DDC Evolution - Agile Network for the AI Era

The emergence of Distributed Disaggregated Chassis (DDC) technology signifies a major shift in network architectures amid ongoing digital transformation. Introduced in 2019 for Data Center Interconnect (DCI), DDC has evolved significantly, becoming standardized in 2024 and applicable in diverse scenarios. This five-year journey from concept to industry implementation heralds a new era of disaggregated network design and operation.

2019 - 2020: Backbone Network Technology (DCI)

In 2019, the rapid growth of data center network (DCN) traffic drove an urgent need for upgrades in data center interconnect (DCI) networks. Traditional chassis-based routers faced limitations in scalability due to fixed chassis sizes, high power consumption, and costly upgrades to meet power and cooling requirements. To address these challenges, AT&T submitted a specification for a merchant silicon-based router to the Open Compute Project (OCP) and introduced the concept of DDC.

DDC technology disaggregates​​ traditional high-capacity routers into multiple box-based devices, leveraging cell switching and distributed architectures to enhance flexibility and scalability. The core idea is to break down monolithic chassis-based systems​ into smaller, modular box-based units to replace line cards and switch fabrics of chassis, interconnected via cables and managed by a centralized or distributed Network Operating System (NOS).

Replacing chassis-based routers with box routers

By 2020, DriveNets' DDC devices were successfully deployed in AT&T's backbone network. By the end of 2022, over 30% of AT&T's core backbone traffic was running on customized Broadcom StrataDNX white-box routers (powered by DriveNets' software). DDC technology proved to be a "Transformer," demonstrating its capabilities in backbone networks and enabling efficient operations.

2023 – 2024: Data Center Interconnect (DCN)

With the successful application of DDC in backbone networks (DCI), its flexibility and scalability gained attention. Meanwhile, the rapid development of big data, cloud computing, and AI technologies led to surging data center traffic, exposing the limitations of traditional network architectures in terms of scalability and power efficiency.

Traditional data center core switches were often closed, centralized chassis-based systems, integrating routing engines, switching boards, and interface cards into a single large enclosure. While easy to manage, they struggled with scalability, flexibility, and cost-effectiveness. As switching capacities evolved from 100G to 400G, power consumption surged—for example, a 16-slot, fully loaded 400G switch could consume 40,000–50,000 watts. This posed significant challenges for legacy data centers in terms of power supply, cooling, and rack deployment.

In this context, DDC technology expanded into data center networks (DCN). It decomposed traditional large switches into smaller, more flexible components. By physically separating the data plane, control plane, and other functional components across different hardware devices interconnected via high-speed networks, DDC enabled independent scaling, offering flexibility, scalability, and cost efficiency. Moreover, components could be distributed across multiple racks, optimizing cooling and power management while overcoming space constraints and facilitating larger-scale networking.

Decomposing traditional large switches

2024 - 2025: Standardization of Next-Gen DDC

In 2025, H3C collaborated with industry partners to establish the DDC core framework standard based on OSF (Open Scheduling Framework for AI Networks). This standardization provided comprehensive guidance—from requirements and frameworks to technical solutions—facilitating ecosystem growth and adoption. The standardized design broke vendor lock-in, enabling seamless interoperability between hardware from different brands and truly achieving "hardware-defined freedom." In April, H3C launched its next-gen DDC switch, the S12500AI series, marking the maturity and commercialization of the technology.

Contributing to DDC technology standardization

In AI training scenarios, next-gen DDC can scale network capacity by adding NCFs (Network Connectivity Fabric) and NCPs (Network Connectivity Processor). This adaptability enables DDC to support networks of all sizes, from small to hyperscale.

In terms of architecture, DDC adopts a design with independent NCF and NCP network elements. NCFs handle cell forwarding tables, while NCPs manage VoQ tables, with self-negotiation and self-networking achieved via BGP EVPN protocols. This approach simplifies network management and eliminates single points of failure. DDC entirely abandons centralized network control units in favor of a distributed control plane, further enhancing reliability and scalability. It is particularly suited for large-scale AI training and inference, supporting ultra-large-scale networking, with multi-cluster solutions scaling up to 73,728 ports.

Conclusion

DDC technology has systematically addressed key bottlenecks in traditional networks—scalability, energy efficiency, and operational complexity—while achieving paradigm-shifting innovation through standardized architectures. As a core driver of this transformation, H3C has embedded the philosophy of "Diversified Dynamic-Connectivity" (DDC) into its next-gen DDC product architecture.

On the "Diversified" front, H3C innovatively integrates cell-switching technology with Ethernet protocols, enabling compatibility with diverse computing resources (supporting multi-brand GPUs and NICs) and heterogeneous network devices, fostering an open intelligent network ecosystem.

On the "Dynamic" front, intelligent scheduling algorithms enable microsecond-level real-time path optimization and zero-congestion networking, supporting elastic scaling from small clusters to hyperscale deployments (with up to 70,000+ 400G ports). This compatibility and flexibility not only meet diverse computing needs but also provide a robust and efficient network foundation for AI data centers.

As standardization accelerates, next-gen DDC technology—with its open ecosystem—will continue to unleash its potential, leading network architectures into a new era of full disaggregation.