The anticipated wave of agentic AI is set to push the limits of what traditional network infrastructure is capable of. Previous AI technologies have been well catered for with static networks and downlink-heavy, non-time-sensitive ML or generative AI applications, running on top of traditional clouds and designed largely for consumption by human users. Agentic AI changes all of this, with a potentially exponential rise in intelligent, proactive machine-to-machine communication. This will demand not only low latency but also high agility and scalability, and simple and automated pay-as-you-use consumption models. And it will result in increased traffic on the uplink. For real-time use cases across industries – from the factory floor to the boardroom, from financial services to the healthcare sector – outdated network architectures are causing headaches and preventing success of fledgling AI agent projects.
Growth of AI agents in companies
According to a study by London-based market research company 3Gem Research, in 2025 almost three quarters (74%) of companies in the US, the UK, Germany, and Australia have either implemented their first AI agents or plan to do so in the next 12 months. These companies are currently operating on average more than 30 agents, with plans to double that number in the coming year. The Boston Consulting Group forecasts market growth of 45% CAGR by 2030. This development requires network connectivity that enables agents to continuously interact – with each other, with databases and systems, with autonomous machines and production plants, with the Internet, and also with humans. It therefore needs to learn, manage, advise, and enrich – all in real time.
However, care must be taken in the investment in agentic AI. The Gartner Hype Cycle 2025 currently places AI agents at the “Peak of Inflated Expectations”, forecasting true and lasting productivity to be achieved in two to five years. Key to this productivity will be adapting network design to the needs of the new era of agile machine-to-machine communication. According to one report, 44% of organizations cite IT infrastructure constraints as their greatest roadblock for expanding AI initiatives, while 3Gem Research found that nearly a quarter (24%) of companies are held back by legacy technology in their endeavors to implement AI agents. More than a fifth (22%) of companies are hampered by a lack of integration and a further 21% by their lack of technology/workflow capabilities.
The bottleneck for AI success: outdated infrastructure
Enterprises are currently burdened with excessive spending on bandwidth, provisioning, and operational network expenses, while struggling with slow, complex provisioning and managing connections across different providers or platforms. Often, existing network solutions cannot scale or adapt to changing business needs quickly enough. Agility is quickly becoming key.
Software definition, automation, and the ability to scale up and down according to current demands are required to manage the complexities that come with AI integration from the cloud or from dedicated data centers. Companies require tailored and seamless, reliable, and easy to consume connectivity across their entire ecosystem. AI demands access to multiple data sources, integration of multi-cloud and hybrid cloud environments, the sharing of data and insights within value chains, and low latency inference wherever it is needed.
In contrast, traditional network architectures are characterized by slow provisioning, inflexible architectures, and costly investment. A static network architecture undermines AI development – and here, Network as a Service (NaaS), including the use-case Interconnection as a Service, provides the answer.
Interconnection as a Service – Future-proofing network design for AI demands
Network as a Service provides on-demand and agile connectivity, easily consumable, programmable, and tailored to individual needs. Combined with the optimized pathways and low latency of programmable Interconnection as a Service, NaaS is a powerful tool for enterprise network architects. It is characterized by flexibility in terms of bandwidth, routes, destinations, network interconnections, and quality of service, and it can be integrated into software stacks (AI or commodity) that automatically use it the way the respective software needs it. As such, it is able to respond to the current needs of applications and AI agents. And enterprise software can use the network autonomously as a simple service. As well as being on-demand, NaaSenables external monitoring of network performance/SLAs and adjustment of services, secure service consumption, and interoperability across multiple systems. It enables the cloudification of networks, while keeping costs down in a usage-based subscription model.
Interconnection as a Service takes the principles of traditional interconnection – direct, private, secure connectivity between networks, clouds, and partners – and makes them on-demand, flexible, and software-driven. This enables the seamless interconnection of stakeholders and digital assets along the AI value chain. Take multi-modal AI agents as an example: these are able to access diverse models as required, communicate with other systems and databases, and interact with human and non-human end users using cloud and AI routing technology. Interconnection as a Service allows companies to extend their network footprint to the interconnection platform, allowing for the integration of multi cloud and hybrid cloud environments, virtual access to networks and data infrastructure in data centers, and remote access to networks in other cities, using the provider’s backbone.
The route to success on your AI roadmap
The Network as a Service market size is currently valued at $33 billion, with projections of close to 30% CAGR up to 2030, rising to $115 billion. And it’s easy to see why: Instead of investing in physical infrastructure and lengthy provisioning processes, companies can consume network capacity and interconnection services like a utility. It gives enterprises the agility of the cloud applied to connectivity and transforms interconnection from a static, infrastructure-heavy investment into a dynamic enabler of business growth. That’s essential for the coming wave of agentic AI, because failing to achieve the goals laid out in by companies in their AI roadmapsis not an option: 17% believe that they will then not meet their revenue targets, 26% that they will need to delay their time-to-market for new products and services, and 28% that they will lose market share to their competitors.
