The unique characteristics of AI workloads are fundamentally rewriting the design logic of data center interconnect technologies. Traditional network architectures were built for general-purpose computing scenarios and struggle to adapt to the special requirements of AI. Whether in terms of data transfer patterns, bandwidth scale, or latency sensitivity, AI scenarios far exceed the boundaries of traditional data center applications, pushing interconnect technologies to the very limits of their original designs.<\/p>\n
To understand these challenges, we first need to compare the core differences between AI workloads and traditional workloads:<\/p>\n
<\/p>\n
![\"Comparison](\"https:\/\/ascentoptics.com\/blog\/wp-content\/uploads\/2026\/01\/\u6838\u5fc3\u5dee\u5f02.png\")
<\/p>\n
<\/p>\n
This difference directly translates into the four major pain points of interconnect technologies:\u00a0How to support the high-density interactions of all-to-all communication?\u00a0How to cope with the non-linear growth of bandwidth demands?\u00a0How to compress latency down to sub-microsecond levels?\u00a0And how to sustain continuous high traffic pressure?\u00a0These are precisely the core directions that current AI data center interconnect technologies must break through.<\/p>\n
<\/p>\n
The Evolution Path of the Three Major Interconnect Technologies: Ethernet, InfiniBand, and Omni-Path<\/strong><\/h2>\nFaced with the challenges brought by AI, by 2025 the data center interconnect field has formed three mainstream technical routes: Ethernet, InfiniBand, and Omni-Path.<\/p>\n
Rooted in different technical DNA, and through continuous iteration and optimization, these three technologies have developed differentiated competitive advantages in compatibility, performance, and cost respectively.<\/p>\n
<\/p>\n
\n- \n
Ethernet: From General-Purpose to AI-Optimized, Breaking Through to Terabit-Class Speeds<\/strong><\/h3>\n<\/li>\n<\/ol>\nAs a mature standard with decades of history, Ethernet has long dominated enterprise data centers thanks to its strong compatibility, controllable costs, and excellent scalability.<\/p>\n
However, in AI scenarios, traditional Ethernet reveals clear weaknesses: under high traffic loads it is prone to increased latency and packet loss, making it difficult to meet the stringent stability requirements of all-to-all communication patterns.<\/p>\n
To adapt to AI, Ethernet has achieved a complete transformation through two key technological upgrades:<\/p>\n
<\/p>\n
(1) IEEE 802.3df-2024: 800GbE Establishes the Foundation for Next-Generation AI Clusters<\/strong><\/p>\nThe IEEE 802.3df-2024 standard, released in February 2024, can be regarded as a watershed moment for AI data center interconnects. This 800GbE specification not only delivers massive bandwidth capacity, but also introduces an 8-lane parallel architecture that enables highly flexible port breakout configurations. A single 800GbE port can be flexibly split \u2014 depending on workload requirements \u2014 into: 2 \u00d7 400GbE, 4 \u00d7 200GbE or 8 \u00d7 100GbE.<\/p>\n
This capability perfectly matches the hybrid traffic patterns typical in AI training: some accelerators require extremely high-bandwidth peer-to-peer interaction, while others only need lower-bandwidth synchronization.<\/p>\n
<\/p>\n
(2) UEC 1.0: Ethernet\u2019s Dedicated AI Optimization Solution<\/strong><\/p>\nIn 2025, the Ultra Ethernet Consortium (UEC) \u2014 jointly initiated by industry heavyweights \u2014 released the UEC 1.0 specification. This represents the most aggressive optimization effort ever made to adapt Ethernet specifically for AI workloads. The specification directly targets InfiniBand\u2019s performance advantages through three core technical innovations:<\/p>\n
Modern RDMA Deployment<\/strong><\/p>\nIntroduces RDMA (Remote Direct Memory Access) technology, enabling GPUs to directly access the memory of other devices without CPU involvement, dramatically reducing latency.<\/p>\n
Link-Level Retry (LLR)<\/strong><\/p>\nAddresses Ethernet\u2019s long-standing historical weakness of packet loss. Instead of relying on traditional Priority-based Flow Control (PFC), LLR performs loss detection and retransmission directly at the link layer \u2014 eliminating the high recovery cost that would otherwise fall on higher-layer protocols.<\/p>\n
Packet Rate Improvement (PRI)<\/strong><\/p>\nReduces protocol overhead through header compression while adding network probe functionality for real-time congestion visibility. This allows administrators to dynamically adjust traffic distribution based on actual conditions.<\/p>\n
