The global race to build data centers for artificial intelligence is uncovering a less visible limit than GPU availability or electricity: the fiber optic cables needed to connect all that computing capacity. Only the United States might need an additional 66 million miles of fiber to connect new data centers before 2029, while Europe and Asia are accelerating their investments in backbone networks and submarine cables to prevent connectivity from becoming a bottleneck in AI infrastructure deployment.
The key points about fiber optics for AI in 30 seconds
- The U.S. could go from 159 to 373 million miles of fiber by 2029.
- An additional 66 million miles of fiber would be needed solely to connect new data centers.
- The Fiber Broadband Association estimates that North America will require around 180,000 additional technicians over the next decade.
- Europe is preparing AI Gigafactories with high-speed connectivity and is allocating €200 million to new backbone networks and cables.
- Asia is expanding submarine routes as large data center markets like India and Southeast Asia grow.
The issue has a straightforward explanation. Large AI clusters don’t operate in isolation. They need to communicate with other facilities, cloud regions, enterprise networks, users, and exchange points. Redundant routes are also essential so that outages or physical cuts don’t isolate a facility that may have involved billions of dollars in investment.
The larger the concentration of computation, the greater the needs for bandwidth, low latency, and redundancy.
Hence, fiber availability is starting to be considered alongside electricity, land, water, and permits as factors that determine where new data center campuses can be built.
The U.S. will need more than double the fiber before 2029
The scale of the American challenge is detailed in a study by the Fiber Broadband Association (FBA), conducted with RVA and with estimates from Corning.
The research estimates that fiber routes could increase from approximately 95,000 to 187,000 miles by 2029. When measuring the total fiber installed within those routes, the increase is even greater: from 159 million to roughly 373 million miles of fiber.
Not all of this infrastructure corresponds to new digital highways.
The estimate includes about 109 million miles of additional fiber on existing routes, where capacity upgrades are necessary; another 66 million miles for connecting new data centers; and around 38 million miles for new long-distance routes.
The total deployment would be approximately 2.3 times the current size.
This is driven by the anticipated growth of hyperscale data centers. The FBA estimates their capacity could triple before 2029, mainly due to the expansion of AI applications.
The cable amount alone doesn’t tell the whole story.
An AI campus needs multiple physical connections and alternate routes. An excavator accidentally damaging a conduit shouldn’t be able to interrupt a cluster with tens of thousands of accelerators.
This requires designing independent entry points, redundant underground routes, and connections with multiple operators. The FBA also highlights cables with 1,728 fibers or more and technologies like Dense Wavelength Division Multiplexing (DWDM), capable of transmitting multiple optical channels simultaneously over a single fiber.
Fiber can thus become a bottleneck even when the data center already has land and power supply.
Having cable isn’t enough: thousands of technicians are also needed
The second challenge is human.
Deploying a fiber network requires specialized equipment capable of cable splicing, connector installation, optical measurements, troubleshooting, and ongoing maintenance.
The Fiber Broadband Association estimates that around 180,000 additional technicians will be needed over the next decade in North America to meet the demands of construction, maintenance, and network expansion.
This figure requires precision.
A previous study by the FBA and the Power & Communication Contractors Association (PCCA) projected immediate needs of about 28,000 workers for broadband deployment and another 30,000 technicians, totaling roughly 58,000 professionals for planned projects.
However, over a decade, the problem expands: around 119,200 workers could retire or leave the sector, potentially creating a shortfall of approximately 180,000 professionals.
The association is working to expand its talent pool with its OpTIC Path program.
As of May 2026, the initiative operated in more than 20 states, with eight more joining and over 38 educational organizations involved. More than 1,550 students had already completed the training.
The program includes 144 hours of training covering installation, splicing, testing, troubleshooting, and optical network maintenance. An additional specific module for data center technicians is scheduled to launch in Q3 2026.
This scenario reveals a little-noted aspect of AI expansion: automating certain digital tasks doesn’t eliminate the need for a vast amount of physical work.
Someone must build data centers, install cooling systems, lay cables, and splice fibers.
Europe has plenty of fiber but AI demands a different scale of connectivity
Europe starts from a different position.
As of 2025, fiber-to-the-premises (FTTP) coverage reached 74.1% of households in the European Union, while very high-capacity fixed networks reached 85.6%. Fiber has become the most widespread fixed broadband technology in Europe.
But connecting homes and linking AI data centers are distinct challenges.
A cluster with tens of thousands of accelerators needs backbone routes capable of transporting enormous traffic volumes between data centers, exchange points, cloud networks, and across countries.
The European AI strategy itself recognizes this dependency.
In July, the EU launched a call to establish up to seven AI Gigafactories, backed by up to €10 billion in public and national funding, with expectations to mobilize at least another €20 billion from the private sector.
These facilities will combine over 100,000 advanced AI processors with data centers, software, cloud infrastructure, power supply, and explicitly advanced networks and high-speed connectivity.
It also includes the 19 AI Factories already in the European strategy.
Therefore, fiber becomes a vital component for enabling this capacity to operate as a distributed infrastructure rather than a collection of isolated supercomputers.
The European Commission is also working to reduce deployment hurdles, particularly permits.
The Gigabit Infrastructure Act, fully effective from May 12, 2026, aims to cut costs and streamline the deployment of ultra-high-capacity networks. The Commission explicitly links the need for greater bandwidth with the growth of cloud computing and artificial intelligence.
Europe is also looking to submarine cables
The bottleneck doesn’t end at land borders.
Much of the international traffic between data centers travels via submarine cables. The European Commission notes that such infrastructure and other backbone fiber networks carry around 99% of global internet traffic.
In March 2026, Brussels announced calls for €200 million to strengthen this infrastructure.
About €180 million is allocated to deploying or upgrading backbone networks, especially submarine cables, but also terrestrial fiber and satellite communication stations. The remaining €20 million will fund intelligent cabling systems capable of enhancing infrastructure monitoring.
Some of these projects include direct links to Spain.
The future PISCES system will connect Ireland, Portugal, Spain, and France through roughly 3,000 kilometers of submarine cable. It will feature 16 fiber pairs and an open-access model for telecom operators, cloud providers, and academic institutions.
In addition to capacity, Europe seeks route diversity. For example, Ireland currently relies heavily on connections crossing the UK, while parts of Europe’s submarine infrastructure are aging.
Expanding data centers makes this redundancy an economic as well as a technical issue.
Asia is building its own submarine highways for AI
The Asia-Pacific region presents a different scenario. Here, some of the world’s largest digital markets coexist with geography that makes submarine cables especially crucial.
The growth of data centers in India, Malaysia, Indonesia, Japan, and other Southeast Asian markets is increasing regional connectivity needs.
India provides a clear example.
Its operational data center capacity has reached approximately 1.7 GW, four times more than six years ago, according to BloombergNEF data published this August. This includes about 1.3 GW under construction and 3.2 GW of approved projects.
International connectivity must grow accordingly.
One of the projects announced this summer is I-2SEA, a new approximately 3,600-kilometer submarine cable system connecting India, Malaysia, and Singapore. The consortium includes Microsoft, Lightstorm, Tata Communications, Singtel, ASEAN Cableship, and NEC, with service expected to start in Q4 2029.
India currently has 17 active submarine cables, totaling approximately 960 Tbps of capacity, with at least ten more projects in planning.
In Southeast Asia, investment is also rapidly increasing. AWS, Google, and Microsoft have collectively committed over $50 billion in cloud infrastructure and AI-ready data centers across the region, according to a report backed by Singapore’s Economic Development Board.
Projects like Apricot and SJC2 add new routes connecting markets such as Japan, Singapore, Taiwan, the Philippines, Indonesia, Hong Kong, and other parts of Asia.
A key difference compared to the United States is that in Asia, expanding AI infrastructure requires simultaneously upgrading terrestrial fiber and deploying an enormous network of undersea cables spanning thousands of kilometers across the ocean.
From GPUs to the infrastructure that connects them
The AI boom shows that a data center can no longer be judged solely by its number of GPUs.
Initially, the problem was electric power. Then came cooling, transformers, substations, turbines, and energy storage systems. Land availability and connection times to electric grids also entered the equation.
Now, optical connectivity is beginning to play a similar role.
All these resources are interconnected. Data centers are moving to locations with reliable power, but these sites don’t always have the fiber routes that traditionally accompanied major data hub markets.
Building where energy is available may later require deploying telecommunications infrastructure to reach that site.
The fastest-growing AI centers—particularly in the U.S.—highlight how the physical infrastructure continues to expand. GPUs, electricity, and data centers all need a fourth piece: millions of kilometers of fiber capable of moving data between them.
FAQs
How much additional fiber will the U.S. need for data centers?
The Fiber Broadband Association estimates that around 66 million miles of fiber will be needed solely to connect new data centers before 2029, aside from upgrades to existing routes and new long-distance connections.
Why do AI data centers require so much fiber?
Large clusters need to exchange data with other facilities, cloud networks, users, and internet services with high bandwidth and low latency. They also require multiple independent physical routes to maintain service during outages.
Is there also a technician shortage?
Yes. The FBA estimates that approximately 180,000 additional technicians will be needed in North America over the next decade, to support network construction, maintenance, and expansion efforts.
Do Europe and Asia face the same issue?
While the situation varies, the need is similar. Europe is strengthening backbone networks and submarine cables while building AI Factories and future Gigafactories. Asia is deploying new terrestrial and submarine routes to connect rapidly growing data markets across vast distances.
