CiXin Technology's 2026 Customer Conference, held in Shenzhen on September 18, marked a significant milestone as the company rolled out its latest innovations. The event highlighted how AI has progressed through voice recognition, interactive AI, and generative AI, now entering a new phase defined by intelligent agents, where CPUs have become as crucial as AI inference and training in the computing foundation.
During his keynote, CiXin Technology's founder and CEO, Sun Wenjian, emphasized that the agent era opens a fresh industrial window for terminal hardware, agent software, and data center integrators, presenting a historic opportunity for China's local ecosystem to rise. At the conference, the company formally introduced the AGX X2 (AGXcelerator) platform, alongside several terminal products and industry solutions based on its SoC, developed with ecosystem partners, showcasing the collaborative strength and large-scale deployment of the computing foundation in the agent era.
Danny Zhang, Chief Marketing Officer at CiXin Technology, detailed the platform's capabilities, noting that the AGX X2 is purpose-built for edge inference and Agentic AI. It delivers 80 to 300 TOPS of on-device peak computing power, supports native inference for large models ranging from 7B to 122B parameters, and features 64 to 160GB of LPDDR5 unified memory. This configuration achieves a over 50% reduction in per-token energy consumption compared to industry standards, making it the only domestic solution to halve energy usage.
The platform addresses three major industry hurdles for edge devices—limited model scale, insufficient memory bandwidth, and power and thermal constraints—through systematic solutions including high compute with large memory, high-bandwidth unified memory, and advanced process technology paired with heterogeneous scheduling. Beyond hardware performance, the AGX X2 embodies CiXin's vision for an open computing ecosystem, defining the Agentic Compute open architecture to expand multi-card and multi-model compute capabilities.
Positioned as an open AI supercomputing platform, the AGX X2 is compatible with multiple domestic AI accelerators, GPUs, and VPUs, supporting XPU cascading to scale computing power. Leveraging its proprietary Agentic OS, it adopts a "one operating system, dual-end coordination" model to enable cloud-device synergy, using Model-as-a-Service (MaaS) for unified token production and metering. This builds a complete business loop, converting every unit of computing contribution into tradable value.
In consumer scenarios, the AGXcelerator powers a new-generation home intelligent agent AI hub, integrating computing, storage, and whole-home intelligence. It offers a Lego-style heterogeneous paradigm that enables freedom across multiple chips, heterogeneous systems, and tokens, with three deployment modes: office agent hosts, personal AI supercomputers, and hybrid multimodal agent clusters. It also serves as a personal fun and work digital companion, covering multimodal interactions for both office and entertainment.
Additionally, the AGX X2 extends to edge applications such as AGX desktop supercomputing, smart manufacturing, array servers, intelligent cockpits, and embodied intelligence, delivering integrated inference solutions with end-edge-cloud coordination. This establishes a collaborative loop of "real-time edge inference, edge aggregation computing, and cloud training iteration," providing localized, low-latency, and privacy-compliant agent computing services for various industries.
The launch of the AGX X2 represents a key strategic step for CiXin Technology, which is building a market strategy centered on Agentic Compute. This includes three pillars: first, iterative development of high-performance Agentic SoCs, enhancing architecture, process upgrades, functional integration, and software ecosystems; second, leveraging T-NPU to create strategic leverage by coordinating SoC and NPU capabilities, through multi-vendor collaboration, dynamic compute scheduling, edge AI deployment, and model optimization to boost on-device AI and token generation; and third, token creation, where SoC-NPU synergy transforms computing processes into token generation and evolution, helping define a new computing currency across value quantification, incentive mechanisms, ecosystem circulation, and economic closure to build a self-sustaining digital economy.