At Computex 2025 in Taipei, chipmakers Nvidia, AMD, and Intel unveiled desktop GPUs designed to run generative AI models locally — marking a pivotal shift away from cloud dependence. As generative AI models grow leaner and more optimized, these new GPUs bring supercomputing power to the desktop, reshaping the AI development and deployment landscape.
Nvidia Leads with Project DIGITS for AI-Only Desktops
Nvidia has unveiled the DGX Spark desktop, previously known as Project DIGITS. This compact, AI-focused PC will launch in July through partners such as Dell, HP, and Lenovo. Featuring Nvidia’s new Blackwell GPU, the DGX Spark delivers an impressive 1 petaflop of performance, effectively acting as a personal AI cloud. Designed specifically to run advanced reasoning models like DeepSeek R1, it brings powerful AI computing directly to the desktops of developers and researchers.
AMD and Intel Join the Local AI Race with New Desktop GPUs
AMD’s new Radeon 9700 AI Pro GPU packs 128 AI accelerators and can run the DeepSeek R1 model with up to 32 billion parameters. In high-performance setups using four of these GPUs, it can even handle Mistral’s massive 123B parameter model — all locally, without relying on the cloud.
Intel’s Arc Pro B50 offers a more budget-friendly option at $299. It also features 128 AI engines but is limited to 16GB of memory, allowing it to run DeepSeek R1 models up to 14 billion parameters. It’s a solid entry point for local AI model development.
AI PCs Drive New Desktop Market Momentum
Industry experts believe AI-powered desktops are redefining the PC landscape. “AI development is shifting back to personal computing,” said Jack Gold, Principal Analyst at J. Gold Associates.
While large models may still run in the cloud for deployment, tasks like development, training, and inference are increasingly happening on local machines—especially with edge-optimized tools.
Generative AI Is Going Local
Generative AI workloads are becoming more efficient and optimized for local execution, reducing energy and compute requirements. According to Anshel Sag of Moor Strategy, businesses are hesitant to send sensitive AI workloads to the cloud. These new GPUs offer a secure and cost-efficient alternative
During his Computex address, Nvidia’s Huang called DeepSeek R1 “a gift to the world’s AI industry” — underlining how the new wave of desktop GPUs for generative AI is more than just hardware; it’s a path to AI democratization.
Conclusion
With powerful new GPUs launching in July and models like DeepSeek becoming mainstream, generative AI is no longer just for the cloud. The rise of local inference and AI PCs marks a major shift in how, where, and by whom AI can be built and used.
About The Author
