Lengthy-Haul Networks and Bandwidth Progress

How a lot long-haul community bandwidth is required to energy the AI revolution?

Whereas knowledge middle investments presently command nearly all of capital expenditure, the long-haul subsea and terrestrial networks that join these services are more and more important. These networks require a substantial amount of funding and planning for AI infrastructure, as they sew knowledge facilities collectively and allow customers around the globe to entry AI providers.

The long-term affect of AI on long-haul community bandwidth necessities remains to be evolving and lacks a easy, uniform reply.

Powered by our Transport Networks Analysis Service, this evaluation discusses the place we see the affect of AI on worldwide bandwidth demand, what the networks most impacted by AI can inform us about AI-specific demand, and future capability development in main cloud areas that would bolster AI utilization.

Sizing up the AI affect on worldwide bandwidth

All worldwide networks will find yourself carrying a point of AI-related site visitors. Used worldwide bandwidth is the first metric we use to evaluate the amount of world demand. Globally, our forecasts present that used worldwide bandwidth is predicted to develop at a compounded annual fee of 24% between 2025 and 2035. This fee of development implies a doubling nearly each 3 years.

Used Worldwide Bandwidth Progress by Area

Whereas it isn’t attainable to disaggregate AI-specific demand, our content material supplier bandwidth class consists of the networks of main hyperscalers, cloud suppliers, in addition to AI platform corporations and neoclouds. These are the networks most closely impacted by AI.

The businesses on this class are the dominant customers of worldwide bandwidth and accounted for 75% of all used capability globally in 2025. Worldwide content material supplier bandwidth is predicted to extend 9-fold from 2025 to 2035.

Used Worldwide Bandwidth by Supply

Mapping future capability development

Main cloud service suppliers (Amazon, Microsoft, Google) together with Meta (which lacks a cloud enterprise) have world networks that carry a large mixture of site visitors sorts, together with AI-based demand. Since AI inference is basically cloud-based, trying broadly at main cloud area places is a helpful method to see which places in every area are key nodes for future capability development. The map beneath reveals the give attention to cloud areas within the U.S., Europe, and Asia. The funding from these corporations in new terrestrial fiber and submarine cable methods is closely influenced by the necessity to interconnect these websites.

Choose Proprietary and Cloud Area Information Heart Areas, 2026

Notes: Map depicts main proprietary knowledge middle and cloud area places. Corporations depicted listed below are additionally current in lots of different shared places. Campus expansions in current places usually are not depicted with the “Deliberate” icon.

Two of the foremost AI platform corporations depend on the foremost hyperscalers’ networks for world attain and big compute, they don’t seem to be content material simply being customary cloud tenants. Each OpenAI and Anthropic are actively constructing out their very own personal networks to interconnect important knowledge facilities, which means they’re more and more appearing like community operators in their very own proper.

As a result of AI’s urge for food for energy is seemingly limitless, main hyperscalers aren’t simply letting workloads sit idle. Operators like Google are actively routing computing workloads dynamically throughout totally different knowledge facilities worldwide to optimize for energy and compute availability. Participating on this strategy requires a strong, high-capacity mesh community appearing as a world load balancer.

It isn’t all AI

AI has the potential to reshape world community topologies by forcing networks to comply with the search for vitality and compute effectivity. However let’s watch out to not attribute growing worldwide bandwidth demand necessities merely to AI. Bandwidth demand for cloud providers, video, social media, and the long-tail of functions that aren’t immediately tied to AI is very large and nonetheless rising. International community growth may also proceed to be wanted alongside conventional corridors.

Get extra intelligence like this

Our Transport Networks Analysis Service delivers knowledge and evaluation on long-haul networks and the undersea cable market, with forecasts of worldwide bandwidth provide, demand, costs, and revenues. Be taught extra and get analysis samples delivered to your e-mail.