Lengthy-Haul Networks and Bandwidth Development

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

Whereas knowledge heart investments presently command nearly all of capital expenditure, the long-haul subsea and terrestrial networks that join these services are more and more essential. These networks require quite a lot of funding and planning for AI infrastructure, as they sew knowledge facilities collectively and allow customers all over the world to entry AI providers.

The long-term influence 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 influence 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 influence 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 quantity of world demand. Globally, our forecasts present that used worldwide bandwidth is anticipated to develop at a compounded annual price of 24% between 2025 and 2035. This price of development implies a doubling virtually each 3 years.

Used Worldwide Bandwidth Development by Area

Whereas it’s not attainable to disaggregate AI-specific demand, our content material supplier bandwidth class contains 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 anticipated 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 international networks that carry a large mixture of site visitors varieties, together with AI-based demand. Since AI inference is basically cloud-based, wanting broadly at main cloud area places is a helpful strategy to see which places in every area are key nodes for future capability development. The map beneath reveals the deal with 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 Knowledge Heart Areas, 2026

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

Two of the most important AI platform corporations depend on the most important hyperscalers’ networks for international attain and big compute, they don’t seem to be content material simply being commonplace cloud tenants. Each OpenAI and Anthropic are actively constructing out their very own non-public networks to interconnect essential 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 countless, 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 method requires a strong, high-capacity mesh community appearing as a world load balancer.

It is not all AI

AI has the potential to reshape international community topologies by forcing networks to observe the hunt for power and compute effectivity. However let’s watch out to not attribute rising 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 can even 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. Study extra and get analysis samples delivered to your e mail.