FusionXpark M Series Desktop AI Supercomputer Global Debut

Zhengzhou, China, Sep 21, 2026 — xFusion officially launched FusionXpark M Series desktop AI supercomputers for the global market during xFusion Explorer 2026. The series embodies xFusion's vision and roadmap for bringing localized AI computing to end devices in the era of intelligent enterprises.

Jovi Chi, AMD's Vice President of Marketing for Greater China, and Zhou Xun, xFusion's General Manager of Edge Computing PDT, attended the release ceremony. Built around the proposition "Transform the Way You Work", the productivity enabled by large models is moving from the cloud to every desktop.

Strategic Positioning
Bringing Token Factory Closer to Users

Since introducing the AI value chain of "WATT → FLOPS → TOKENS → AGENTS → VALUES" in May 2026, xFusion has continued to advance its token factory system around the concept of "Token as the unit of cost and agent as the engine of profit". To address the needs of different scenarios, xFusion has established a three-tier production model: large-scale production in data centers with FusionServer, agile production in offices with TokenBox™, and desktop production with FusionXpark. While the three tiers differ in form, they are rooted in the same core capabilities: the continuous production, governance, and operation of tokens. The global debut of FusionXpark M Series marks the latest milestone in this strategy.

"Edge computing is not simply an extension of centralized computing. It is where AI meets the physical world to create real industrial value." Zhou Xun has reiterated this point in previous speeches. In this context, computing is moving from the center to the edge, and from the edge to the individual. FusionXpark M Series represents this evolution at the personal level. It brings AI closer to each individual, making it more accessible and easier to use.

Zhou Xun, xFusion's General Manager of Edge Computing PDT

FusionXpark M495
AI Agents for the Workplace, Unlocking AI Productivity for Everyone

FusionXpark M Series offers enterprise-grade desktop AI supercomputers built for office AI agents, on-premises AI deployment, and on-device model inference. Powered by AMD's Ryzen™ AI Max+ platform, it features an open hardware architecture that helps enterprises build a controllable edge AI computing foundation.

This release features two models: FusionXpark M495 and FusionXpark M395

As a key highlight of this event, FusionXpark M495 is positioned as the enterprise flagship for intelligent office work. Powered by AMD's Ryzen™ AI Max+ PRO 495 processor, it integrates a 16-core, 32-thread Zen 5 CPU and a Radeon™ 8065S GPU with 40 RDNA 3.5 compute units. Equipped with up to 192 GB of unified memory and up to 160 GB allocable to graphics memory, it delivers powerful local AI inference performance. Native support for Windows 11, together with the pre-installed FusionXplay platform, enables one-click model deployment and streamlined local AI management. It supports smooth on-device inference for models with up to 300B parameters, providing an all-in-one solution for design, AI creation, office productivity, and meeting collaboration.

For enterprise users, controllable means certainty in two key dimensions: models run locally, so responsiveness is no longer limited by network conditions or cloud compute availability; and data stays on the device, ensuring that core business assets stay securely within the enterprise. As AI agents take on scheduling, document work, meetings, and content creation, FusionXpark M Series becomes the foundation for every employee's personal AI agent.

To support enterprise AI workloads and large-scale private knowledge base processing, the entire FusionXpark M Series features 40 Gbit/s daisy-chain clustering and OCuLink GPU expansion, enabling multi-device clusters to locally deploy models with up to 671B parameters.

Also unveiled at the event was FusionXpark M395, featuring 128 GB of unified memory and access to AI applications through FusionXplay, bringing the power of AI from enterprise productivity to personal creativity and daily use scenarios.

Partnering with AMD for Global Expansion

Deep collaboration with the AMD platform is a key foundation of the FusionXpark M Series launch. xFusion and AMD have a strong track record of joint innovation in high-performance computing and enterprise-grade computing. The joint unveiling of the new products by senior executives from both companies marks the partnership's further expansion into the edge AI agent market.

Supported by more than 30,000 ecosystem partners and the application ecosystem of FusionXplay, FusionXpark M Series will leverage xFusion Explorer 2026 as a platform to accelerate its global expansion, enabling more enterprise and individual users to access AI agent capabilities locally with ease.

From SuperNodes in data centers, to TokenBox™ on-premises AI superstations, and to FusionXpark desktop AI supercomputers and FusionXtation™ X3 intelligent workstations, xFusion is extending the capabilities of its token factories to every scenario in the agentic era. "Transform the way you work" is more than a proposition — it represents xFusion's systematic vision for the future of work in the agentic era.