Honor Released Tiangong AXB35 Compact AI Workstations
The new series of compact workstations aims to support local multi-agent workloads and large language models.
Updated on Sept. 18, 2026 in Computers
Live Poll
Would you consider using a compact, high-powered AI workstation for your personal or professional tasks?
Honor unveiled the Tiangong AXB35 workstation family, a new line of compact desktop systems designed for running local large language models. The lineup includes an Nvidia-powered Ultra model, an AMD-based Pro, and an Intel-based Standard version.
Why it matters
The AXB35 series addresses the rising demand for local compute capacity to handle large language models and multi-agent workflows without relying solely on cloud infrastructure. By packing high-performance silicon into a 2-liter chassis, the hardware targets specialized AI tasks in space-constrained professional environments.
The Ultra model delivers up to 1 PFLOP of FP4 AI performance via its Nvidia RTX Spark platform, which operates at a 140W power rating. Supporting models up to 300 billion parameters, the system includes 128GB of unified memory in a 1.5kg chassis measuring 70 by 192 by 203 millimeters.
The players
Honor
A global technology company focused on mobile devices, tablets, and increasingly compact AI-capable computing hardware.
Nvidia
A leading designer of graphics processing units and AI hardware platforms powering data centers and edge compute applications.
The details
The Ultra model integrates the Nvidia RTX Spark, a platform that pairs a Blackwell-based GPU with a Grace CPU. The Blackwell architecture is a specialized design for high-performance AI computation, while the Grace CPU uses an Arm-based design to handle massive memory throughput. To handle data, the system utilizes three M.2 slots—connectors for high-speed non-volatile storage—for internal capacity expansion.
Timeline
Honor unveiled the Tiangong AXB35 workstation family on September 18, 2026.
The Tech Race
Honor is entering the high-density workstation market to compete with established small-form-factor devices like the Apple Mac Studio. By prioritizing AI-specific throughput, the AXB35 series differentiates itself from traditional general-purpose desktops in the race for local model deployment.
These systems will support professional users who need to run 300-billion-parameter models locally rather than through external APIs. Users can expect compatibility with both Windows 11 and Ubuntu, with connectivity options including Wi-Fi 7 and dual 10GbE ports for high-speed networking.
The takeaway
The AXB35 series demonstrates the increasing feasibility of deploying high-parameter models on desktop hardware. Watch for Honor’s forthcoming announcement of retail pricing and regional availability to confirm if this platform will become a standard tool for AI development.
Further reading
For more on the current landscape of desktop computing, visit Computers.
Live Poll
Would you consider using a compact, high-powered AI workstation for your personal or professional tasks?
