GPUaaS represents a major opportunity — but ABI Research Analyst Larbi Belkhit said that revenue isn’t the sole motivation behind telcos’ AI factory buildouts
Telecom operators are positioning themselves as major players in the AI infrastructure space, with ABI Research projecting they could generate more than $21 billion in revenue from GPU-as-a-Service (GPUaaS) by 2030.
“Almost 100% of this is going to be AI inference in the workloads, not training,” said ABI Research Analyst Larbi Belkhit, author of a new report on the trend, in an interview with RCR Wireless News.
The report identified roughly 17 telcos worldwide that are currently building AI factories using Nvidia GPUs. The majority are located in the Asia-Pacific (APAC) region, with a few underway in Europe, the Middle East and Africa, and one announced in North America.
APAC telcos are taking the lead, Belkhit explained, because many are shifting away from traditional telecom services to embrace AI as a core business: “In our research, we refer to this strategy as telcos becoming sovereign hyperscalers” — national or regional players that own and operate their own AI infrastructure.
Standout operators in the region include Singtel, which announced in March 2024 a partnership with Nvidia to democratize AI through GPUaaS. Other APAC leaders include SoftBank, SK Telecom and Korea Telecom. In Europe, Telenor is a front-runner, though Belkhit said other European telcos are beginning to seriously evaluate the opportunity.
In North America, Verizon has taken a step in this direction with the launch of Verizon AI Connect, a collaboration with GPUaaS and cloud computing provider Vultr. The platform enables enterprises to deploy AI workloads at scale. However, Belkhit noted, “They don’t have a direct play yet — they’re more of a facilitator.”
So, how will hyperscalers react to telcos moving into GPUaaS? They probably don’t see them as much of a threat, said Belkhit — at least not right now. “In the last few years, the newer cloud providers or neoclouds have been gaining a lot of traction.” He added that telcos are partnering with these neocloud players, which could pose a more strategic challenge to the hyperscalers. “From a hyperscaler’s perspective, it’s about dealing with these neocloud companies first … then perhaps telcos partner with the hyperscalers instead.”
GPUaaS — not priority number one, and not the final step
While GPUaaS represents a major business opportunity, Belkhit clarified that revenue isn’t the sole motivation behind telcos’ AI factory buildouts. “They’re first and foremost built for internal workflow purposes and integrating AI,” he said. “The GPU-as-a-service opportunity …. is an entryway into the AI market for telcos [and] not the final step … because it goes back to what telcos are known for — managing infrastructure.”
