Asia is a hotbed of telco AI factories

• Telcos around the world are investing in AI factories
• Many are partnering with AI technology giant Nvidia
• Asia is home to many of the operators developing such facilities

A growing number of telcos are investing in ‘AI factory’ facilities in an effort to capitalise on the broad business potential of AI infrastructure, with Asia home to many of the operators that are adopting such strategies. 

That’s one of the key takeaways from a new DSP Leaders Report, Trends in Telco AI Infrastructure, which can be downloaded for free here. 

The report focuses on three key areas: AI infrastructure – datacentres and AI factories; the role of the radio access network (RAN) in telco AI infrastructure strategies; and the role that telcos can play in connecting AI datacentre facilities. 

An AI factory, as defined in the report, is “a datacentre facility optimised for AI workloads, from data ingestion to training, fine-tuning and AI inference” and, as such, requires specialised infrastructure, most notably graphics processing units (GPUs).  

Unsurprisingly, the most common factor linking many of the telco AI factory developments is AI tech giant Nvidia: While it isn’t the only chip company with datacentre GPU products, it does dominate the sector, with a share of more than 90% of the datacentre GPU market, based on an analysis of 2024 sales by research firm IoT Analytics. As the report notes: “The plain fact is that, right now, Nvidia is the AI kingmaker. If you want AI infrastructure that can handle AI training workloads and large-scale, centralised inferencing, you’re almost certainly going to do business with Nvidia, either directly or indirectly.”

As you’d imagine, Nvidia isn’t shy about sharing which telcos it’s engaging with and, as the report was published, Nvidia had 18 publicly announced telco AI factory engagements, of which half are in Asia – with various telcos in India, Indonesia, Japan, Malaysia, Singapore and Vietnam – and another in Central Asia (Kazakhstan). A combination of sovereign service potential and local language AI model developments appear to be fuelling many of these deployments. 

Of the remaining Nvidia telco AI factory announcements, four are in Europe (France, Italy, Norway, Switzerland), one in the Middle East (Qatar), one in South Africa, one in North America (Canada) and one in Latin America (Ecuador). 

In addition, Nvidia has forged multiple other engagements (not classified as AI factory deployments) with telcos, including the likes of SK Telecom in South Korea, which has a broad range of deployments underway as part of its AI infrastructure superhighway plan, and Deutsche Telekom, which has teamed up with Nvidia to develop an AI gigafactory (essentially, a very large AI factory). 

Nvidia doesn’t have a monopoly position, though. For example, in Canada, where Telus is Nvidia’s AI factory partner, Bell Canada has unveiled its Bell AI Fabric strategy that will see the operator invest in a national network of AI infrastructure facilities, “starting with a datacentre supercluster in British Columbia that will aim to provide upwards of 500 MW of hydro-electric-powered AI compute capacity across six facilities”. The first Bell AI Fabric facility, a 7 MW AI inference facility in Kamloops, British Columbia, is being deployed in partnership with Groq, which develops and builds hardware specifically designed to accelerate the inference of large language models (LLMs).

For more on these developments and trends, download the Trends in Telco AI Infrastructure report now (and for free!). 

- Ray Le Maistre, Editorial Director, TelecomTV