In addition, UEC 1.0 introduces support for switch-side packet spraying + NIC-side reordering \u2014 a mechanism previously seen only in proprietary systems. This capability now enables standard Ethernet to effectively handle the intense all-to-all communication pressure characteristic of large-scale AI training workloads.<\/p>\n
<\/p>\n
2. InfiniBand: From High Performance to 800Gb\/s, Solidifying Its Low-Latency Advantage<\/strong><\/h3>\nInfiniBand<\/a> was born in the late 1990s. From the very beginning of its design, it targeted high-speed server-to-server communication within data centers. Unlike Ethernet, which evolved from local area network (LAN) origins, InfiniBand was purpose-built from the ground up to meet the stringent demands of cluster computing.<\/p>\nIts core strengths lie in lossless transmission and ultra-low latency. Through hardware-level flow control and dedicated network adapters, InfiniBand directly addresses one of the most critical pain points in AI training: the cascading failures and stalls caused by packet loss.<\/p>\n
<\/p>\n
(1) Credit-Based Flow Control: The Performance Cornerstone of InfiniBand<\/strong><\/h3>\nThe fundamental difference between InfiniBand and Ethernet lies in InfiniBand\u2019s credit-based flow control mechanism. Before transmitting packets, the sender first verifies that the receiver has sufficient buffer space, preventing packet loss at the source. This design is critical for AI training: in large-scale training clusters involving thousands of accelerators, the loss of even a single packet can force an entire batch to be recomputed. Credit-based flow control effectively eliminates this risk.<\/p>\n
(2) XDR Evolution: Sustaining Low Latency at 800 Gb\/s Bandwidth<\/strong><\/h3>\nIn October 2023, the InfiniBand Trade Association (IBTA) released the InfiniBand Specification Version 1.7, ushering InfiniBand into the XDR (Extended Data Rate) era. This upgrade increases single-port bandwidth to 800 Gb\/s, while inter-switch link speeds reach up to 1.6 Tb\/s<\/a>, enabled by 200 Gb\/s per-lane SerDes technology.<\/p>\n <\/p>\n
3. Omni-Path: From Dormancy to Revival, Targeting Cost-Sensitive AI Scenarios<\/strong><\/h3>\nOmni-Path has had a notably dramatic journey. Originally introduced by Intel in the mid-2010s, it was designed to challenge NVIDIA InfiniBand\u2019s dominance in the HPC market. With features such as adaptive routing, integrated fabric management, and competitive performance, Omni-Path once attracted significant industry attention. However, in 2019, Intel announced the discontinuation of the Omni-Path project to refocus on its core processor business, and the technology subsequently fell into dormancy.<\/p>\n
In 2020, Omni-Path saw a turning point when the original Intel Omni-Path engineering team spun off to form Cornelis Networks. The company revived development of the technology and launched the CN5000 series, positioning it as a cost-competitive AI interconnect solution.<\/p>\n
<\/p>\n
Ethernet: From General-Purpose to AI-Optimized, Breaking Through to Terabit-Class Speeds<\/strong><\/h3>\n<\/li>\n<\/ol>\nAs a mature standard with decades of history, Ethernet has long dominated enterprise data centers thanks to its strong compatibility, controllable costs, and excellent scalability.<\/p>\n
However, in AI scenarios, traditional Ethernet reveals clear weaknesses: under high traffic loads it is prone to increased latency and packet loss, making it difficult to meet the stringent stability requirements of all-to-all communication patterns.<\/p>\n
To adapt to AI, Ethernet has achieved a complete transformation through two key technological upgrades:<\/p>\n
<\/p>\n
(1) IEEE 802.3df-2024: 800GbE Establishes the Foundation for Next-Generation AI Clusters<\/strong><\/p>\nThe IEEE 802.3df-2024 standard, released in February 2024, can be regarded as a watershed moment for AI data center interconnects. This 800GbE specification not only delivers massive bandwidth capacity, but also introduces an 8-lane parallel architecture that enables highly flexible port breakout configurations. A single 800GbE port can be flexibly split \u2014 depending on workload requirements \u2014 into: 2 \u00d7 400GbE, 4 \u00d7 200GbE or 8 \u00d7 100GbE.<\/p>\n
This capability perfectly matches the hybrid traffic patterns typical in AI training: some accelerators require extremely high-bandwidth peer-to-peer interaction, while others only need lower-bandwidth synchronization.<\/p>\n
<\/p>\n
