Carrier Ethernet for the AI Era: Why Connectivity Must Evolve
Mar 3, 2026
Key Takeaways
Artificial intelligence is rapidly reshaping how organizations handle data. Yet behind every model trained, every inference made, and every new AI-driven business workflow underpins a critical foundation: the network. And today, that network must transform. AI workloads are no longer confined to centralized hyperscale data centers. They are increasingly distributed across multiple environments: On‑premises, cloud, colocation, telco edge, and enterprise edge, requiring performance‑assured connectivity across all domains.
This blog explores why AI demands new connectivity models and how Mplify (formerly MEF)’s new Carrier Ethernet for AI program, in which RAD is directly involved, is setting the industry standard for AI-ready transport.
What is the impact of distributed AI workloads on connectivity?
Organizations are deploying AI across a wide variety of environments, reflecting highly distributed operational requirements. According to a recent AI infrastructure survey by 451 Research, now part of S&P Global:
- 57% of responding enterprises perform model training on-premises versus 48% in public cloud.
- 53% deploy production AI models on-premises versus 47% in the cloud.
- More than half use managed service providers, and over a third use colocation facilities for AI workloads.
- 22% train AI models at the edge, and 27% run inference at the edge.
- 63% plan to continue using a hybrid mix of on-premises and public cloud for new AI/ML workloads.
This distributed approach means workloads, and the enormous datasets that feed them, must move across data centers, cloud regions, enterprise sites, and edge locations with predictability, consistency, and security.
How CSPs stand to benefit from AI-driven, new connectivity demand?
Historically, much of the AI connectivity conversation centered on hyperscaler data center interconnect. CSPs mostly provided dark fiber links between cloud data centers.
But this focus is shifting. It’s enterprise adoption of AI, not hyperscalers, that are expected to drive the newest network demand in the next several years.
As enterprises scale AI for automation, analytics, real time insights, and distributed processing, they will need a new generation of transport services between edge, data center, and the cloud.
This creates a powerful opportunity for CSPs to deliver premium connectivity services that support next-generation AI operations.
What are AI’s requirements for network performance?
AI workloads behave very differently from traditional enterprise applications. They are data-intensive, involving large training and inference datasets, latency sensitive, especially for real-time inference, distributed across many domains and access types, and involve different architectures, such as Point to Point, Point to Multi-point, Multipoint to Multipoint etc. Above all, they are performance-critical and expensive GPU clusters must be kept fully utilized at all times.
This places far more stringent demands on network performance than pre-AI services. In addition to high throughput (100G/400G/800G), the network needs to deliver extremely low and predictable end-to-end latency across multi-domain environments, with very low jitter and minimal packet loss or near-lossless transport.
Carrier Ethernet is uniquely positioned to meet these needs. It provides:
- SLA based performance assurance
- Operational visibility (OAM, PM)
- Hierarchical QoS and traffic management
- Multi-domain service consistency
- Standards-based interoperability
- Flexible automation at scale
As enterprises increasingly adopt hybrid and multi-site AI architectures, Carrier Ethernet becomes the trusted transport foundation capable of meeting both current and emerging AI connectivity requirements.
What is the Mplify Carrier Ethernet for AI program?
Mplify (formerly MEF) has introduced the Carrier Ethernet for AI program to define how networks must evolve to support AI and agentic AI workloads. The initiative establishes AI‑specific use cases, updates performance specifications, and creates new certifications to validate whether providers can meet AI’s far more demanding transport requirements, well beyond those of traditional MEF 3.0 services.
By modernizing long‑standing MEF performance criteria for the AI era, Mplify fills a critical gap: Although many vendors claim to be “AI‑ready,” there has been no standardized way to verify those claims, until now.
The program focuses on three essential AI connectivity scenarios:
- Subscriber-to-AI edge connectivity
- Edge-to-data-center interconnect
- Data-center-to-data-center transport
The first use case to be addressed under the program’s framework is DCI, given its central role in providing the high‑capacity, low‑latency links required for distributed AI architectures.
Ultimately, the program offers the industry’s first independent method for validating AI‑ready transport, ensuring Carrier Ethernet continues to support large‑scale, high‑performance AI deployments.
What is RAD’s role in evolving Carrier Ethernet for AI?
As AI workloads surge, Carrier Ethernet must evolve beyond static connectivity. RAD is helping lead this transformation, both through industry collaboration in the Mplify Alliance and by advancing its own portfolio to support AI traffic at scale.
A key example is the new ETX‑2i‑400G, a purpose‑built, AI‑era platform engineered for high‑capacity DCI, ultra‑low latency, advanced traffic management, and quantum‑safe security. It ensures GPU clusters stay fully utilized by keeping latency and jitter tightly controlled, while its SLA enforcement keeps AI’s bursty traffic from impacting other services.
RAD is also preparing carriers for the future of security with a crypto‑agile architecture that includes Post‑Quantum Cryptography (PQC), Quantum Key Distribution (QKD), and secure IPsec‑based access for SASE environments and remote GPU workloads.
Beyond transport, RAD brings elasticity and intelligence to the edge. Its programmable Carrier Ethernet devices support Ethernet On‑Demand, allowing bandwidth to scale dynamically for dataset ingestion, training bursts, or inference spikes.
RAD’s new ASIC family, purpose‑built for its Carrier Ethernet devices, is engineered to deliver higher capacity, intelligent buffering, real‑time telemetry, line‑rate MACsec and IPsec encryption, and programmable control to support adaptive, automated, AI‑driven traffic patterns. In addition, the ASICs integrate a processing unit and on‑chip memory designed for low‑latency, real‑time, on‑device AI inference, enabling a wide range of AI‑native networking use cases.
RAD devices generate rich telemetry and OAM data that is modeled in the operations layer and can be exposed through the Model Context Protocol (MCP). This provides AI agents with the structured context they need to automate root‑cause analysis, anticipate service impacts, and optimize the network in real time.
Together, these capabilities enable operators to create networks that sense, adapt, and respond to AI‑driven demands, ensuring high performance for both AI workloads and traditional services.
Summary
As AI workloads spread across data centers, cloud regions, and edge sites, enterprises will depend on CSPs more than ever to deliver high-capacity, predictable, and secure connectivity.
CSPs that are embracing AI-optimized Carrier Ethernet, high-capacity optical services, and Mplify’s new AI certification will be ideally positioned to:
- Differentiate with validated AI-ready performance
- Monetize premium connectivity services
- Support enterprise AI transformation
- Build trust in a rapidly evolving market
Mplify’s Carrier Ethernet for AI program provides the standards-based, independently validated framework needed to prove AI-ready transport, and as a long-standing Mplify member and a global leader in Carrier Ethernet solutions, RAD is contributing directly to the evolution of these standards while introducing new products and features that meet the networking requirements of the AI era.
Carrier Ethernet has been the backbone of business connectivity for more than 20 years. Now it is being reimagined for the future of AI-driven connectivity and RAD is proud to help lead the industry forward.