AI 重塑海底与陆地网络 适配数据中心互联容量增长需求

Finance, business operations, national security, entertainment, and many other sectors depend on the connectivity provided by global fiber optic networks. Subsea fiber cables carry that connectivity across oceans to link continents and island nations. Increasingly, those intercontinental links are being extended directly into the data centers that run cloud and AI services.

Today, tens of thousands of data centers rely on that backbone to support AI training, inference, agents, and applications – a number set to grow further as global data center capacity, currently roughly 100 GW, is expected to double to 200 GW by 2030. The resulting increase in data center interconnections – and the rapid rise in their capacity – continues to drive corresponding growth in subsea cable capacity and terrestrial connectivity.

According to industry trackers, more than 570 subsea cables span about 1.5 million kilometers. This includes over 40 new subsea cable systems scheduled to enter service in 2026 alone, with additional cables planned through the end of this decade. This expansion is forcing subsea network operators and their terrestrial partners to address new challenges, including how to accommodate substantially higher capacity at cable landing stations (CLS) and across backhaul networks.

Related:Why AI Performance Starts Long Before GPUs

How AI and Cloud Services Are Changing Subsea-Terrestrial Networks

Historically, telecommunications providers, wholesale carriers, and cable consortia drove most subsea deployments. Submarine line terminating equipment (SLTE) was installed at the CLS, which functioned as the junction where subsea traffic was handed off to terrestrial networks.

Today’s dynamics look different. Hyperscale cloud and AI providers are increasingly driving deployments, laying cables to connect specific data centers across oceans. In this model, the CLS is not necessarily an endpoint; traffic can travel straight from one data center to another.

To meet capacity demands, cable systems that once carried four, six, or eight pairs are giving way to spatial division multiplexing (SDM) architectures with 24 or more pairs. A single CLS may now handle at least four times the previous capacity, intensifying pressure to reduce space, power, and cost at the landing sites while increasing the capacity required across terrestrial backhaul.

As a result, optimizing the overall network architecture is now just as important as adding or upgrading equipment. Operators are moving toward designs that treat the subsea and terrestrial segments as a single end-to-end network rather than as separate domains.

Related:Power Drives the AI Data Center Boom, but Connectivity Cannot be Overlooked

800G Coherent Pluggables Advance in Subsea Networks

New generations of 800G coherent pluggable optics are expanding from metro deployments into regional and long-haul roles. Deployed directly into a router for IP-over-WDM or in “thin” transponder setups that handle functions like demarcation and lower-speed service aggregation (Figure 1), coherent pluggables reduce footprint, power consumption, and capital costs compared to performance-optimized coherent optics.

Figure 1. 800G coherent pluggable optics enable power, space, and cost savings across terrestrial backhaul and for subsea links.

These advantages make 800G coherent pluggables well-suited for terrestrial backhaul, where they offer cost-effective, lower-power options and operation in combined C and L bands (“C+L”) to increase fiber capacity. Operators are also proving the technology on subsea routes, from short unrepeatered festoon links to repeatered systems running thousands of kilometers. A recent trial on a subsea route from the US to Puerto Rico saw the pluggables reach 600G per wavelength at a distance of 2,841 kilometers and 400G per wavelength at 5,862 km. New versions in development aim to extend reach to transatlantic distances.

Offloading Landing Sites with Optical Pass-Through

When the same operator controls both its subsea and terrestrial fiber, deploying reconfigurable optical add/drop multiplexers (ROADMs) at the CLS (Figure 2) enables optical pass-through. Instead of stopping at the CLS, wavelengths can continue directly into the terrestrial backhaul and terminate within the destination data centers.

Related:From Highways to Country Roads: How Backbone Networks Support AI Infrastructure

Making this work requires several ROADM functions to operate together, including amplified spontaneous emission (ASE) spectrum power insertion, spectrum sharing, and optical channel monitoring, as well as constant-power operation in the terrestrial inline amplifiers.

Figure 2. By eliminating back-to-back transponders at the CLS, operators can reduce power, space, and costs associated with transponder equipment.

Maximizing Backhaul Capacity with C+L WDM Spectrum

Reaching the data center itself is not always something a subsea operator can do on its own fiber. When another party owns the backhaul segment, dark fiber leases or managed optical fiber network (MOFN) services fill the gap. However, neither of these comes cheap once SDM cables supporting 24 or more fiber pairs are involved.

Figure 3. Using expanded C+L spectrum in terrestrial backhaul links can reduce the number of fibers used and any associated lease costs.

Using both C+L bands concurrently can increase the backhaul fiber’s capacity and improve the economics. While a subsea fiber is typically limited to roughly 4.5 THz of wavelength division multiplexing (WDM) spectrum, terrestrial fibers can support approximately 9.6 THz with C+L, and even more with extended bands such as Super-C and Super-L bands (Figure 3). That lets backhaul routes carry double the subsea capacity per fiber – or more.

Terrestrial spans also tend to be shorter than their subsea counterparts, which allows wavelengths to run at somewhat higher data rates. Layered on top of the C+L spectrum gains, this allows the subsea cable operator to reduce its backhaul fiber needs by 50%-75%, a meaningful cut to an operator’s leased-fiber or MOFN costs.

Conclusion

Subsea networks are now integral to end-to-end data center fabrics, enabling global scale across interconnection for AI and cloud traffic. By leveraging new technologies and optimized architectures, subsea network operators can stay ahead of capacity growth while lowering costs and power consumption and maximizing operational efficiency across both subsea and terrestrial domains.

About the Authors

Nokia

Serge Melle is Director of Optical Networks Product Marketing at Nokia. Serge has been at Nokia since 2019 and joined the Optical Networks division in 2021, where he has led the ON product marketing team, new product launches, campaigns, and a broad range of digital marketing activities. Prior to joining Nokia, Serge worked at Infinera and Nortel Networks in product/solutions marketing and business development, and at Pirelli Telecom Systems, where he was involved in implementing the industry’s first wavelength division multiplexing (WDM) network deployments. Serge is extensively published in the field of fiber-optic communications and sensing and holds a BSc in physics from Concordia University, Montréal, and an MASc in applied physics from the University of Toronto.