(2) UEC 1.0: Ethernet\u2019s Dedicated AI Optimization Solution<\/strong><\/p>\nIn 2025, the Ultra Ethernet Consortium (UEC) \u2014 jointly initiated by industry heavyweights \u2014 released the UEC 1.0 specification. This represents the most aggressive optimization effort ever made to adapt Ethernet specifically for AI workloads. The specification directly targets InfiniBand\u2019s performance advantages through three core technical innovations:<\/p>\n
Modern RDMA Deployment<\/strong><\/p>\nIntroduces RDMA (Remote Direct Memory Access) technology, enabling GPUs to directly access the memory of other devices without CPU involvement, dramatically reducing latency.<\/p>\n
Link-Level Retry (LLR)<\/strong><\/p>\nAddresses Ethernet\u2019s long-standing historical weakness of packet loss. Instead of relying on traditional Priority-based Flow Control (PFC), LLR performs loss detection and retransmission directly at the link layer \u2014 eliminating the high recovery cost that would otherwise fall on higher-layer protocols.<\/p>\n
Packet Rate Improvement (PRI)<\/strong><\/p>\nReduces protocol overhead through header compression while adding network probe functionality for real-time congestion visibility. This allows administrators to dynamically adjust traffic distribution based on actual conditions.<\/p>\n
In addition, UEC 1.0 introduces support for switch-side packet spraying + NIC-side reordering \u2014 a mechanism previously seen only in proprietary systems. This capability now enables standard Ethernet to effectively handle the intense all-to-all communication pressure characteristic of large-scale AI training workloads.<\/p>\n
<\/p>\n
2. InfiniBand: From High Performance to 800Gb\/s, Solidifying Its Low-Latency Advantage<\/strong><\/h3>\nInfiniBand<\/a> was born in the late 1990s. From the very beginning of its design, it targeted high-speed server-to-server communication within data centers. Unlike Ethernet, which evolved from local area network (LAN) origins, InfiniBand was purpose-built from the ground up to meet the stringent demands of cluster computing.<\/p>\nIts core strengths lie in lossless transmission and ultra-low latency. Through hardware-level flow control and dedicated network adapters, InfiniBand directly addresses one of the most critical pain points in AI training: the cascading failures and stalls caused by packet loss.<\/p>\n
<\/p>\n
(1) Credit-Based Flow Control: The Performance Cornerstone of InfiniBand<\/strong><\/h3>\nThe fundamental difference between InfiniBand and Ethernet lies in InfiniBand\u2019s credit-based flow control mechanism. Before transmitting packets, the sender first verifies that the receiver has sufficient buffer space, preventing packet loss at the source. This design is critical for AI training: in large-scale training clusters involving thousands of accelerators, the loss of even a single packet can force an entire batch to be recomputed. Credit-based flow control effectively eliminates this risk.<\/p>\n
(2) XDR Evolution: Sustaining Low Latency at 800 Gb\/s Bandwidth<\/strong><\/h3>\nIn October 2023, the InfiniBand Trade Association (IBTA) released the InfiniBand Specification Version 1.7, ushering InfiniBand into the XDR (Extended Data Rate) era. This upgrade increases single-port bandwidth to 800 Gb\/s, while inter-switch link speeds reach up to 1.6 Tb\/s<\/a>, enabled by 200 Gb\/s per-lane SerDes technology.<\/p>\n <\/p>\n
3. Omni-Path: From Dormancy to Revival, Targeting Cost-Sensitive AI Scenarios<\/strong><\/h3>\nOmni-Path has had a notably dramatic journey. Originally introduced by Intel in the mid-2010s, it was designed to challenge NVIDIA InfiniBand\u2019s dominance in the HPC market. With features such as adaptive routing, integrated fabric management, and competitive performance, Omni-Path once attracted significant industry attention. However, in 2019, Intel announced the discontinuation of the Omni-Path project to refocus on its core processor business, and the technology subsequently fell into dormancy.<\/p>\n
In 2020, Omni-Path saw a turning point when the original Intel Omni-Path engineering team spun off to form Cornelis Networks. The company revived development of the technology and launched the CN5000 series, positioning it as a cost-competitive AI interconnect solution.<\/p>\n
<\/p>\n
The IEEE 802.3df-2024 standard, released in February 2024, can be regarded as a watershed moment for AI data center interconnects. This 800GbE specification not only delivers massive bandwidth capacity, but also introduces an 8-lane parallel architecture that enables highly flexible port breakout configurations. A single 800GbE port can be flexibly split \u2014 depending on workload requirements \u2014 into: 2 \u00d7 400GbE, 4 \u00d7 200GbE or 8 \u00d7 100GbE.<\/p>\n
