Spirent Communications looks at the future of telco AI across the RAN, devices, data centers and operating models
In its position as a provider of test, measurement and service assurance solutions to operators worldwide, Spirent Communications has a unique view on where operators are directing investments and on what time scales. As it relates to artificial intelligence (AI), Spirent’s Stephen Douglas, head of market strategy, sees a present focus on telco AI as moving “from a tactical point solution to strategic implementation across all our business domains and the lifecycle which could unlock tremendous value.”
Focusing his commentary on the use of AI in the network, Douglas said that “incredible benefits could be achieving from driving new efficiencies in capital spending to optimize the peaks and the troughs, capital intensity periods, to delighting customers with new experiences and services, while creating those elusive new revenue streams.”
Generally speaking, Douglas said, operators will implement AI across customer service, marketing and sales, the network, IT and support; the goals here, also generally speaking, are to reduce operational and capital costs, enhance customer experience and increase revenue. In terms of specific high-priority applications/use cases, Douglas delineated:
- Network management tasks like network design, traffic prediction, capacity planning and radio map generation
- Operations and management, including network optimization, predictive maintenance, fault prediction, anomaly detection and root cause analysis.
- For security AI could assist with threat prediction, fraud detection and resiliency planning
- In the 5G radio access network (RAN), AI focuses will be around energy optimization, spectral efficiency, traffic steering, load balancing and mobility optimization.
- And for user equipment, think about AI and machine learning (ML) in the 5G air interface, and on-device generative AI (gen AI) capabilities.
The three As of telco AI—analytics, automation and AI
“I think a key thing is really that this direction of travel,” Douglas said, “ which is going to outline my predictions, it’s being set by ongoing advancements unlocked really through three interweaving functions over time: you have analytics, you have automation, and of course you have AI, both predictive and generative…Analytics and insights have been the network mainstay for…management for many years. And with the growth in data volume and our growing ability to actually harvest and access that data, both in real-real time and also historically, has allowed us to unlock the potential initially of predictive AI, but now also for generative AI.”
Specific to gen AI, Douglas highlighted its role in helping operators with content creation like troubleshooting guides, incident reports, network topologies and coverage maps, configuration scripts and test traffic generation for adversarial scenarios; querying documentation for field and support teams, and consulting historical data for similar issues and resolutions; and supporting predictive models for anomaly detection, augmenting data sets and preventing overfitting.
He gave the example of AT&T’s internal Ask AT&T gen AI tool being used by engineers for document interrogation and faster, more precise root cause analysis. “Traditionally root cause analysis has been a very manual and human-intensive process,” Douglas said, “sort of looking for a needle in a haystack of data and within lots and lots of network-oriented documents and guides. The use of generative AI is not only helping the AT&T teams act and resolve issues quicker. It’s actually freeing up engineering teams’ time…It’s also aiding them to move to become a more proactive organization rather than reactive. So these are substantial benefits that are already being demonstrated today.”
Some other factors Douglas called out that serve as a backdrop to his predictions are an explosion in data center capex, reckoned by Dell’Oro Group to go from $260 billion in 2023 to more than $500 billion 2027. The large language models (LLMs) that support gen AI will expand from billions of parameters to more than 1 trillion parameters. Graphics processing unit (GPU) cluster sizes will commensurately grow which will impact the data center interconnect fabrics needed to support lossless communications between GPUs and GPU clusters.
More on that data/data center point, Douglas said, “You simply can’t have AI without data…And it’s the data center which is becoming the evolving workhorse for AI…This is requiring a complete re-architecting of the physical design and the network fabric of the data centers today.”
For operators eyeing telco AI, “The time for talking is past”
Now for the predictions. Douglas sees AI in the RAN being marked by continued advancements in AI/ML-based energy efficiency then, further out, gen AI supporting semantic communications, reducing the need to transmit raw data in full. For devices, we’ll see an expansion of on-device gen AI across modalities supported by power efficient neural processing units (NPUs), then longer-term the emergence of a hybrid AI architecture bringing together on-device and cloud/edge processing. In the data center, Douglas sees growing use of Ethernet alongside InfiniBand for backend AI networking, followed by mass adoption of Ultra Ethernet Transport. And from an operating perspective, increasing propagation of automation frameworks and the use of AIOps will give way to full autonomy enabled by network digital twins for reinforcement learning and AI transparency.
Expanding on some of those predictions, Douglas said semantic communications could “revolutionize” telecoms “by reducing the need to transmit raw data in full. Imagine a world where you only needed to send 5% of a message and yet 100% could be accurately regenerated at the other end. This could be transformational.”
On full autonomy and transparency, he said, “I believe you may see full autonomy but it’s not going to be in the dark. We humans struggle with black boxes and AI needs to be explainable. Hence I predict full autonomy is going to be governed by network digital twins used to continuously test, to reinforce with learning and, most importantly, to provide AI transparency and understandability.”
Final thoughts from Douglas: “I don’t think telco should stop because other industries are moving so fast in this regard. I think the time for talking is past. We need to just move forward with it because the actual benefits are substantial…It would be, I think, detrimental to telco to wait until it’s got the most holistic, perfect sort of end-to-end architecture in place for it.”
This article is adapted from the recent Telco AI Forum which is available to view on demand.
