Singtel, SKT target the cloud GPU businessSingtel, SKT target the cloud GPU business

As Singtel and SKT position themselves in the cloud-based GPU business, the biggest customers may well be other telcos.

Robert Clark, Contributing Editor

August 30, 2024

2 Min Read
Singtel headquarters in Singapore
(Source: towfiqu Barbhuiya/Alamy Stock Photo)

Singtel and SK Telecom have broken new ground with the world's first telco-provided GPU-as-a-service (GPU-aaS) offerings. The two Asian operators have made a string of announcements to position themselves for a piece of this fast-growing AI-driven business.

But while there's plenty of demand for Nvidia GPUs, it's not clear just how much value the cloud service can yield for telcos. The two companies are taking different paths. Singtel has just struck an agreement with regional operator group Bridge Alliance to sell the service to southeast Asia telcos, and another with Nscale to support it in Europe as well.

Bill Chang, CEO of Singtel's Digital InfraCo, said the company was forging a series of strategic partnerships to broaden service availability.

SK Telecom has teamed up with Lambda, a Silicon Valley startup in which it holds a small stake, to support AI cloud services just in South Korea. SKT and Lambda said by year-end they would deploy Nvidia GPU clusters on Lambda's cloud platform.

'Significant enterprise demand'

"SKT will launch AI cloud services, including GPUaaS, to enable enterprises to access GPU cloud services on an as-needed basis to develop or utilize AI services," the press release said.

Singtel foreshadowed its GPUaaS foray in February, providing access to Nvidia GPUs via its Paragon platform.

A GPU, with its hundreds or thousands of cores, can process complex graphic tasks and data calculations much faster than a traditional CPU, making it an affordable option for businesses that want to take advantage of AI-related apps or services.

One estimate is the market was worth $3.2 billion in 2023, will grow 35% this year and be worth $49.8 billion by 2032. But at this early stage, those numbers are just guesses.

Brian Washburn, a service provider analyst at Omdia, said there would likely be "significant enterprise demand" for GPUaaS, as it offered a cost-effective access to scarce and expensive Nvidia chips.

He said with this level of demand telcos that can launch GPUaaS "will have no trouble finding clients (more likely other types of service providers)." Telcos would be able to conveniently bundle in network connectivity and some orchestration.

But profit margins would likely get pressured as supply and demand start to balance out.

"The bigger plan is to get tenants who buy packages of hosting, network, and managed services. The total customer spend makes the margins work," he said.

Telecom operators' most valuable assets in the AI-GPU market could be their edge sites, he suggested. These can offer higher performance and lower latency, with possibly a more distributed power draw and infrastructure efficiency, he said in a note to Light Reading.

He also said the GPUaaS market most likely would be led by big cloud providers who will obtain GPUs in bulk, along with specialists like Coreweave, and IT cloud services players like IBM and Oracle.

Read more about:

AsiaAI

About the Author

Robert Clark

Contributing Editor, Light Reading

Robert Clark is an independent technology editor and researcher based in Hong Kong. 

Subscribe and receive the latest news from the industry.
Join 62,000+ members. Yes it's completely free.

You May Also Like