Token economy heats up as Xingwang Power targets AI Factory intelligent interconnect | Yunqi Partners

In the AI Factory era, the race for compute is shifting from "stacking GPUs" to "producing tokens efficiently." As demand for large model training and inference continues to grow, network interconnect, data processing, and heterogeneous compute coordination are becoming critical factors in overall AI infrastructure efficiency.

Recently, Yunqi Capital Pre-A round lead portfolio company "Xinwang Power" (芯网动力) made its debut at the 2026 Global Computing Conference, showcasing its three major product lines — sNIC smart NICs, DPUs, and IO-IP/Die — along with intelligent interconnect solutions for cloud-edge-device scenarios. In this edition of "Yunqi Partners," we explore how Xinwang Power is positioning itself in AI Factory's underlying interconnect technology and the new opportunities in compute efficiency behind the token economy.

In October, the global AI compute industry's attention once again turned to underlying infrastructure.

At the recently concluded NVIDIA GTC Berlin 2026, AI Factory emerged as one of the key topics. From high-speed network interconnect and compute resource scheduling to token production efficiency, system-level optimization of AI infrastructure received increasing attention.

Almost concurrently, the 2026 Global Computing Conference (CGC2026) took place in Shenzhen from October 21–24. With the theme "Megawatt Foundations, Super-Nodes Unleashing Surging Compute Power," the conference focused on super-nodes, megawatt-scale intelligent computing infrastructure, and full-stack technology coordination.

Both conferences pointed to an important trend in the AI compute industry: as demand for large model training and inference continues to grow, the competition for compute is moving from single-point hardware performance toward system-level coordination across computing, networking, storage, and security.

Against this backdrop, Yunqi Capital Pre-A round lead portfolio company "Xinwang Power" made its first appearance at the 2026 Global Computing Conference, showcasing its technical accumulation in intelligent interconnect, three major product systems, and four scenario-based solutions. The company was also invited to participate in the Super-Node Technology and Ecosystem Forum and the Computing Security Ecosystem Forum to share its technical explorations for next-generation AI infrastructure.

From GPU Scale to Token Output

AI Factory Is Reshaping Compute Efficiency

As the AI industry shifts from compute resource construction to large-scale token production, the way we evaluate infrastructure efficiency is also changing.

Large-scale GPU clusters require continuous data exchange, task coordination, and compute scheduling. As cluster scale expands, network bandwidth, communication latency, data transfer efficiency, and resource utilization all directly impact overall system performance.

Particularly as AI applications move toward large-scale inference, improving effective compute utilization and reducing token production and delivery costs have become persistent engineering challenges for AI infrastructure.

At this conference, Xinwang Power shared around the theme "Interconnect Enabling Intelligent Compute, Efficient Token Supply," exploring the technical path for AI infrastructure to evolve from resource stacking to efficient coordination.

In the company's technical layout, intelligent interconnect serves as a foundational capability connecting compute, data, and applications. Through the coordination of network interconnect technology and heterogeneous computing architectures, it is exploring more efficient and trustworthy AI compute delivery methods — a direction of sustained investment.

Three Major Product Systems

Building Cloud-Edge-Device Intelligent Interconnect Capabilities

Centered on the evolution of AI compute infrastructure, Xinwang Power has developed an intelligent interconnect product portfolio covering cloud-edge-device scenarios through sustained R&D.

At this conference, Xinwang Power highlighted three major product systems:

• sNIC Smart NICs: Based on self-developed core technology, addressing high-speed data transfer needs in AI computing scenarios. Through RDMA and other network technologies, it explores low-latency, high-throughput intelligent interconnect capabilities, providing underlying support for intelligent computing networks.
• DPU: Focused on data processing, resource management, and infrastructure security. Around trusted computing, resource isolation, and data protection requirements, it provides corresponding hardware and technical support for AI infrastructure.
• IO-IP/Die: Providing IO-IP/Die products and solutions. Through integration with compute chips, it meets the high-speed interconnect and data exchange needs of diverse computing scenarios.

These three products form synergy from network connectivity, data processing and security, to chip-level interconnect capabilities, offering multi-layered technical options for AI computing systems of different scales and deployment environments.

For AI Scale Infra

Exploring Four Scenario-Based Solutions

Building on underlying technology and product systems, Xinwang Power further showcased four solutions for different AI computing scenarios.

① Compute-Network Coordinated Super-Intelligent Interconnect Solution

Addressing interconnect needs for large-scale compute clusters, supporting Scale-Up, Scale-Out, and other network architectures. Through high-speed data transfer and compute-network coordination, it explores technical paths to improve cluster interconnect efficiency and compute utilization.

② Trusted AI Scale Infra Solution

For trusted computing and security needs in AI infrastructure. Based on DPU and other underlying technologies, it provides support in resource isolation, trusted execution, and data protection, exploring AI infrastructure architectures that balance compute efficiency with system security.

③ Edge Heterogeneous Super-Node Interconnect Solution

For edge-side heterogeneous computing and multi-device coordination needs. Combining memory and message semantic coordination, GPU Direct, and other technical capabilities, it explores more efficient interconnect and resource coordination methods between edge computing nodes.

④ AI Terminal Network Solution

For the diverse performance, power consumption, form factor, and cost requirements of AI terminals. It provides small-form-factor, low-power smart NICs and IO-IP products, supporting the extension of AI computing capabilities to more terminal scenarios.

From cloud-scale intelligent computing clusters to edge heterogeneous computing and AI terminals, Xinwang Power is advancing the productization and application exploration of underlying technology around interconnect needs for different scenarios.

The Next Chapter for AI Infra

System Coordination Remains Key

As AI model capabilities continue to evolve, inference demand grows, and intelligent computing extends to edge and terminal devices, underlying infrastructure is undergoing a new round of architectural upgrades.

From core compute chips like GPUs to high-speed networking, data processing, security, and chip-level interconnect technology, the coordination capability across each link will collectively impact AI system performance, efficiency, and reliability.

For continuously evolving AI infrastructure, the importance of intelligent interconnect will only increase.

Yunqi Capital has long paid attention to innovation and development in AI and underlying computing technology. We look forward to Xinwang Power continuing to advance R&D and product deployment of core technologies including sNIC, DPU, and IO-IP/Die, providing technical support for more efficient and trustworthy compute infrastructure in the AI era.

Content adapted from "Xinwang Power"