The theme of the Huawei Analyst Summit 2024 is "Thrive with Intelligence,” showcasing the significant progress in integrating artificial intelligence (AI) to innovate intelligent products and solutions.
In line with this, Telecom Review conducted an exclusive interview with Neil Shah, Vice President Research & Co-Founder of Counterpoint Research, to discuss the development and convergence of mobile communications and AI technologies (and the value this brings) as well as the progress and challenges concerning network intelligence.
Convergence of AI and Wireless Networks
Shah expressed from the start that “the timing is right” to merge AI and wireless networks together. “If you look at what has been driving this; it is the tremendous work which has been done across the stack, from semiconductors to software to cloud. On the network side, we are entering the 5.5G or 5G-Advanced era, and in future 2030 and later, 6G.”
The foundation which has been laid for 5.5G and 6G is ripe for more intelligent, software-driven innovation across the networks. Simultaneously, there has been a substantial surge in semiconductor development, enabling the deployment of increasingly sophisticated artificial intelligence (AI) capabilities both within and beyond network environments. Additionally, there is the capability to analyze data and perform inferencing processes on the data and data models under construction.
He elaborated that this represents the “spark” or driving force behind the convergence, not only across various use cases and applications but also in creating new opportunities for operators to monetize.
Agreeing that this convergence will bring more value, Shah highlighted that “we are just at the start, or nascent stages, of the integration of different technologies such as wireless networks and AI.” Together, both technologies will be “more powerful and they can create different value propositions for operators.”
Advocating for increased AI investments, these competitive advantages begin with optimizing or sustaining networks and extend to quicker deployment and creation of new services, ultimately resulting in enhanced user experiences for customers.
To illustrate this point, Shah said that it used to take one operator weeks to deploy a service, but with AI, coupled with smart policy management and the gathered data, “service deployment can happen in days, or even in hours.”
Goals of Operators
Telecom Review probed Shah regarding his opinion on the wireless networks’ goals in the next three to five years when AI and networks are converged. He answered, “there are two goals for operators: automate the network and make the networks more intelligent.”
Automating the network refers to automating all the processes. For that, AI needs to collect and assess a lot of data from the network which could impact different policies and processes that make the decisions faster without manual intervention.
On the other hand, there is AI itself. Shah cited the numerous devices that are using generative AI (GenAI) such as PCs and smartphones. This could potentially lead to an increase in AI workloads on the network.
“Earlier, it used to be just video—70% of the workloads in the networks—but now we'll see AI workload also taking that share, thus, operators must optimize the AI workloads across the networks,” he noted.
Operators must innovate across various domains, forming a comprehensive circle of improvement. This involves building new infrastructure and capabilities to fully automate and enhance the intelligence of their networks. Multiple elements, including the radio access network (RAN), core network, and a lot of other user plane-related tasks, have various speeds of AI adoption. Shah pointed out that the lowest hanging opportunities are within the network management side where operators can optimize the networks and save energy.
“To attain power savings, they strive to make all the processes very optimal. That adoption will be really fast, and then we'll move on to other parts of the network—the user plane or the UE side of things,” he clarified.
Overall, “it is a journey, and it will keep on evolving,” Shah continued. “It's not that in five years we'll reach a stage where everything is good, but the amount of workloads will gradually increase so operators will have to work with vendors like Huawei and Ericsson to build on more efficient networks.”
Progress of Network Intelligence
Shah outlined the adoption of AI across the network, emphasizing that vendors like Huawei have been working on this AI journey for almost four or five years, and many other operators have adopted AI in different ways.
“We have moved from software-defined networking to AI-based networks and now we are going to a new level of AI, which is using large language models (LLMs) as well as more automated, intelligent AI, which can compute even better, faster data processes,” Shah rationalized. “That is the level of AI which we are going to reach, and that is how the evolution is going to be—from software-defined networking, to basic AI, to more advanced AI.”
He described that many operators have leveraged software-defined networking and basic AI in optimizing networks and saving power. This evolution follows the same pattern with the generation of cellular technologies.
“In 3G, there was not much of AI; in 4G and 4G-Advanced (LTE), we had some amount of AI; in basic 5G, deployed in the last three years, AI has developed. With 5.5G and other new 3GPP releases, along with what other equipment vendors are adding on top of it, that is driving a lot of value for operators,” Shah indicated.
With more advanced wireless networks, the network intelligence progress is looking more positive, noted Shah. Considering the convergence of upgraded architectures and standards across proprietary technologies, the timing aligns perfectly as networks become increasingly programmable. This is made possible by 5.5G and beyond.
Challenges and Direction of Network Intelligence Innovation
Within the context of conversation, AI requires training on multiple data sets. As a result, various entities within the ecosystem will contribute foundational models. “Players such as Huawei, Meta, or Ericsson will build these foundational models. But how these equipment vendors can add more value is going to be a challenge for them and for the ecosystem, because all the vendors won't have access to all the data in the network, in the same way that not every operator will share this data to each other,” Shah underscored.
One of the biggest challenges foreseen within the AI ecosystem right now is transparency and ethics. Transparency should be at the core of every data training set across regions. “Whatever it is trained on in China, may not work in the Middle East; or whatever it is trained on in the US, may not work in Europe.” Furthermore, there are “AI regulations at a macro level as well as at a country level, which involves data sovereignty.”
Apart from that, computational capabilities and costs are also difficulties. “Right now, we are seeing a lot of advancement in semiconductors, which is allowing us to train those billion parameters and models, but there is a cost associated with it,” stated Shah.
In reality, training a model requires hundreds of millions of dollars. A significant challenge arises in determining how to distribute this cost, especially considering that the size of models typically increases by almost 3x/5x/10x annually. “Hence, the computational capability to train that model will also grow,” Shah added.
Furthermore, Shah identified another bottleneck: the environmental sustainability of AI operations. He noted discussions among companies exploring the possibility of integrating nuclear plants into their data centers to meet the substantial power requirements for training their models. Moving forward, Shah implied that many vendors and operators will have to take an “ecosystem and partnership approach” to optimize the usage of AI models.
In addition, “you will see a lot of calibration from different players in the ecosystem to make sure the AI used is efficient and ethical at the same time,” he remarked. “You can't put AI in everything but obviously in some use cases like Intelligent RAN, you'll have it more than in some other use cases, where it might not be prudent enough to use AI.”
Readiness for the Convergence of AI and Wireless Networks
Many operators have already deployed 5G and they're looking for the next killer use case to monetize. “AI is one of the key technologies, which, along with the convergence of 5.5G or advanced wireless technologies, will give them not only ways to save cost, but also ways to build new business models and services,” echoed Shah.
To do this successfully, operators will have to prioritize and align internally to adopt AI and understand how to deploy AI by working with ecosystem partners. This involves the talent pool within their teams. Shah emphasized two key factors. Firstly, it's primarily a mindset shift. Secondly, it involves education and understanding how AI can bring benefits to operators.
Given the current advancements, Shah also offered his perspective on the preparedness of carriers who are progressing towards the convergence of AI and wireless networks. He highlighted that advanced operators like China Mobile and China Unicom are ahead of different operators who have adopted AI. They have more advanced networks and have deployed Intelligent RAN and level 3 automation capabilities.
Taking this into account, some operators are ready to deploy, but there are still some operators which might have cost constraints or are not prudent enough to understand the benefits. It's not a matter of unwillingness; they intend to do it, and progress is evident because this evolution is natural for any network. “Similar to how they evolved from 3G to 4G to 5G, AI plus 5G and 5.5G is going to be the next evolution, which they cannot ignore,” Shah affirmed.
Right Combination, Perfect Timing: 5G + AI
In a closing remark, Shah emphasized that “adoption is going to be there.” He foresees that the rate at which adoption will occur “will depend and differentiate from region to region, from operator to operator.”
Shah envisions the level of convergence between AI and 5.5G being contingent on the deployment of 5G SA (Standalone). Once a 5G standalone network is in place, operators can leverage spectrum, policy management, and SLA-driven business models to a greater extent, with AI playing a pivotal role in accelerating these advancements. “This is a perfect combination, right now,” concluded Shah.