Artificial intelligence is transforming not only applications and business processes – it is also creating entirely new demands on the underlying infrastructure. In particular, data centre networks must keep pace with the rapidly increasing flood of data, new types of workloads, and extremely low latency requirements. Ahead of his presentation on “AI and its impact on data centre networking” at this year’s Data Centre Expert Summit, eco spoke with Nils Kleemann, CTO – Central Europe at NOKIA, about current developments, the need for strategic shifts, and his vision of an AI-ready network.
What changes are you currently observing in the area of data centre networking due to the increasing use of AI?
The growing data traffic associated with AI is certainly one of the key developments. According to its latest Traffic Report, Nokia Bell Labs expects that by 2033, AI-generated global Wide Area Network (WAN) traffic will account for 33% of total WAN traffic. The lion’s share of this will be driven by the consumer segment (38% of total global consumer WAN traffic), although the enterprise segment is expected to see the highest growth rate (57% CAGR). As a result, the interconnection of data centres is becoming increasingly important and part of the critical infrastructure, which must be scalable (fibre), redundant, and high-performance.
Alongside the very large language models (LLMs) that are trained in large-scale data centres and support AI applications, there are also smaller LLMs that run on end devices such as smartphones or IoT devices. However, not all such devices have the processing power to handle the relevant AI computations. For this reason, such AI processes need to be outsourced to nearby edge data centres, thereby driving demand for edge data centre infrastructure.
To what extent does the AI boom require a rethinking of network architecture – both for hyperscalers and smaller providers?
We are certainly witnessing a transformation of today’s networks. This starts within the data centres themselves. To achieve the extremely high computational performance of GPUs in an AI data centre, these processors operate in parallel – placing extreme demands on the backbone (e.g. 400G/800G interfaces) and requiring ultra-low latency. This is why the Ultra Ethernet Consortium (UEC) was established, of which Nokia is a member. Entirely new standards are being developed here.
In operating such AI data centres, automation – particularly intent-based automation tools – will play an increasingly important role in avoiding human input errors. Regarding security, especially in the context of rising cyberattacks, a shift in thinking towards scalability is also essential. For instance, in the case of DDoS attacks, the corresponding IP packets have traditionally been routed to a dedicated scrubbing centre. A more scalable approach would be to implement this scrubbing directly on an IP router (e.g. at the data centre gateway).
With the expected deployment of quantum computers in the coming years, quantum-secure data transmission – such as between data centres – will also move into the spotlight. Furthermore, due to the growing importance of edge data centres, we are seeing an increase in ‘east–west’ traffic and a greater demand for support of more symmetrical traffic patterns – which must be given greater consideration in network architecture.
How are network providers such as Nokia strategically preparing for the increasing demand for AI-capable networks?
Nokia is addressing these developments with dedicated solutions that can be grouped into five areas:
▶ Datacentre Fabric (DCF) – high-performance, scalable interconnection within data centres
▶ Datacentre Gateway (DC GW) – scalable, multi-service routing solution for a wide range of application scenarios
▶ Datacentre Interconnect (DCI) – optical and IP-based transmission solutions between data centres
▶ Automation – intent-based automation solutions based on Kubernetes
▶ Security – solutions for DDoS mitigation and quantum-safe encryptionIn addition to these solutions for expanding and transforming network technologies to transport AI data streams and applications between data centres and end users, Nokia is also investing in the use of AI within network technologies themselves. The central question here is how AI can simplify or optimise network operation, end-user services, and performance. This leads to the concept of “AI-native networking” – an approach that breaks through the limitations of traditional network technologies. In this respect, Nokia is pursuing a two-pronged strategy: networks for AI-enabled services, and AI-enabled networks.
What is your vision of an AI-ready network in 2030 – which principles, architectures, or paradigms will dominate?
When AI changes the network: New directions in data centre networking. The approach should be to provide “AI-enabled networks for AI-enabled services” that are scalable, highly automated, and secured against cyberattacks (DDoS, quantum). Only in this way can critical network infrastructure support a country’s digital sovereignty.
