Artificial intelligence is no longer simply increasing demand for compute inside data centres. It is reshaping the assumptions that underpin how networks are built, operated and automated. Nokia’s latest expansion of its data centre networking portfolio reflects how deeply that shift is now embedded in infrastructure strategy.
Nokia has announced a new family of high-performance data centre switches alongside enhanced AI-enabled automation tools, positioning the network itself as both a performance bottleneck and an operational control plane for large-scale AI workloads. The move targets the growing requirements of advanced training and inference environments, particularly as agentic AI applications begin to place sustained, real-time demands on network fabric reliability and scale.
At the centre of the announcement is the introduction of Nokia’s 7220 Interconnect Router H6 switch family, combined with new Artificial Intelligence for Operations capabilities within its Event-Driven Automation platform. Together, the company argues, these technologies address two parallel challenges emerging from the AI super-cycle: the need for extreme network throughput and the need to operate increasingly complex environments with far fewer human interventions.
Networks under pressure from agentic AI
AI workloads are changing the profile of traffic inside data centres. Training large models and running distributed inference pipelines require predictable, ultra-low latency communication across thousands of accelerators. Even minor congestion or packet loss can delay jobs, inflate costs and leave expensive compute underutilised.
Nokia’s new 7220 IXR-H6 switches are designed to address this pressure. The platform delivers throughput of up to 102.4 terabits per second, with interface speeds of 800 gigabit Ethernet and 1.6 terabit Ethernet, effectively doubling capacity within the same physical footprint. The switches are compliant with Ultra Ethernet Consortium specifications, aligning them with emerging efforts to adapt Ethernet for high-performance AI and HPC environments.
Flexibility is also a central theme. The switches are available in both liquid-cooled and air-cooled variants and can be deployed across different rack configurations. Nokia is positioning this as a response to the varied physical constraints of AI data centres, where power density and thermal management increasingly shape infrastructure decisions.
A notable differentiator is software choice. Nokia says it is the only vendor offering hardware that supports both an embedded network operating system and the open-source SONiC platform. Operators can run Nokia’s SR Linux or Community SONiC, an approach intended to balance openness with vendor-supported reliability as AI networks scale.
Automation becomes a necessity not a feature
If raw performance is one side of the challenge, operations is the other. As AI environments grow, manual network management becomes untenable. Downtime carries a high financial cost, particularly when clusters are running continuously rather than in batch mode.
Nokia’s response is to embed agentic AI capabilities into its Event-Driven Automation platform. The enhanced EDA AIOps tools use natural language interaction and reasoning to help operators identify issues, perform root cause analysis and execute remediation actions. These capabilities are designed to work alongside EDA’s real-time telemetry, digital twin modelling, dry-run testing and instant rollback features.
According to a recent study by Bell Labs Consulting and Futurum referenced by Nokia, such approaches can reduce data centre network downtime by as much as 96 per cent. While that figure reflects a specific analysis rather than a universal outcome, it underlines the direction of travel. In AI-driven data centres, automation is becoming a prerequisite for reliability rather than an efficiency upgrade.
The emphasis on agentic AI is also telling. Rather than static rule-based automation, Nokia is positioning its platform around systems that can reason about network state and act with greater autonomy, mirroring broader trends in AI software development.
A wider signal for the AI infrastructure market
Industry partners see the announcement as part of a broader realignment. Tom Burke, chief revenue officer at Nscale, said Nokia’s technology strengthens the ability to deliver advanced AI platforms for shared customers, reinforcing confidence in infrastructure that must support rapid AI innovation.
From an analyst perspective, Alan Weckel of 650 Group highlighted the significance of 1.6 terabit Ethernet and Nokia’s commitment to Ultra Ethernet, noting projections that Ethernet will remain the dominant protocol for AI networking.
Availability timelines suggest the shift is imminent rather than theoretical. The 7220 IXR-H6 switches are scheduled for availability in the first quarter of 2026, with EDA AIOps features available for demonstration now and deployment by the end of 2025.
Taken together, the announcement underscores a simple reality. As AI workloads industrialise, the network is no longer a passive layer beneath compute. It is becoming a central lever for performance, resilience and cost control, forcing data centre operators to rethink not just how fast their networks are, but how intelligently they are run.
