Nvidia unveiled DGX Spark, the new tiniest supercomputer in the world, on Wednesday. The trillion-dollar tech giant offers data center-class performance and includes the GB10 Grace Blackwell superchip, as well as its ConnectX-7 networking capabilities and software stack.
This initiative was introduced to provide small-scale businesses and developers with access to an AI computing system without worrying about spending money on renting an AI data center service or buying new AI servers
The Spark is rolling out with third-party partners, including ASUS and Dell, which will allow customers to develop and use AI models on their own local devices.
DGX Spark is part of a new class of computers Nvidia launched with third-party partners like ASUS and DELL, which allows users to develop and use AI models on their local devices.
Nvidia said the Spark can be equipped with up to 128GB of memory, which should help when running giant AI models.
“In 2016, we built DGX-1 to give AI researchers their own supercomputer. I hand-delivered the first system to Elon Musk at a small startup called OpenAI — and from it came ChatGPT, kickstarting the AI revolution,” Nvidia CEO Jensen Huang expressed in a statement.
“DGX-1 launched in the age of AI supercomputers and unlocked the scaling principles that drive modern AI. With DGX Spark, we return to that mission — placing an AI computer in the hands of every developer to kindle the next wave of breakthroughs.”
Like the DGX-1, Huang also hand-delivered one of the first Spark systems to Musk.
He mentioned that his company has a larger and more powerful DGX Station, which runs on Nvidia’s GB300 Grace Blackwell Ultra desktop superchip.
The Spark and Station systems are specifically designed for building and running AI models, utilizing Nvidia’s Linux-based DGX OS and its AI software stack.
Read More: World’s fastest supercomputer showcased at GITEX 2025
They are not intended for running video games or other general applications. These systems complement Nvidia’s existing AI offerings, allowing users to prototype AI models before broader deployment.
Additionally, two Spark systems can be connected to accommodate even larger AI models, supporting up to 405 billion parameters.
Nvidia maintains a top position in the AI chip race thanks to both its powerful processors and CUDA software service.
In addition to its desktop systems, the company has secured important deals with AI companies. These include a $100 billion investment in OpenAI, which will involve the ChatGPT developer acquiring 10 gigawatts of GPUs.
Last month, Nvidia also finalized an agreement for CoreWeave to purchase $6.3 billion worth of GPUs.
The company is a key provider of GPUs to major tech firms, including Elon Musk’s xAI, Amazon, Google, Meta, and Microsoft, as well as the Stargate Project, a joint venture between OpenAI and Oracle.