Quick Takeaways
- Autolink AI Link 3.0 Deep Fusion EEA marks a major step toward unified, scalable vehicle electronics.
- The architecture addresses computing fragmentation and bandwidth limits in software-defined vehicles.
On January 6, 2026, Autolink hosted a launch event at its CES 2026 exhibit to introduce AI Link 3.0, its next-generation Deep Fusion electronic and electrical vehicle architecture. The event included an opening address by founder and Chairman Yang Hongze and a presentation by Vice President and Chief Strategy Officer Li Zhigang. Technology partners AMD and ReinOCS also shared insights into their contributions to the Deep Fusion EEA.
Why next-generation vehicle architectures are needed
Yang Hongze emphasized that the automotive industry is rapidly advancing toward intelligent vehicles, electrification, and software-defined platforms. The growing adoption of cockpit domain controllers, integrated cockpit-driving controllers, and zonal controllers, combined with rising sensor counts and software complexity, is pushing traditional domain-based architectures to their limits.
He noted that existing systems suffer from fragmented computing resources and operational inefficiencies. Autolink AI Link 3.0 Deep Fusion EEA has been developed as a unified and scalable foundation to overcome these constraints and support the evolving requirements of intelligent vehicles.
Core elements of Autolink AI Link 3.0 Deep Fusion EEA
Li Zhigang outlined the essential capabilities required for future vehicle architectures, including real-time sensor fusion, high-bandwidth communication, cross-domain data sharing, ADAS redundancy, and shared compute resources. The Deep Fusion EEA enables real-time cross-domain data exchange through an ultra-low-latency network and dynamic coordination of computing power.
The architecture is built on three core technologies:
Central computing platform with Qualcomm Snapdragon SoC
The central computing platform in AI Link 3.0 is powered by the Qualcomm Snapdragon SA8797 SoC. It delivers more than 560 thousand DMIPS, features a GPU rated at 8.1 TFLOPS, and an NPU capable of 640 TOPS, enabling support for SAE Level 3 and Level 4 autonomous driving.
The platform can also run large language models with up to 13 billion parameters. It supports up to eight 8K displays and connectivity for as many as 20 cameras, addressing both cockpit and automated driving workloads within a single computing hub.
Zonal controllers featuring AMD Versal AI Edge Series Gen 2 SoCs
Wayne Lyons, Senior Director of Automotive Markets at AMD, highlighted the integration of second-generation Versal AI Edge Series ADAS domain controllers within AI Link 3.0 zonal controllers. These units are based on the AMD 3558 chipset, offering 200 thousand DMIPS on the A Core, 28 thousand DMIPS on the R Core, 268 GFLOPs of GPU performance, and 369 TOPS from the NPU.
Designed for ADAS applications, the chipset interfaces with cameras, radars, speakers, and body electronics modules. It provides power-efficient edge AI processing, supports future architectural scalability, and complies with automotive safety and security standards. The zonal controllers also enable redundancy for SAE Level 3 and Level 4 autonomous functions.
ReinOCS PCIe optical communication backbone network
Xuezhe Zheng, CTO of ReinOCS, addressed the growing challenge of bandwidth demand in modern vehicles. AI Link 3.0 employs a PCIe-over-fiber ring network to deliver a stable, efficient, and flexible high-speed interconnect. The network integrates RDMA technology, enabling synchronized, real-time decision-making across vehicle domains.
In this setup, the central computing platform supports up to 370 I/O interfaces, while each zonal controller can scale to 580 I/O interfaces. ReinOCS also showcased its FTTV-eDP optical module, supporting up to 32.4 Gbps at low power and cost, FTTV-MIPI modules enabling up to 100 Gbps for camera data transmission, and the FTTV-PCIe optical module supporting PCIe 4.0 over distances up to 30 meters with bandwidths reaching 64 Gbps and ultra-low latency.
Why next-generation vehicle architectures are needed
Yang Hongze emphasized that the automotive industry is rapidly advancing toward intelligent vehicles, electrification, and software-defined platforms. The growing adoption of cockpit domain controllers, integrated cockpit-driving controllers, and zonal controllers, combined with rising sensor counts and software complexity, is pushing traditional domain-based architectures to their limits.
He noted that existing systems suffer from fragmented computing resources and operational inefficiencies. Autolink AI Link 3.0 Deep Fusion EEA has been developed as a unified and scalable foundation to overcome these constraints and support the evolving requirements of intelligent vehicles.
Core elements of Autolink AI Link 3.0 Deep Fusion EEA
Li Zhigang outlined the essential capabilities required for future vehicle architectures, including real-time sensor fusion, high-bandwidth communication, cross-domain data sharing, ADAS redundancy, and shared compute resources. The Deep Fusion EEA enables real-time cross-domain data exchange through an ultra-low-latency network and dynamic coordination of computing power.
The architecture is built on three core technologies:
- A central computing platform
- Zonal controllers
- A PCIe optical communication backbone network
Central computing platform with Qualcomm Snapdragon SoC
The central computing platform in AI Link 3.0 is powered by the Qualcomm Snapdragon SA8797 SoC. It delivers more than 560 thousand DMIPS, features a GPU rated at 8.1 TFLOPS, and an NPU capable of 640 TOPS, enabling support for SAE Level 3 and Level 4 autonomous driving.
The platform can also run large language models with up to 13 billion parameters. It supports up to eight 8K displays and connectivity for as many as 20 cameras, addressing both cockpit and automated driving workloads within a single computing hub.
Zonal controllers featuring AMD Versal AI Edge Series Gen 2 SoCs
Wayne Lyons, Senior Director of Automotive Markets at AMD, highlighted the integration of second-generation Versal AI Edge Series ADAS domain controllers within AI Link 3.0 zonal controllers. These units are based on the AMD 3558 chipset, offering 200 thousand DMIPS on the A Core, 28 thousand DMIPS on the R Core, 268 GFLOPs of GPU performance, and 369 TOPS from the NPU.
Designed for ADAS applications, the chipset interfaces with cameras, radars, speakers, and body electronics modules. It provides power-efficient edge AI processing, supports future architectural scalability, and complies with automotive safety and security standards. The zonal controllers also enable redundancy for SAE Level 3 and Level 4 autonomous functions.
ReinOCS PCIe optical communication backbone network
Xuezhe Zheng, CTO of ReinOCS, addressed the growing challenge of bandwidth demand in modern vehicles. AI Link 3.0 employs a PCIe-over-fiber ring network to deliver a stable, efficient, and flexible high-speed interconnect. The network integrates RDMA technology, enabling synchronized, real-time decision-making across vehicle domains.
In this setup, the central computing platform supports up to 370 I/O interfaces, while each zonal controller can scale to 580 I/O interfaces. ReinOCS also showcased its FTTV-eDP optical module, supporting up to 32.4 Gbps at low power and cost, FTTV-MIPI modules enabling up to 100 Gbps for camera data transmission, and the FTTV-PCIe optical module supporting PCIe 4.0 over distances up to 30 meters with bandwidths reaching 64 Gbps and ultra-low latency.
Company Press Release
Click above to visit the official source.
Share: