MediaTek has deepened its collaboration with NVIDIA through the co-design of the GB10 Grace Blackwell Superchip, the core of NVIDIA’s newly launched DGX Spark — a personal AI supercomputer that brings the power of large-model development to the desktop.
Announced earlier this year, the DGX Spark system officially becomes available to the public starting October 15, enabling developers to prototype, fine-tune, and inference large-scale AI models locally. The collaboration marks another milestone in the long-standing partnership between the two companies, which spans data center computing, automotive platforms, IoT solutions, and consumer devices.
At the heart of DGX Spark lies the GB10 Grace Blackwell Superchip, designed jointly by NVIDIA and MediaTek. The chip leverages MediaTek’s advanced design expertise in CPU, memory subsystems, and high-speed interface technologies, delivering both power efficiency and high performance. The GB10 features a 20-core Arm-based Grace CPU, integrated with NVIDIA’s next-generation Blackwell GPU and 128GB of unified memory, achieving up to 1 petaFLOP of AI performance.
This level of compute power allows developers to work with AI models of up to 200 billion parameters directly on their desktops. For even larger workloads, DGX Spark includes ConnectX-7 networking technology, which allows two DGX Spark systems to connect and jointly handle inference on models up to 405 billion parameters. Despite its immense performance capabilities, the DGX Spark is compact enough to fit on a standard desk and power-efficient enough to run on a standard electrical outlet, making it one of the most accessible AI supercomputers ever built.
“DGX Spark will usher in the next era of AI prototyping and advance our mission of making great technology more accessible from the edge to the cloud, while solving performance and power consumption challenges,” said Vince Hu, Corporate Vice President of MediaTek’s Data Center and Compute Business Group. “The GB10 Superchip utilizes our high-performance computing expertise for the data center in combination with our power-saving technologies for consumer devices, custom-built to run AI workloads.”
The DGX Spark’s design reflects NVIDIA’s ongoing mission to democratize AI compute — extending its DGX portfolio from hyperscale data centers to the desktop, where developers, researchers, and startups can experiment with advanced models locally. For MediaTek, the collaboration highlights its expanding role as a key player in AI infrastructure and edge computing, a strategic evolution beyond its traditional leadership in mobile and consumer electronics.
The GB10 Superchip draws upon MediaTek’s deep experience in delivering high-performance, power-efficient silicon architectures. The company’s engineers worked closely with NVIDIA to optimize the GB10’s thermal design, interconnect architecture, and memory subsystem, ensuring efficient handling of massive model training and inferencing workloads. The result is a platform that offers exceptional performance per watt — a key metric for sustainable AI infrastructure.
DGX Spark also represents a significant shift in how developers interact with AI models. By providing a personal AI supercomputer, NVIDIA aims to empower individual creators, research teams, and enterprises to fine-tune and test models without relying exclusively on cloud resources. This approach enables faster iteration, greater data privacy, and cost control, particularly in regulated sectors such as healthcare, finance, and defense, where on-premise AI capabilities are essential.
Beyond the desktop, the MediaTek–NVIDIA collaboration has far-reaching implications for industries building next-generation intelligent systems. The companies have already worked together on AI platforms for autonomous vehicles, IoT applications, and high-efficiency computing in edge environments. The GB10 Superchip now extends that synergy into the data center and enterprise AI space, paving the way for scalable solutions that integrate seamlessly across environments — from personal computing to cloud infrastructure.
“Our collaboration with NVIDIA is rooted in a shared vision of performance, efficiency, and scalability,” added Hu. “By combining our leadership in low-power system design with NVIDIA’s world-class GPU and AI expertise, we’re enabling the next wave of intelligent computing across every device and deployment model.”
As the AI landscape accelerates, the launch of the DGX Spark underscores a growing demand for localized, high-performance AI computing that bridges the gap between consumer-grade hardware and hyperscale infrastructure. With the GB10 Superchip at its core, the DGX Spark brings supercomputer-class AI capabilities within reach of developers worldwide — redefining what’s possible for model innovation, experimentation, and deployment at the edge.
