While these figures, a projected 10GW of potential power demand are planning estimates rather than current operational loads, they highlight the scale of AI infrastructure now reshaping global networks.
OpenAI’s “Stargate” initiative, reportedly backed by partners including Oracle, SoftBank, and MGX, envisions hundreds of new sites worldwide to power next-generation AI models.
If realised, these facilities will place unprecedented demands on data centres, fibre networks, internet exchanges, and interconnection hubs. For carriers and network operators, such growth translates into a surge in bandwidth requirements, low-latency backhaul needs, and greater redundancy obligations.
Energy analysts caution that Fortune’s headline figures are speculative. The International Energy Agency projects global data centre electricity consumption to more than double by 2030, yet still represents a modest fraction of total global power.
BloombergNEF forecasts global data centre capacity reaching 277GW by 2035, or roughly 4% of electricity demand, which represents significant growth, but is still below the dramatic levels suggested in some media coverage.
Nonetheless, there is certainly a clear trajectory here: AI workloads are driving hyperscale demand for both compute and connectivity.
The industry is responding with expanded internet exchanges, including 400GE interconnection ports in London and Manchester, and strategic data centre deployments near renewable energy hubs and grid interconnection points. These measures aim to ensure resilient, scalable networks capable of supporting massive AI training and inference operations.
Hyperscalers, edge providers, and colocation operators are increasingly focusing on clustering infrastructure to meet latency, throughput, and sustainability requirements.
While the 10GW figure remains aspirational, it underscores that AI infrastructure is now a central driver of network planning. Operators and service providers will now anticipate demand spikes, integrate renewable energy sources, and ensure their interconnection strategies can support continuous, high-capacity workloads.
