Telcos are starting to realise the role they can play in the nascent but fast-developing generative AI market. Beyond deploying the technology for their own needs, some of the most innovative operators are already working on creating propositions around supporting enterprises to adopt generative AI.
Telcos wake up for the AI opportunity
If you attended our STL “The Telco in 2030” event in November 2023, you heard us warning the telecom industry to be more ambitious when it comes to AI. Some AI giants were getting frustrated with telcos focusing too much on AI as only a cost-cutting instrument, and failing to see the opportunity to drive revenue growth.
Fast-forward to May 2024, and in our newly launched Telco generative AI adoption tracker, we were excited to see a significant number of generative AI projects within telcos designed to offer AI-related services.
AI-as-a-service is present in one in four generative AI projects
As the hype around AI, and especially generative AI, shows no sign of subsiding, companies are jumping onto the bandwagon, at the very least to demonstrate to investors that they are part of the AI frenzy. To explore what telcos are actually doing with generative AI, STL Partners created a tracker of public announcements from telecom operators worldwide showing plans, trials and actual implementations of this new technology.
In the first edition of the tracker, we collected 56 announcements from 30 operators. Most are large multi-national groups, but smaller, innovative telcos are also on the list. With group-wide implementations taken into account, the map of generative AI work within telcos looks encouraging.
Global distribution of generative AI implementation by telcos (May 2024)
Source: STL Partners
Importantly, 25% of the generative AI projects we have recorded are not just about building internal tools and improving internal processes, but about offering AI-related services. There are three groups of such projects:
1. Developing proprietary large language models (LLMs)
2. Offering generative AI solutions to enterprises
3. Building AI infrastructure to serve the enterprise market
Developing large language models for the industry
This is an ambitious undertaking. Building an LLM is expensive and requires time and talent. Given that several big LLMs exist, and improved versions are released regularly, one can wonder why a telco might want to develop its own. In addition, telcos tend to be short of AI skills, especially in markets where the hyperscalers are better than telcos at attracting AI developers.
Yet, China Telecom is reported to be working on its own LLM for the telecom industry. The telco intends to use the model itself, but also offer it to enterprise clients. Earlier this year, five operators – SK Telecom, Deutsche Telekom, Softbank, e& and Singtel – announced the Global Telco AI Alliance, which will create an LLM for the industry.
There are multiple reasons why these telcos chose to build their own LLMs instead of working with partners. China Telecom has a large scale and more resources, including access to talent, than many other telcos around the world. The size of its domestic market makes the return on investment easier to prove. Critically, the geopolitical confrontation between the US and China pushes Chinese companies to develop their own technology.
When it comes to the Global Telco AI Alliance, one of the reasons the five founding members cited for the venture was the desire to create an LLM that is optimised for the telecom industry. The unwillingness to be locked in a strategically critical relationship with a hyperscaler also plays a role.
It will be interesting to see if these two projects will succeed. They likely both have the data they need to train an LLM, but time is also of the essence. OpenAI, Microsoft, Google and Meta already have a huge foundation, on which they are building at speed. A telco-specific LLM might be of interest to the industry but if its features and capabilities are sub-par, it will find it challenging to attract other telcos.
Offering generative AI solutions to enterprises
This is the most natural avenue for innovative telcos to take as they look for ways to grow their revenues from the enterprise market. Large companies can partner directly with hyperscalers, but small and medium businesses might find it difficult to work with such large suppliers. For one, they may not even know where to start with AI, and need flexible small-scale approaches. Telcos are already a trusted supplier of connectivity and some IT services to them, and therefore should make a play for the nascent AI market for small and medium businesses.
STL’s tracker shows examples from Vodafone, Orange and KDDI, among others, building such propositions for the enterprise market in areas like document management, content creation, customer services improvement and others.
Building AI infrastructure for the market
Going beyond providing AI solutions, Swisscom is an example of a telco building infrastructure to support local businesses in their quest to deploy AI. Chipset shortage means that only big players have access to the latest crop, and even some telcos struggle to get their hands on enough silicon. Swisscom has partnered with Nvidia to create generative AI “supercomputers”, which it will use internally but also offer to its enterprise clients.
Telcos’ future in as AI-as-a-service
At STL, we will continue to track the deployments of generative AI by telcos, and expect to see more examples of building generative AI capabilities to offer as a service to their enterprise client base. The hyperscalers dominate the young AI market, but the speed at which consumers and enterprises are adopting, or wish to adopt, the technology suggests that the market is forming right now. Telcos are in a good position to benefit from this development, as long as they move fast and with a solid understanding of their customers’ needs.
