Europe’s AI Ambitions Risk Bottleneck as Fibre Spending Captures Barely 1% of $700 Billion Capex

While the artificial intelligence sector has spent the past two years consumed by power availability and transformer shortages, an equally dangerous blind spot has emerged across the European continent: a critical deficit in high-capacity fibre-optic infrastructure.

Speaking at Capacity Europe 2026, network and cloud architects from Oracle, TELXIUS, SIPARTECH, and euNetworks sounded the alarm that Europe's AI ambitions could stall unless connectivity evolves at the same breakneck pace as compute and electrical procurement.

The 1% Infrastructure Imbalance The core of the problem lies in an extreme capital allocation mismatch across hyperscale budgets:

The $700 Billion Blind Spot: Global hyperscalers have directed more than $700 billion into digital infrastructure through 2026, yet only 1% of that capital has targeted network connectivity, according to Monica Martinez Quero, Chief Marketing Officer at TELXIUS.

The "Two Cans and String" Reality: Telecom leaders cautioned that without robust backhaul routing, high-density GPU campuses cannot communicate or ingest distributed training sets. If the 1% connectivity layer fails to materialize, the remaining 99% of compute infrastructure sits idle.

"If that 1% doesn't exist, then the 99% is not going to work. Connectivity cannot be left as an afterthought." — Monica Martinez Quero, CMO, TELXIUS

• Going Off-Grid: The Nuclear Pivot Urban power grids in primary data center markets have become so severely congested that developers are relocating to the periphery:

• Chasing Nuclear Capacity: To secure uninterrupted megawatt capacity, hyperscalers are increasingly constructing campuses adjacent to operational nuclear power stations, often situated in remote, rural areas.

• The Rural Connectivity Divide: Vincent Rousseau, Deputy Director of Operations at SIPARTECH, highlighted that these off-grid locations require extensive, brand-new dedicated fibre trunks to link back to major metropolitan aggregation nodes.

• Lead Times and Red Tape: Deploying intercity optical routes demands complex civil engineering, long-haul trenching, and lengthy municipal permitting processes—timelines that Shaunak Thakkar, cloud infrastructure lead at Oracle, noted are routinely underestimated by data center developers.

Inference as Europe's Competitive Moat Despite these physical bottlenecks, panellists argued Europe maintains an opening to dominate the next wave of computing: real-time AI inference.

Unlike foundation model pre-training, which can occur asynchronously in isolated geographic pockets, agentic inference demands sub-millisecond proximity to corporate headquarters and end-users. Industry leaders project that traditional tier-one FLAP-D hubs—Frankfurt, London, Amsterdam, Paris, and Madrid—will remain critical command centers for low-latency inference routing, provided operators and hyperscalers coordinate long-term bandwidth investments before capacity crunches hit.