We are moving rapidly towards a hyperconnected digital world in which artificial intelligence (AI) helps us to manage, monetize and materialize data in innovative ways. My previous blog post in this series examined several emerging AI applications and their impact on IP network evolution. This post focuses on the role network service providers (NSPs) can play—and the revenue opportunities they can unlock—in helping to deliver these AI applications cost-efficiently and reliably.

Where’s the money in AI?

AI technology is evolving fast but is in many ways still in startup mode. Hyperscalers like Apple, Alibaba, Amazon, Google, Meta and Microsoft are investing billions to build the huge data centers (and power plants) they need to train large language models such as ChatGPT, Gemini, Llama and Qwen. This is the quest for artificial general intelligence (AGI): to build ever-smarter digital agents that can mirror human cognitive abilities such as reading, writing, listening, speech, learning, reasoning, operating machines and performing complex tasks.

Figure 1. AI inferencing is essential for monetizing investments in AI training

While AGI is fundamentally important for human-machine interactions and applications such as AI chatbots and virtual assistants, the biggest payback will come from the countless applications that will use these pre-trained AI models for specific tasks, functions and data queries in what is called “inferencing.”

Consumer applications such as photo editing and smart home automation increasingly rely on AI to interpret private data and infer critical decisions. The same is true for numerous applications in commerce, finance, healthcare, manufacturing, transport and public safety. These inferencing applications favor more specialized or “narrow” AI models that are optimized for specific tasks, environments and data sets. Narrow models have a smaller resource footprint and can even function independently of the cloud in some cases.

Bridging the gap with the cloud

Currently, AI inferencing logic resides either in a data center or on the user device or premises. Carrying data back and forth between user devices and data centers takes time and money and comes with risks. And aside from practical scaling limitations imposed by power and space constraints, we don’t want data centers to become too big to fail because of man-made or natural disasters. On the user side, there are billions of highly diverse and widely dispersed devices and thousands of organizations that could benefit from AI. However, these organizations may not always have the necessary hardware or IT resources, or must still rely on external AI compute and storage resources for some applications  or functions (i.e., hybrid cloud and split inferencing).

Hosting AI inferencing workloads at the network edge, in between centralized data centers and user devices, will bridge this gap and address the following challenges:

• Reduce the dependency on centralized data centers by distributing the load of millions of daily user interactions over a large set of geo-redundant AI edge compute servers.

• Reduce network bandwidth costs and latency (data packet round-trip times) by bringing AI inferencing logic closer to user devices and data sources.

• Reduce the exposure to network congestion, cybersecurity and data privacy or sovereignty risks by keeping data traffic within the perimeters of a single managed network.

Figure 2. Bridging the gap between users and the cloud with AI edge inferencing

Unlike centralized data centers that must be carefully planned and dimensioned in advance to handle anticipated demand, building out the AI edge cloud can be largely demand driven. Moreover, AI edge compute can offload data centers and network traffic by preprocessing and curating raw data inputs. But the question is, who would build and operate it?

At the edge of a hyperconnected world

Cloud service providers (CSPs) can’t easily bridge the gap between their cloud data centers and end users on their own. They have the technology, but most typically rely on data center colocation exchange providers (CXPs) such as Equinix to host their server equipment so they can extend their cloud presence into large metros and cities. To go even further would be a bridge too far because of data sovereignty laws and the unsurmountable costs and logistical challenges of acquiring, equipping, operating and maintaining suitable edge locations on a global scale.

Network service providers (NSPs)—telcos, cable operators and mobile operators—literally live at the edge of this hyperconnected digital world. They can leverage their local presence, real estate assets, network infrastructure and professional services to enable cloud builders and digital infrastructure providers to scale out the AI edge cloud. The transition from digital subscriber line (DSL) connections over legacy copper loops to passive optical networking (PON) over fiber access is freeing up valuable space and power resources in distributed and central office locations. NSPs can use these resources to host AI server infrastructure for cloud builders, digital infrastructure partners, large enterprises and their own private use.

Based on their capabilities and comfort zone, NSPs may consider supporting a range of value-added services. For example, an NSP could:

• Offer colocation services with fiber access for container data centers in business parks, shopping malls, airports, hospitals and other venues that require local AI compute services.

• Lease out floor space or rack space for hosting AI compute servers in telco central offices and offer power, cooling, installation and maintenance services, secure (quantum-safe) fiber connectivity and virtual private network (VPN) services.

• Offer AI compute and software hosting services on behalf of wholesale partners and customers, as well as for in-house telco applications such as AI RAN and AI NetOps.

Driving revenue growth with services for AI edge inferencing

AI edge inferencing presents tremendous growth opportunities for network service providers. Connectivity is a key enabler for the AI era but that alone will not be enough to unlock its full value for our hyperconnected digital world. Actively participating in building out the AI edge cloud with value-added hosting services will enable NSPs to monetize connectivity services more effectively. It empowers them to move from being the “pipes” to becoming facilitators of value-added services that create new revenue streams, enhance the customer experience and inspire loyalty.

To learn more about this topic, please check out my talk at MPLS SDN AI world congress. For more information about our IP network solutions for AI, please visit https://www.nokia.com/ip-networks/ai/.