Telecom operators are rebuilding networks for the AI era as GPU clusters, distributed inference, cloud applications and AI data centers create new requirements for fiber capacity, optical transport, computing and low-latency connectivity.
The shift is changing telecom Capex. Operators are no longer investing only in spectrum, radio networks and consumer broadband. AT&T and Verizon are securing massive volumes of fiber, SK Telecom is targeting 15 GW of AI data-center capacity, Airtel is expanding toward 1 GW of data centers, and SoftBank is combining GPU clouds with AI-RAN.
The emerging network architecture increasingly connects AI data centers → long-haul fiber → metro networks → edge computing → 5G → devices.
AT&T Signs $3 Billion Fiber Deal for AI Era
AT&T’s September 2026 agreement with Corning provides one of the clearest examples of the scale of physical infrastructure required.
The multi-year contract is valued at more than $3 billion and will supply fiber and cable for AT&T’s network expansion.
Demand is already rising sharply. The average AT&T Fiber household consumes more than 1 TB per month, five times the 2016 level. AT&T expects usage to reach 2-2.5 TB per month by 2030 as streaming, cloud applications and AI increase consumption.
AT&T is targeting connectivity for 60 million Americans by the end of 2030.
The AT&T-Corning $3 billion fiber agreement shows how AI demand is becoming part of the business case for expanding physical fiber infrastructure.
Verizon Orders 80 Million Miles of Fiber
Verizon is making an even larger commitment in physical fiber volume.
Its agreement with Corning covers more than 80 million miles of high-density optical fiber and connectivity solutions between 2027 and 2032.
The infrastructure will serve two purposes: expanding broadband to homes and businesses and creating long-haul routes connecting AI data centers.
Through AI Connect, Verizon is deploying ultra-dense fiber across major corridors to support hyperscalers. The operator specifically identifies direct data-center connectivity as a requirement for AI workloads demanding massive throughput.
The Verizon-Corning AI fiber project demonstrates an important network shift: broadband and AI infrastructure can increasingly share the same high-capacity fiber investment.
Lumen Builds 58 Million-Mile AI Fiber Network
Lumen is restructuring its long-haul network around AI and hyperscaler demand.
The operator had 17 million intercity fiber miles deployed at the end of 2025 and plans to reach 47 million by 2028 and approximately 58 million by 2031.
Lumen has secured nearly $13 billion in Private Connectivity Fabric contracts, including demand associated with AI and hyperscaler customers.
Anthropic selected Lumen to expand high-capacity fiber connectivity supporting its AI operations.
For Lumen, AI infrastructure therefore means monetizing physical routes connecting large computing clusters rather than competing primarily for conventional consumer telecom traffic.
China Mobile Shifts Capex Toward AI Computing
China Mobile demonstrates how AI is changing the composition of telecom investment.
The operator plans approximately RMB136.6 billion of Capex in 2026.
Within that investment program, computing-network spending is expected to increase 62.4 percent, while AI-network investment is expected to rise 19.8 percent.
This represents a fundamental change from the traditional telecom model. Computing resources are increasingly being treated as part of the network itself.
China Mobile can combine one of the world’s largest 5G networks with cloud, computing and AI infrastructure, allowing enterprise workloads to move between centralized and distributed resources.
China Telecom Builds 118.8 EFLOPS of AI Capacity
China Telecom is following a similar strategy.
Its intelligent-computing revenue increased 95 percent year over year in H1 2026, while self-owned and accessed intelligent-computing capacity exceeded 118.8 EFLOPS.
China Telecom Cloud revenue reached RMB61.8 billion, up 7.8 percent, while AIDC revenue increased 9.3 percent.
Its Xirang 2.0 platform supports more than 20 types of heterogeneous AI chips and integrates computing, networking and AI-model scheduling.
The architecture demonstrates how telecom operators can potentially become orchestration platforms linking networks, GPUs, cloud infrastructure and AI models.
Airtel’s Nxtra Targets 1 GW of Data Centers
Bharti Airtel is building a similar network-compute model in India through Nxtra.
A $1 billion investment announced in March 2026 valued Nxtra at approximately $3.1 billion.