This capability perfectly matches the hybrid traffic patterns typical in AI training: some accelerators require extremely high-bandwidth peer-to-peer interaction, while others only need lower-bandwidth synchronization.<\/p>\n
<\/p>\n
(2) UEC 1.0: Ethernet\u2019s Dedicated AI Optimization Solution<\/strong><\/p>\n In 2025, the Ultra Ethernet Consortium (UEC) \u2014 jointly initiated by industry heavyweights \u2014 released the UEC 1.0 specification. This represents the most aggressive optimization effort ever made to adapt Ethernet specifically for AI workloads. The specification directly targets InfiniBand\u2019s performance advantages through three core technical innovations:<\/p>\n Modern RDMA Deployment<\/strong><\/p>\n Introduces RDMA (Remote Direct Memory Access) technology, enabling GPUs to directly access the memory of other devices without CPU involvement, dramatically reducing latency.<\/p>\n Link-Level Retry (LLR)<\/strong><\/p>\n Addresses Ethernet\u2019s long-standing historical weakness of packet loss. Instead of relying on traditional Priority-based Flow Control (PFC), LLR performs loss detection and retransmission directly at the link layer \u2014 eliminating the high recovery cost that would otherwise fall on higher-layer protocols.<\/p>\n Packet Rate Improvement (PRI)<\/strong><\/p>\n Reduces protocol overhead through header compression while adding network probe functionality for real-time congestion visibility. This allows administrators to dynamically adjust traffic distribution based on actual conditions.<\/p>\n In addition, UEC 1.0 introduces support for switch-side packet spraying + NIC-side reordering \u2014 a mechanism previously seen only in proprietary systems. This capability now enables standard Ethernet to effectively handle the intense all-to-all communication pressure characteristic of large-scale AI training workloads.<\/p>\n <\/p>\n InfiniBand<\/a> was born in the late 1990s. From the very beginning of its design, it targeted high-speed server-to-server communication within data centers. Unlike Ethernet, which evolved from local area network (LAN) origins, InfiniBand was purpose-built from the ground up to meet the stringent demands of cluster computing.<\/p>\n Its core strengths lie in lossless transmission and ultra-low latency. Through hardware-level flow control and dedicated network adapters, InfiniBand directly addresses one of the most critical pain points in AI training: the cascading failures and stalls caused by packet loss.<\/p>\n <\/p>\n The fundamental difference between InfiniBand and Ethernet lies in InfiniBand\u2019s credit-based flow control mechanism. Before transmitting packets, the sender first verifies that the receiver has sufficient buffer space, preventing packet loss at the source. This design is critical for AI training: in large-scale training clusters involving thousands of accelerators, the loss of even a single packet can force an entire batch to be recomputed. Credit-based flow control effectively eliminates this risk.<\/p>\n In October 2023, the InfiniBand Trade Association (IBTA) released the InfiniBand Specification Version 1.7, ushering InfiniBand into the XDR (Extended Data Rate) era. This upgrade increases single-port bandwidth to 800 Gb\/s, while inter-switch link speeds reach up to 1.6 Tb\/s<\/a>, enabled by 200 Gb\/s per-lane SerDes technology.<\/p>\n <\/p>\n Omni-Path has had a notably dramatic journey. Originally introduced by Intel in the mid-2010s, it was designed to challenge NVIDIA InfiniBand\u2019s dominance in the HPC market. With features such as adaptive routing, integrated fabric management, and competitive performance, Omni-Path once attracted significant industry attention. However, in 2019, Intel announced the discontinuation of the Omni-Path project to refocus on its core processor business, and the technology subsequently fell into dormancy.<\/p>\n In 2020, Omni-Path saw a turning point when the original Intel Omni-Path engineering team spun off to form Cornelis Networks. The company revived development of the technology and launched the CN5000 series, positioning it as a cost-competitive AI interconnect solution.<\/p>\n <\/p>\n2. InfiniBand: From High Performance to 800Gb\/s, Solidifying Its Low-Latency Advantage<\/strong><\/h3>\n
(1) Credit-Based Flow Control: The Performance Cornerstone of InfiniBand<\/strong><\/h3>\n
(2) XDR Evolution: Sustaining Low Latency at 800 Gb\/s Bandwidth<\/strong><\/h3>\n
3. Omni-Path: From Dormancy to Revival, Targeting Cost-Sensitive AI Scenarios<\/strong><\/h3>\n
![\"Three](\"https:\/\/ascentoptics.com\/blog\/wp-content\/uploads\/2026\/01\/AI-DCI\u5bf9\u6bd4.png\")
<\/p>\n