation, set to retail starting at $4,999. The release comes as severe memory supply constraints continue to drive up component costs across the tech industry, forcing a major price adjustment on the original 128GB version.

By offering a lower memory threshold, Nvidia aims to maintain an entry level option for researchers, software engineers, and local AI developers who require on-device inference and autonomous agent processing without paying escalating cloud subscription fees or higher enterprise hardware tags.

Nvidia 64gb dgx spark workstation: Summary of the Announcement

The newly announced workstation package delivers the full architectural capabilities of the original system, retaining the centerpiece GB10 Grace Blackwell Superchip, high-speed ConnectX-7 networking, and the complete Nvidia AI software stack. However, to offset soaring memory market dynamics, Nvidia halved the coherent system memory capacity to 64GB. The system is slated to begin shipping on October 23, 2026, exclusively through official original equipment manufacturer (OEM) partners.

Nvidia Expands DGX Spark Workstation Lineup with 64GB Model

Designed as a compact desktop unit capable of running sophisticated generative models locally, the DGX Spark series addresses growing demand for private, offline AI workloads. By launching the 64GB SKU, Nvidia creates a secondary tier that accommodates smaller budgets while keeping the overarching hardware architecture intact.

Industry analysts view the move as a strategic necessity rather than a planned product refresh. As component costs escalate, Acer warned PC prices could increase up to 20 percent in Q4 2026, demonstrating how widespread memory cost pressures have become across consumer and workstation segments alike.

Despite the cut in coherent memory capacity, Nvidia emphasizes that the 64GB workstation remains a robust platform for modern developer workflows. It natively supports popular local runtime environments like vLLM, Ollama, LM Studio, and llama.cpp. Users can execute parameter-dense local models up to 35 billion parameters at high precision, or run quantized 100-billion-parameter variants right from their desks.

Specifications and AI Workload Performance Capabilities

At the core of the new 64GB system is the multi-die GB10 Grace Blackwell Superchip, combining an enterprise Arm processor with Blackwell graphics architecture. The silicon includes fifth-generation Tensor Cores and native FP4 precision acceleration, achieving up to 1 petaflop of AI compute output.

Key hardware and performance specifications include:

  • Processor: Nvidia GB10 Grace Blackwell Superchip.
  • System Memory: 64GB coherent unified memory.
  • Compute Performance: Up to 1 petaFLOP at FP4 AI precision.
  • Networking: Integrated Nvidia ConnectX-7 200GbE networking.
  • Software Ecosystem: Pre-loaded DGX OS with Nvidia AI Enterprise tools, NIM microservices, and OpenShell agent frameworks.

The integrated ConnectX-7 networking stack allows developers to interconnect multiple systems directly. Using software tools like Nvidia Sync Cluster Assistant, two 64GB workstations can be combined over high-bandwidth links to form a unified 128GB local cluster. This multi-node capability enables teams to scale compute capacity incrementally as model sizes and project requirements grow.

Memory Supply Crunch Drives Up Price of 128GB DGX Spark

The launch of a $4,999 entry tier comes alongside significant price increases for the premium 128GB model. Originally debuted in late 2025 around $3,999, the 128GB version saw retail pricing move up dramatically toward $6,950 across various retail channels due to global supply shortages in high-performance memory chips.

Market conditions across memory modules have tightened significantly throughout the year. Reports like those from TrendForce projecting PC DRAM contract prices to rise up to 18 percent in 4Q26 highlight the intense demand for high-bandwidth DRAM. AI server deployments by tech giants and massive infrastructure expansions, such as xAI expanding its Colossus 2 cluster to target 1.21 million Nvidia GPUs, have consumed huge swaths of global memory production capacity.

The resulting supply crunch has affected hardware builders across every tier. Other hardware sectors are experiencing similar strains, as seen when Raspberry Pi increased 2GB Pi 4 and Pi 5 prices by $12.50 citing memory costs. Even mobile workstation vendors have been forced to recalculate pricing models, such as when XMG raised laptop prices as DDR5 SO-DIMM RAM costs continued to surge.

OEM Partner Availability and Ship Dates for October 2026

Unlike standard consumer graphics cards, the 64GB Nvidia DGX Spark workstation will not be sold directly as an unbranded reference design. Instead, Nvidia is distributing the new system exclusively through its ecosystem of OEM partners.

Beginning October 23, 2026, system builders Acer, ASUS, Dell, Gigabyte, HP, and MSI will begin shipping custom pre-built configurations of the 64GB unit. Depending on the vendor, system storage configurations are expected to range between 1TB and 4TB NVMe solid-state drives, which may influence final end-user shelf pricing.

Reactions from the AI community remain mixed. While developers appreciate having a hardware platform under $5,000 to experiment with local autonomous agents and fine-tuning, many express concern that rising component costs could restrict local research. Industry analysts note that until memory fabs increase production yield in late 2027, hardware vendors will likely rely on segmented product tiers like the 64GB Spark to navigate component scarcity.

Overall, Nvidia's introduction of the 64GB DGX Spark workstation illustrates how hardware makers are adapting to economic realities while preserving access to desktop-class AI hardware.