Nxtra currently has around 300 MW of capacity and plans to scale toward 1 GW, with Airtel targeting approximately 25 percent of India’s data-center market.
The company operates 14 large data centers and more than 120 edge facilities.
That footprint is strategically important because Airtel can potentially combine mobile and enterprise connectivity, subsea cable capacity, fiber, edge locations and hyperscale data centers.
AI workloads therefore create an opportunity to monetize multiple layers of telecom infrastructure rather than connectivity alone.
SK Telecom Targets 15 GW of AI Data Centers
SK Telecom is taking one of the industry’s most aggressive approaches.
The Korean operator plans to develop 15 GW of AI data-center capacity by 2035, with an initial 5 GW targeted for phased opening from 2029. It established SK Hyper and committed KRW750 billion through 2030 toward securing sites, substations and other infrastructure.
The first phase includes a GW-scale AI data-center cluster around Ulsan, followed by additional regional developments.
SK Telecom has also created SK Horizon, combining data-center and submarine-cable infrastructure. KKR and the IMM Investment-Stonebridge consortium are investing a combined KRW3.08 trillion in the business.
The SK Telecom AI infrastructure strategy illustrates how a mobile operator can expand into power-intensive digital infrastructure while using its telecom network to connect computing sites.
SoftBank Connects NVIDIA GPUs with AI-RAN
SoftBank is developing another model: distributed AI infrastructure integrated directly with the mobile network.
Its AI Data Center GPU Cloud, scheduled to launch in October 2026, uses NVIDIA GB200 NVL72 systems deployed in Japanese data centers.
The infrastructure supports workloads ranging from large-model training to latency-sensitive inference.
SoftBank intends to integrate the GPU cloud with AI-RAN, creating distributed infrastructure in which computing workloads can move between centralized AI data centers and mobile edge locations.
That architecture could become particularly important for robotics, autonomous systems, industrial AI and other applications requiring faster inference closer to users.
Singtel Adds 58 MW of AI-Ready Capacity
Singtel’s Nxera provides another Asian example.
Its DC Tuas facility in Singapore opened with 58 MW of AI-ready capacity, taking Nxera’s Singapore capacity to around 120 MW.
More than 90 percent of the new facility’s capacity had been committed before launch.
DC Tuas uses direct-to-chip liquid cooling, reflecting the higher power densities associated with GPU-based computing.
For Singtel, the opportunity is to connect data-center infrastructure with its regional enterprise networks and international submarine-cable connectivity.
Optical Networks Move Toward 1.6 Tbps
AI traffic also requires more capacity from existing fiber.
In India, Constl and Ciena demonstrated a 1.6 Tbps optical channel between Mumbai and Pune in September 2026, following a 1 Tbps achievement in 2025.
Vodafone Idea and Ciena have also demonstrated 1.6 Tbps over a single optical channel on Vi’s data-center interconnect network.
Higher-capacity wavelengths allow operators to carry more AI and cloud traffic over installed fiber before constructing additional physical routes.
NTT is looking further ahead through photonics and optical switching. Its IOWN AI Fund participated in a $125 million funding round for iPronics, which develops programmable optical-networking technology for AI data centers.
AI Is Redefining Telecom Network Architecture
These projects show that AI is not simply adding more internet traffic. It is changing where traffic originates, where it is processed and how telecom networks are designed.
AT&T and Verizon are expanding physical fiber. Lumen is building long-haul AI corridors. China Mobile and China Telecom are integrating computing into their networks. Airtel and Singtel are scaling AI-ready data centers, while SK Telecom is pursuing a 15 GW AI infrastructure platform.
SoftBank is pushing computing closer to mobile users through AI-RAN.
The emerging telecom architecture can increasingly be described as:
GPU clusters → AI data centers → optical interconnect → long-haul fiber → metro networks → edge computing → AI-RAN → devices.
For telecom operators, this creates a business opportunity extending well beyond mobile connectivity.
The operators that successfully combine fiber, data centers, GPU computing, edge infrastructure, submarine cables and 5G could become critical infrastructure providers for the AI economy.
In 2026, that is why telecom operators are rebuilding networks: AI is turning connectivity and computing into parts of the same infrastructure platform.
FASNA SHABEER
