Modern AI Requires NaaS: The Future of Distributed Networking

The days of running AI operations in a single data center or centralized cloud facility are gone. Modern AI applications must make use of highly distributed services and capabilities. Bringing these resources seamlessly together requires a new approach to networking. Enter NaaS, network-as-a-service, which is (as the name implies) a cloud services model where providers deliver networking capabilities, including connectivity, security, and management, to enterprises on demand and typically via a subscription.

A quick look at how modern AI applications work illustrates the need for NaaS. To start, data used to train AI models is typically stored and generated in different places. Compute resources for training models and running inference are often dispersed, especially with the rise of new offerings like GPUs-as-a-Service and Neoclouds. The need for fast analysis and actionable insights in real-time applications means that data processing and inference must be done close to where the data is generated or the action (e.g., an autonomous car maneuvering around a road hazard) takes place.

Noting these factors, the convergence of AI and networking got considerable attention this year. Mplify (formerly MEF), the organization that has been driving the NaaS market for years, brought AI into the fold. “We have moved into the AI role,” said Pascal Menezes, CTO of Mplify. He noted that with AI agents and models of all sorts everywhere, networks have become very critical.

See also: Why Real-Time AI Needs Distributed Cloud Compute at the Edge

Enter NaaS

Today, enterprises and the providers delivering services to support AI efforts essentially need AI connectivity as a service. That’s where NaaS can play a critical role.

NaaS is a cloud-like consumption model for networking. Instead of enterprises owning, configuring, and managing their own networking hardware and software, they can consume connectivity, bandwidth, security, and optimization features on demand from a provider.

Key aspects include:

• On-demand and subscription-based: Similar to SaaS or IaaS, enterprises can scale up or down based on usage.
• Policy-driven automation: Networking functions are exposed as APIs and services rather than manual configuration.
• Abstracted infrastructure: Enterprises do not need to worry about physical links, routing, or vendor-specific hardware.

See also: What Are Neoclouds and Why Does AI Need Them?

Mplify’s Role in NaaS

Mplify has a long history of bringing connectivity technologies to market. Its focus on frameworks, lifecycle service orchestration (LSO) APIs, and certifications helped make Carrier Ethernet a pervasive offering that enterprises relied on for decades.

The group has followed a similar path with NaaS. It has developed LSO APIs and frameworks that allow service providers to expose network services, such as SD-WAN, Secure Access Service Edge (SASE), and more, in a standardized way.

With Mplify’s standards, enterprises can order, activate, and monitor network services across multiple operators as if they were dealing with a single cloud provider. Additionally, providers can certify their offerings against Mplify’s standards, which increases trust and accelerates adoption.

Why NaaS is Important for AI

AI, edge computing, and GPU acceleration are highly distributed and resource-intensive. They require flexible connectivity. AI workloads may burst between data centers, clouds, and edge sites. NaaS lets enterprises and providers spin up high-bandwidth links only when needed. To that end, NaaS supports fluctuating requirements based on predictable and unforeseen changes in business demand.

Modern AI applications also need low-latency and assured Quality of Service (QoS). That is another area where NaaS can play a role. For example, inference at the edge (e.g., in manufacturing or autonomous systems) needs deterministic, reliable networking. NaaS allows policy-based performance guarantees. To this point, NaaS providers generally offer service level agreements (SLAs) that guarantee specific levels of performance, reliability, and availability.

One additional aspect of NaaS is that it tightly integrates security. That is important because, with distributed GPUs and edge nodes, the attack surface expands. NaaS integrates security as part of the service fabric. Most NaaS offerings support SASE, which combines wide-area networking (WAN) connectivity capabilities with network security functions, such as secure web gateways, zero-trust network access (ZTNA), firewall-as-a-service (FWaaS), and cloud access security broker (CASB).

Bottom Line

Enterprises are used to compute-as-a-service offerings. However, as they expand their AI initiatives and utilize new offerings like GPUs-as-a-Service, the compute capabilities quickly overwhelm the underlying network. That creates choke points and introduces performance problems.

NaaS helps address these issues by offering equally elastic, high-bandwidth interconnects. Such services ensure enterprises and providers can move large AI datasets and run distributed models without introducing latency and other delays.