Giga Computing shows at Computex 2026 from mini PC for agents to a desktop supercomputer with GB300
In it Computex 2026which we are covering from Taipei, Giga Computing, a subsidiary of GIGABYTE, has taken advantage of its stand to show a fairly broad portfolio of AI infrastructure based on NVIDIA. The proposal ranges from rack platforms for large-scale training and inference to desktop systems, mini PCs for agents and modular solutions to deploy computing capacity in a container format.
The most ambitious piece in the showcase is the NVIDIA Vera Rubin NVL72a platform aimed at training, inference and uploads of agentic artificial intelligence. Giga also accompanies it with an 800 VDC electrical distribution solution, a sign of where the sector is moving as racks gain density and their energy needs increase.
Along with that more part of the data center, the company has set up a flow demonstration real-to-sim-to-real. The process starts with servers with NVIDIA RTX PRO 6000 Blackwell to capture real-world data and recreate simulated environments, then moves to a system with NVIDIA HGX Rubin NVL8 to train and optimize models, and ends with deployment in physical systems supported by Jetson Orin NX, such as robots or smart factories.
From data center to desktop to modular deployment
The most striking part outside the rack is the GIGABYTE W775-V10a desktop supercomputer with closed-loop cooling based on the NVIDIA GB300 Grace Blackwell Ultra Desktop Superchip. It is designed for local development of AI, inference, data science and agents, and Giga already anticipates that its GB300 systems will also arrive on Windows later this year.
Below in size appears the AI TOP ATOM, a mini PC of around one liter designed for developersresearchers and data scientists. Giga proposes it as a platform to run multi-model inference and autonomous flows with NemoClaw, within the NVIDIA software ecosystem.
The strategy is completed with GAIFA and GADU. The first will be a new AI Factory in Taiwan scheduled for the end of the third quarter, while the second proposes a modular and containerized infrastructure, something like a deployable AI factory in a box. With this, Giga Computing tries to cover the entire chain, from the laboratory and the edge to the rapid launch of capacity at scale.
