Minisforum unveils compact AI infrastructure set to challenge traditional enterprise hardware

Minisforum launches a dual-device local AI stack based on AMD Ryzen AI Max+ Pro 495, promising high performance and increased privacy for model execution and data storage, disrupting traditional server and cloud reliance.

Minisforum is turning its latest artificial intelligence push into a small on-premises stack rather than a single box: a compact workstation for model execution at the front end and a companion NAS for keeping data, models and agents local at the back end, both built around AMD’s Ryzen AI Max+ Pro 495. The pair was unveiled in Berlin on 4 September at IFA 2026, and by 5 September the workstation had already appeared on Minisforum’s own regional web shops even though a final public price was still missing. (minisforum.com)

That positioning matters because Minisforum is not presenting the launch simply as two faster compact PCs. In its own release, the company described the MS-S1 Max-P495 as the compute front end and the N5 Max-P495 as the storage-and-data back end of a local AI system. Across the pair, Minisforum is claiming up to 131 TOPS of AI performance, up to 192GB of LPDDR5X memory running at 8,533MT/s, and as much as 160GB that can be assigned to graphics. In plain terms, that means a small machine can hold much larger models in memory and process them without shuttling work out to a remote cloud service, which is the privacy and latency argument running through the whole launch. (minisforum.com)

The storage half of that strategy is the N5 Max-P495. Minisforum says it can be configured with up to 200TB of local capacity and used for RAG knowledge bases, embeddings, databases and agents including OpenClaw and Hermes. In the company’s words, it is meant to act as a “centralized backend for data, models, knowledge, and long-running AI workloads”. Regional coverage in Spain and France pushed the same point from a different angle, describing the machine as part of a local-AI infrastructure pair aimed at keeping vector databases, persistent agents and sensitive information on site rather than in third-party services. (minisforum.com)

The MS-S1 Max-P495 is the more outward-facing product because it is the system intended to sit on a desk or in a rack and perform the live inference work. Russian and Chinese reports described it as an update to the earlier MS-S1 rather than an all-new platform, with 55 TOPS coming from a dedicated NPU and support for models from Gemma, OpenAI and Qwen. GeekPark said Minisforum is pitching the box for “millisecond-level” local inference in agent workloads, computer vision and generative AI. On Minisforum’s own product page, the company goes further, claiming that two 192GB units can run Qwen3.5-397B locally and that a four-unit cluster in a 2U rack has already been used to run a quantised DeepSeek-R1 671B model, though those figures are vendor claims rather than independent benchmarks. (store.minisforum.com)

Notebookcheck’s reporting filled in the hardware details that the launch material left vague. It said the machine keeps the same broad formula as the existing MS-S1: a 16-core, 32-thread AMD APU in a chassis of about 3.5 litres, but with slightly higher clocks, faster memory support and a Radeon 8065S integrated GPU replacing the older 8060S. The publication also reported Minisforum’s claim of a “top-tier thermal solution” with six heat pipes, dual fans and up to 160W of dissipation. For workstation buyers, the more practical details may be the ports and expansion: four USB4 connections in total, including two USB4 v2 ports at 80Gbps, dual 10GbE, HDMI 2.1, Wi‑Fi 7, Bluetooth 5.4 and a full-length PCIe x16 slot wired at PCIe 4.0 x4. (notebookcheck.net)

Availability is still the least settled part of the story. Notebookcheck said the new MS-S1 was expected to launch later in September and reported that Minisforum had teased a four-digit price beginning with 7, implying a figure near $7,000. Minisforum’s own US page currently shows only “$7???” above a subscription prompt, while its UK and EU storefronts listed the model on 5 September with placeholder prices of £0 and €0 and a sold-out label. That suggests the sales pages are live before regional pricing is ready. GeekPark likewise said only that the MS-S1 Max-P495 would go on sale in September 2026. Minisforum has still not published a firm launch timetable or a retail price for the new N5 Max-P495. (notebookcheck.net)

There is also a broader industry angle here. Webdisclosure cast the two machines as a matched local-AI infrastructure set intended to “democratize” access to stronger on-device computing, while Liberal de Castilla linked the announcement to a wider move to shrink enterprise-style AI hardware into formats suitable for offices, professional desks and high-end home setups. The same Spanish report noted that Minisforum had already built visibility at IFA the previous year, when its N5 Pro picked up design recognition. The company is therefore using IFA 2026 not just to refresh specifications, but to argue that compact hardware can now cover workloads that previously implied a rack server or a cloud subscription. (webdisclosure.fr)

For now, visitors in Berlin can see both systems at Hall 6.2-104 through 8 September. The more important test will come after the exhibition floor closes: whether the final pricing makes sense against Minisforum’s current range, where the existing UK-listed MS-S1 Max 128GB sells for £3,399 and the current N5 Max AI NAS for £2,359, and whether buyers actually want a split local-AI stack instead of a single, larger workstation. On specifications alone, Minisforum has made a serious bet that privacy, low latency and very large unified memory pools are becoming product categories in their own right. (minisforum.com)