In addition to announcing collaborations with MediaTek and Microsoft,RTX Spark Super ChipNVIDIA's collaboration with Microsoft extends beyond consumer laptops, with the additional announcement of the "NVIDIA DGX Station for Windows," arguably the most powerful desktop AI supercomputer on the planet. This system, expected to launch in the fourth quarter of this year, seamlessly integrates the ease of daily management of Microsoft Windows with NVIDIA's top-of-the-line GB300 Grace Blackwell Ultra desktop superchip, allowing businesses to run cutting-edge AI models and proactive agent workflows with up to 1 trillion parameters directly from their local workstations.
Because they use the same architecture, AI tasks executed on the RTX Spark platform can be seamlessly migrated to the DGX Station platform, which also supports the Windows operating system environment. This will also attract more users who are accustomed to using the Windows operating interface or whose tasks can only run in the Windows environment.
Powered by GB300 desktop super chip: 20 Petaflops computing power and 748GB unified memory
While maintaining the compact size of a desktop workstation, the DGX Station for Windows boasts monstrous internal hardware specifications:
• Core Hardware Heart:It features the new NVIDIA GB300 Grace Blackwell Ultra desktop-class super chip. Through NVLink-C2C chip interconnect technology, a Blackwell Ultra GPU is tightly connected to a high-performance NVIDIA Grace CPU with a 72-core design, thereby achieving high-performance end-to-end communication and computing performance.
• Rampaging performance:With a system configuration of up to 748GB of unified memory, it can unleash an astonishing 20 Petaflops of FP4 precision AI computing power.
• Physical AI expansion capabilities:To handle edge computing that requires real-time 3D rendering and virtual-to-physical interaction, the system can also be expanded with an additional NVIDIA RTX PRO 6000 Blackwell workstation graphics card, fully combining cutting-edge AI model computing power with ray tracing visualization simulation, and outputting higher AI computing power on enterprise desktops.
• The main artery of transmission:Built with an NVIDIA ConnectX-8 SuperNIC network chip, it natively supports high-speed network transmission of up to 800Gb/s. In addition to being able to transfer massive amounts of data at ultra-high speed, it also supports local cluster interconnection of multiple DGX Stations to handle larger computing tasks.
Building a "resident" AI agent infrastructure: Hundreds of AI agents performing parallel inference simultaneously
As enterprise AI evolves from simple chatbots with a "question and answer" mechanism to autonomous workflows that require 24/7 operation and have real-time reasoning capabilities, DGX Station for Windows is defined as Dedicated Agent Infrastructure.
Pavan Davuluri, Microsoft’s executive vice president of Windows and Devices, said the collaboration will elevate the full power of Windows, from thin and light laptops (RTX Spark) all the way to data center-level workstations.
With this local computing power, enterprise IT and development teams can simultaneously deploy hundreds of AI agents to perform different tasks. These AI agents can be deeply embedded into 3D design, engineering software, or existing automated workflows within an enterprise, becoming truly intelligent "teammates" who understand human productivity tools and processes.
Building a cybersecurity firewall in partnership with Microsoft: NVIDIA OpenShell sandbox mechanism
Running autonomous AI agents within a corporate intranet environment can be quite a headache for IT administrators, as it involves numerous security and privacy concerns. To address this, DGX Station for Windows utilizes the open-source and security-designed runtime environment—NVIDIA OpenShell.
A tamper-proof sandbox environment based on Windows underlying isolation:
OpenShell combines Microsoft's latest Windows Security and Containment Primitives to create a completely independent and isolated virtual sandbox for each AI agent running locally, thereby forcibly isolating application-layer operations from infrastructure-layer policies.
This means that all enterprise cybersecurity and privacy policies are placed at the highest level of the system, and the AI agent is "completely inaccessible" to these rules within the sandbox. Compared to the past reliance on prompts to constrain AI behavior, this hard physical isolation mechanism from the bottom up will ensure that the AI agent can never overwrite policies without authorization, and there is no possibility of tragedies such as credential leaks or leakage of confidential personal information.
Furthermore, for enterprise IT management teams, DGX Station for Windows allows them to leverage familiar Microsoft group principles, compliance checks, and device monitoring tools for unified scheduling. Existing Linux AI development toolchains can also benefit from the same management support and seamless porting through WSL (Windows Subsystem for Linux).
Core features of NVIDIA DGX Station for Windows:
| Application Domain Workflow | Hardware and system-level computing advantages |
| AI Agents | supportHundreds of cutting-edge AI agents are operating in parallel.It can also be directly integrated into enterprise applications within the Windows ecosystem. |
| AI Development | Supports large-scale AI models in the native Windows environment.Pre-training and fine-tuningAnd through WSL, it is perfectly compatible with the Linux AI toolchain. |
| Data Science | Possessing Gundam 748GB of Coherent MemoryIt can directly handle massive datasets, completely eliminating data transmission bottlenecks during data transfer. |
| Cutting-edge inference (AI Inference) | A burst of 20 Petaflops of computing power can be executed smoothly on a local machine.Up to 1 trillion (1 Trillion) parameter scaleThe giant model inference. |
| Physical AI | With the dual support of the GB300 super chip and the external RTX PRO graphics card, it perfectly achieves [the desired effect] within a single desktop system.Cutting-edge AI computing, ray tracing visualization and virtual physical environment simulation. |
Analysis: NVIDIA and Microsoft are pursuing a two-way race, threatening a historic reshuffle for x86 and Linux workstations.
Judging from NVIDIA's DGX Station for Windows, this is not just a top-of-the-line hardware workstation, but also a "two-way collaboration" with a high degree of strategic depth between NVIDIA and Microsoft in the enterprise market.
For a long time, high-end AI development has been dominated by Linux systems, while enterprise operations, cybersecurity compliance, and IT Fleet delivery management have largely been the strengths of Windows environments. This has led to significant compatibility and disconnect challenges for IT departments when delivering and conducting security reviews of internal Windows software clusters after data scientists develop AI models.
This time, NVIDIA directly put the Grace Blackwell Ultra super chip into the Windows platform, and with the cooperation of Microsoft from the Windows underlying layer, OpenShell, the physical sandbox technology is used to isolate the AI agent in hardware form to ensure that the AI agent will not "overstep the bounds" and engage in dangerous behaviors such as tampering with or deleting data.
From a market positioning perspective, DGX Station for Windows not only addresses the internal needs of enterprises for model inference and fine-tuning at scales of up to trillions of parameters, but also solves the problem that most enterprises may face data and privacy leaks, as well as bandwidth and time costs during data transmission, when sending data to the cloud for collaborative computing through remote AI factories.
Even with the same architecture as the RTX Spark platform, AI tools and fine-tuned models developed by enterprises on DGX Station for Windows can run directly on the RTX Spark platform, or conversely, tools used on the RTX Spark platform can be placed on DGX Station for Windows.
In addition, NVIDIA can also use this opportunity to further tap into the workstation market that has long been dominated by the x86 architecture, such as further impacting the competitive advantage of AMD's Ryzen AI Max and Ryzen AI Max PRO processors, which were originally designed to meet the needs of AI applications.
The first wave of DGX Station for Windows products will be launched in the fourth quarter of this year by companies such as ASUS, Dell, Gigabyte, HP, MSI, and Supermicro.
