The surge in AI infrastructure spending is reaching the equipment that moves data between processors, racks, buildings, and regions, with Nokia recording a sharp increase in sales to cloud and AI customers during its second quarter.

Nokia generated net sales of €4.8 billion, up nine per cent at constant currencies, while its Network Infrastructure division grew by 12 per cent. Optical network sales increased by 20 per cent and IP network sales rose by 16 per cent.

Sales to customers classified within AI and cloud expanded by 105 per cent, while orders from that group reached €2.8 billion. Nokia expects roughly half of those orders to convert into revenue within 12 months, with customers making longer commitments partly because some components remain difficult to obtain.

Datacentres have always required networks, and cloud growth has supported optical and routing investment for years. AI clusters are reinforcing that established direction because thousands of expensive accelerators need to exchange information continuously, making network performance part of the economics of the computing system.

Processors cannot work in isolation

An advanced processor delivers little value when it spends too much time waiting for data or for other machines to complete their tasks. Training and operating large models can involve intensive communication across a cluster, which places greater demands on switching, routing, optical transport, and network management.

Latency, congestion, packet loss, and hardware failure can reduce the utilisation of machines costing tens of thousands of pounds each. A small fall in utilisation across a large installation can leave substantial capital sitting idle, which turns network design into a direct influence on the return from AI spending.

Nokia sells into several parts of that chain. Its IP products move traffic within and between networks, while optical systems carry large volumes over longer distances and link separate datacentre campuses.

The growth figures are useful precisely because they do not depend on describing every network product as artificial intelligence. Much of the revenue reflects familiar infrastructure being purchased in greater volume and at higher capacity because AI workloads create new traffic patterns.

Longer orders give Nokia more visibility over demand, although they also expose the company to a relatively small collection of hyperscalers, datacentre operators, and specialist AI infrastructure businesses. Large customers can move projects between quarters, negotiate aggressively, or design more equipment internally.

Component shortages add another tension. Scarcity may encourage customers to reserve capacity earlier, but suppliers risk holding expensive inventory if constraints ease or projects are delayed after equipment has been ordered.

Growth arrives during a wider restructuring

Nokia’s comparable operating margin reached nine per cent, while reported profitability was affected by the acceleration of restructuring. The contrast shows that demand from AI infrastructure is expanding while the group continues to reduce costs and reshape areas exposed to weaker telecoms investment.

Additional restructuring has been concentrated partly in Europe and is expected to create significant charges. Higher sales into cloud infrastructure do not consequently translate into uniform workforce growth, particularly when management is trying to improve returns elsewhere in the portfolio.

The European policy debate around AI capacity has concentrated heavily on access to processors, but the figures expose the wider industrial requirement. Optical equipment, routing, fibre, cooling, power systems, and skilled operations are all needed before an accelerator can produce useful work.

Nokia and other European network suppliers may benefit from public supercomputers, sovereign cloud projects, and private AI campuses, although those customers have different purchasing patterns. A research facility needs open allocation and specialist support, whereas a hyperscaler values standardised designs, rapid deployment, and global purchasing power.

Enterprises building private AI environments create another market, but most will operate at a smaller scale and may buy through integrators rather than directly from equipment manufacturers. Their demand will depend on whether sensitive workloads justify dedicated infrastructure instead of public cloud services.

Current spending can support several years of network expansion without removing the risk of overbuilding. Customers may reserve more capacity than they eventually use, while improvements in processors and model efficiency could change the volume of infrastructure required for a given workload.

Nokia’s quarter offers evidence that the financial benefits of AI investment are reaching beyond the most prominent chip suppliers. Whether that becomes a durable replacement and expansion market will depend on how much of the newly built compute is used, how often it needs upgrading, and whether customers continue to add capacity after the first wave of construction.