NVIDIA's DGX Spark: A Compact AI Powerhouse Set to Launch in July 2025

Nvidia skips Computex main floor to launch mind-blowing DGX AI PCs you probably can't afford yet

Nvidia DGX Spark delivers 1,000 TOPS in a mini PC that targets serious AI developers and coders At Computex 2025, Nvidia took a somewhat unconventional route by sidestepping the main exhibition floor and instead hosting its own "GTC Taipei" event at a nearby hotel. There, Nvidia CEO Jensen Huang delivered a trio of keynote speeches at the event, unveiling new AI-focused hardware. Among the biggest announcements were two devices: the DGX Spark, a compact mini PC aimed at AI developers, and the DGX Station, a more powerful workstation-class system. Though Nvidia-branded units were showcased, the real surprise was the range of OEM partners joining the initiative, with 11 models expected across the Spark and Station lines. The DGX Spark is built around the new Nvidia GB10 Grace Blackwell Superchip, which combines a 20-core Arm CPU co-developed with MediaTek (featuring 10 Cortex-X295 and 10 Cortex-X725 cores) and a GPU based on Nvidia's Blackwell architecture. Designed for developers, the device delivers up to 1,000 TOPS (FP4/sparse) of performance and ships with a Linux-based DGX OS, Nvidia's AI development suite also used in its data center platforms like Blackwell and Hopper. Several partners, including Acer, Asus, Dell, Gigabyte, HP, Lenovo, and MSI, had models on display. At first glance, however, the only visible differences were in the external design. No internal teardowns were permitted during the event, raising a valid question: how different are these OEM versions beyond aesthetics? While the DGX Spark promises to be a strong contender for the title of best mobile workstation for AI development, potential buyers may want to wait for detailed reviews before making a purchase. The DGX Station, aimed more directly at professionals needing high-end workstation PCs, features the GB300 Grace Blackwell Ultra Desktop Superchip. It comes equipped with 288GB of HBM3e memory on the GPU and a 72-core Neoverse V2 CPU paired with 496GB of LPDDR5X RAM, making it far from an ordinary desktop. Like the Spark, it runs on DGX OS and supports Nvidia's full AI development stack. The DGX Station board shown at the exhibition was a mockup, though the actual product was displayed during a separate session. Still, questions remain about how finalized the systems are, especially since full availability isn't expected until late 2025. Notably, the DGX Station will only be available through OEMs, with Asus, Gigabyte, MSI, and Supermicro leading the rollout. This fragmented distribution model could introduce variation in build quality and thermal performance, critical factors for users seeking the best workstation PC.

NVIDIA's high-perf DGX Spark AI supercomputer should launch in July, AIBs preparing systems now

As an Amazon Associate, we earn from qualifying purchases. TweakTown may also earn commissions from other affiliate partners at no extra cost to you. NVIDIA has reportedly told its OEM partners to begin shipping its DGX Spark AI supercomputers in July, with AIB partners to jump head-first into the new business opportunity according to the latest reports. In a new report from UDN, we're hearing that NVIDIA's new DGX Spark system will be joined by the company's other personal AI supercomputer -- the DGX Station -- to be launched later this year. UDN says that analysts are "optimistic" that NVIDIA's personal AI supercomputers are going to be launched one after another, which will help boost Q3 2025 performance of the related suppliers. These new AI systems enable AI developers, researchers, data scientists, and students to prototype, fine-tune, and inference large models on desktops. NVIDIA says that users can run these models locally or deploy them on NVIDIA DGX Cloud or any other accelerated cloud or data center infrastructure. Inside, the new DGX Spark uses NVIDIA's new GB10 Grace Blackwell Superchip, which has been optimized into a desktop form factor. NVIDIA's new GB10 features a powerful NVIDIA Blackwell GPU with fifth-generation Tensor Cores and FP4 support, with 1000 TOPS of AI compute power for fine-tuning and inference with the latest AI reasoning models, including the NVIDIA Cosmos Reason world foundation model and NVIDIA GR00T N1 robot foundation model. NVIDIA's new GB10 Superchip uses NVIDIA NVLink-C2C interconnect technology, which delivers CPU+GPU-coherent memory model with 5x the bandwidth of PCIe Gen5. This lets the Superchip access data between a GPU and CPU to optimize performance for memory-intensive AI developer workloads.

NVIDIA's Highly-Anticipated AI Mini-Supercomputer "DGX Spark" Rumored To Launch By July; Here's a Rundown of What To Expect

NVIDIA has indirectly entered the "AI PC" segment with its first consumer-focused AI supercomputer, the DGX Spark, which will launch soon. Team Green has been focused on catering to the larger side of the AI markets, which include clusters and data centers. Last year, the firm showcased its compact supercomputer, which is essentially targeted towards end users and individuals who want a high-performing AI device. For now, NVIDIA hasn't disclosed the launch date for DGX Spark, but Taiwan Economic Daily reports that DGX Spark could launch by July, with AIBs like Gigabyte and ASUS shipping out units to customers globally. For those unaware, the DGX Spark is NVIDIA's smallest AI device to date, offering performance that almost seems impossible given the device's size. While the specifics of the supercomputer are unknown, it is revealed that DGX Spark features the GB10 Grace Blackwell Superchip, which comes with the powerful NVIDIA Blackwell GPU with fifth-generation Tensor Cores and FP4 support, delivering up to 1,000 trillion operations per second of AI compute for fine-tuning and inference. The GB10 Superchip uses NVIDIA NVLink-C2C interconnect technology to deliver a CPU+GPU-coherent memory model with 5x the bandwidth of fifth-generation PCIe. This results in the device delivering up to 1,000 TOPS of AI power, which is a massive figure considering the system's compactness. However, NVIDIA has managed to squeeze out high performance with the architectures onboard. Interestingly, DGX Spark has been expanded to AIBs, and they will offer separate variants, with MSI and Gigabyte being the leading ones. On our visit to Computex 2025, we saw several manufacturers showcasing their implementation of DGX Spark, with the key difference lying in the overall design language. MSI's version of DGX Spark is called EdgeXpert MS-C931, which features 128 GB LPDDR5X and a simple black and grey design with HDMI and Connect-X ports for multi-connectivity. There's nothing too exciting going around with it since the device is oriented towards professional consumers, but it will be interesting to see how it performs in gaming scenarios. Next is Gigabyte's AI TOP ATOM device, the firm's version of DGX Spark. Yet again, the design was pretty similar to MSI's, since there's nothing AIBs could change in it. However, Gigabyte did show us the internals of their system, where we managed to spot Micron's LPDDR5X memory chips. Gigabyte claimed that their AI TOP ATOM supercomputer offers up to 70 billion parameters, and with two ATOMs working together using NVIDIA Connect-X, they can bring up to 405 billion parameters for AI training.