Honor Unveils Tiangong AXB35, a Tiny AI Workstation Built for Local LLMs
Honor has revealed the Tiangong AXB35 series, a new family of compact AI workstations designed to bring powerful local artificial intelligence processing into a surprisingly small desktop form factor. The standout model is the Tiangong AXB35 Ultra, a miniature 2-liter workstation powered by Nvidia’s RTX Spark platform and aimed at demanding AI development, multi-agent workflows, and enterprise-level large language model deployment.
Despite measuring just 70 × 192 × 203 mm and weighing around 1.5 kg, the Tiangong AXB35 Ultra is built to handle serious AI workloads. Honor is positioning the system as a compact alternative to larger desktop workstations, giving businesses and developers a way to run advanced AI models locally without relying entirely on cloud infrastructure.
The Ultra model uses Nvidia’s RTX Spark platform, which combines a Blackwell-based GPU with a Grace CPU. Nvidia rates the platform at up to 1 PFLOP of FP4 AI performance, making it suitable for high-efficiency inference and local AI processing. The desktop version of the chip can include up to 6,144 CUDA cores, a 20-core Arm processor, and up to 128GB of LPDDR5X unified memory.
Honor pairs this hardware with a 2TB NVMe SSD, giving the compact workstation enough storage for AI models, datasets, and professional applications. Nvidia lists the platform at up to 140W, while Honor has designed the AXB35 family with multiple operating modes rated at 55W, 85W, and 132W. The system is powered by an external 240W adapter.
Honor says the Tiangong AXB35 Ultra is intended for multi-agent AI workloads and large-scale local model deployment. The company mentions support for large language models with up to 300 billion parameters, though real-world performance will depend on several factors, including quantization, context length, software optimization, and the number of simultaneous users or agents.
In addition to the flagship Ultra version, Honor is preparing two more Tiangong AXB35 configurations.
The Standard model features an Intel Core Ultra X7 358H processor, 64GB of memory, and a 512GB SSD. Honor lists this configuration at up to 180 TOPS of AI performance, making it a practical option for lighter AI agent workloads, productivity tasks, and smaller local language models.
The Pro model switches to AMD hardware, using the Ryzen AI Max+ 395 with 128GB of memory and a 1TB SSD. Honor rates this version at up to 126 TOPS, positioning it between traditional compact PCs and more advanced AI workstations.
All three Tiangong AXB35 models share the same small chassis and workstation-focused connectivity. The systems include two 10GbE ports, Wi-Fi 7, two USB4 ports, four USB-A ports, HDMI 2.1, and DisplayPort 1.4. Storage expansion is also supported through three M.2 slots, which is useful for users working with large AI models, media files, or enterprise datasets.
Software support includes both Windows 11 and Ubuntu, giving developers, researchers, and businesses flexibility depending on their workflow. Windows support makes the machine suitable for general productivity and creative applications, while Ubuntu compatibility should appeal to AI developers and teams working with open-source machine learning tools.
Honor has not yet confirmed pricing or general retail availability for the Tiangong AXB35 lineup. For now, interested customers are being directed to contact the company for configuration details, which suggests the systems may initially target enterprise buyers, AI labs, developers, and professional workstation users rather than the mainstream consumer PC market.
With the Tiangong AXB35 series, Honor is clearly aiming at the growing demand for compact local AI hardware. As companies look for faster, more private, and more cost-controlled ways to run AI models outside the cloud, small but powerful workstations like the AXB35 Ultra could become increasingly attractive.
