Page 10 - Connecting the Future How Connectivity and AI Unlock New Potential
P. 10

Connecting the Future: How Connectivity and AI Unlock New Potential



                  broadband in next-generation networks. Together, the standards and frameworks developed by
                  SG13 and SG15 support a hybrid approach to last-mile connectivity, combining the performance
                  of fiber with the reach of wireless technologies to ensure equitable and sustainable access to the
                  Internet worldwide. 8
                  Regardless of the underlying technology – fiber, satellite, or mobile broadband – connectivity is
                  ultimately delivered to end users through wired or wireless access technologies. In most settings,
                  especially in developed countries, most internet traffic is transmitted over Wi-Fi, even when it is
                  caried by mobile operators. In lower-income regions or mobile contexts, cellular networks often
                  serve as the primary means of delivering internet traffic, particularly in outdoor or transit environ-
                  ments.
                  AI relies on strong, modern digital infrastructure; optimizing each segment of connectivity is essen-
                  tial to enabling widespread, effective AI use and demands coordinated investment and policy
                  action.

                  1�1�2  Unique Requirements for Artificial Intelligence

                  AI’s reliance on large amounts of data requires significant investment in both computing and
                  connectivity. However, many parts of the world may still feel less prepared for these upgraded
                  requirements: Cisco’s 2024 AI Readiness Index indicated that more than half (54%) of global
                  companies acknowledged their infrastructure “has limited or moderate scalability and flexibility
                  to accommodate” the increasing needs of AI, while 78% lacked confidence “in the availability of
                                                   9
                  computing resources for AI workloads” . Maximizing AI adoption will depend on three key network
                  characteristics: bandwidth, latency, and stability of network access. Even with recent advancements
                  in connectivity optimization, cloud-based AI applications require bandwidths upwards of 10-100
                  Gbps, while complex AI tools like generative models and computer vision require 100-800 Gbps
                  or more. 10
                  These high data bandwidth requirements allow AI workloads to access, transmit, and process
                  massive datasets between Internet of Things (IoT) devices and specialized AI servers used for data
                  processing. Real-time AI applications, such as fintech, smart grid and smart healthcare, require
                  low latencies (near-instant response time) in data transmission, as even milliseconds of delay
                  could impact decision-making and safety. Equally important, successful AI usage requires stable
                  networks, guaranteeing uninterrupted operations and preventing service disruptions that could
                  erode trust in AI-driven solutions. 11
                  Successful use of AI-powered technologies will depend on reliable, redundant communication
                  networks for real-time data transmission, and the outcomes of AI integration hinge on the speed
                  and volume in which data can be accessed, processed, and acted upon. This network performance
                  requires delivery of services across different infrastructures (e.g., 5G, Wi-Fi, fiber and fixed wireless,
                  satellites) and among disparate systems and devices in the data center, cloud, and edge.
                  While AI is often associated with cloud-dependent models like LLMs, additional AI tools running on
                  edge devices – such as handheld diagnostic kits in rural clinics or smartphone-based soil sensors
                  in agriculture – are transforming services in areas with limited infrastructure. These systems use
                  onboard machine learning models that function offline and sync when network access is available,
                  thereby enabling broader inclusion in underserved rural or low-income regions.
                  As AI systems proliferate across everyday life, robust network infrastructure providing low latency
                  and network availability for high bandwidths will ensure widespread adoption and effectiveness.

                  1�2  Digital Infrastructure and Necessary Improvements


                  1�2�1  Subsea Cable Networks

                  The network of transoceanic data transmission has expanded to cover the globe with 570 in-ser-
                  vice systems  through over 1,450,000 km of cables.  Formerly servicing mostly long-haul Internet
                             12
                                                              13
                  usage, these networks are now dominated by demand from cloud providers, and data processing
                  requirements fuel an increasing portion used by AI applications.  AI data usage will certainly
                                                                            14
                  increase capacity demand in these first-mile networks, however government-imposed data sover-
                  eignty requirements  and localized data processing could ultimately limit significant increases in
                                    15
                  capacity requirements. 16

                                                           4
   5   6   7   8   9   10   11   12   13   14   15