When Vodafone announced in January it would invest $1.5 billion over the next 10 years in cloud and customer-focused AI services developed in conjunction with Microsoft, it said the deal would include replacing multiple physical data centers with virtual ones across Europe. So far, so unsurprising. Over the last decade or so many telcos in mature markets have chosen to shed physical data centers.
The rise of AI, however, is encouraging a handful of network operators, and notably large Asian players, to take a very different approach and instead ramp up their spending on physical computing infrastructure.
In South Korea, for example, SK Telecom is pursuing AI-driven demand for data centers as part of its AI ambitions. Speaking on the company’s recent earnings call. SK Telecom CFO Yang Seob Kim said the operator plans to “drive our data center business more strongly, focused on next-generation AI data centers and global expansion.” For now, its move appears to be paying off given that the company reported a 30% increase in full-year data center revenue in 2023.
In December Japan NTT announced it was setting up a new company with TEPCO Power Grid, Inc. to jointly develop and operate data centers in Greater Tokyo. And its sights don’t stop at the domestic market. NTT Global Data Centers is investing 1.5 trillion yen (approximately 12 billion USD) over the next five years to expand and upgrade its data center business globally, citing a global opportunity to meet demand for data usage linked to rising use of GenAI, among other technologies. Softbank, meanwhile, is building data centers in collaboration with NVIDIA, which are designed to host generative AI and wireless applications on a multi-tenant common server platform.
Reliance industries, the parent company of Jio, India’s largest telco by number of subscribers, has partnered with Nvidia to build supercomputing infrastructure for AI. According to a statement: “Reliance will create AI applications and services for their 450 million Jio customers and provide energy-efficient AI infrastructure to scientists, developers and startups across India.”
Sovereign AI
Nvidia also scored a win in Europe with Swisscom, which is building what it calls a ‘trusted AI factory’. The telco intends to invest up to 100 million Swiss Francs into AI solutions over the next years, including generative AI full-stack supercomputers in Switzerland and Italy.
Swisscom sees an opportunity in helping its enterprise customers, which include some of the world’s largest banks, and government organizations, meet regulatory requirements for data governance, as well as ensuring security.
“A customer must be sure what happens with its data, where it happens and how it happens. Switzerland, with its many multinational organizations, needs trusted and unique sovereign building blocks for AI,” said Christoph Aeschlimann, CEO of Swisscom, in a statement.
Nvidia is often described as selling shovels in the AI gold rush and it turns out that providing data center platforms that include GPUs and software is proving to be big business for the company. The company recorded data center revenue of $14.51 billion, up 41% from Q2, up 279% from the previous year for the financial quarter ended 29 October 2023.
The rush to build new data center facilities for AI doesn’t appear to be letting up any time soon. In the US, according to a report by CBRE published in September, “an all-time high of 2,287.6 MW was under construction in primary markets, a 25% year-over-year increase.” At the same time, “the overall vacancy rate for primary [US] markets remains near a record low, at 3.3%,” according to CBRE, adding that Northern Virginia’s vacancy rate is 0.94%. And “many AI startups are seeking large requirements between 5 to 25 MW.”
Nonetheless, telcos that invest in the physical infrastructure and transmission networks that help them and their customers pan for AI gold have to weigh up carefully whether, where and how to invest.
The energy conundrum
AI, and GenAI in particular, is power-hungry. The IEA’s Electricity 2024 report, for example, forecasts global energy consumption of data centers, AI and cryptocurrencies will increase from 460 Terawatt hours (TWh) in 2022 to a range of between 620 TWh and 1050 TWh in 2026. Therefore a major consideration is access to low-cost and sustainable sources of energy. Indeed telcos are emphasizing the sustainable credentials of their data center plans. Singtel’s Nxera infrastructure subsidiary, for example, says it plans to build “green, sustainable and hyper-connected AI-ready data centers” in the Southeast Asia region.
Linked to the search for energy is the job the data center will do. Equinix points out that although GenAI development workloads require a lot of compute, memory, networking and storage resources, they aren’t latency sensitive, so they can be located near cheap energy and away from urban centers.
GenAI production workloads, however, are a different matter. They may have large numbers of users simultaneously accessing the generative AI model, making it sensible to locate them in edge locations close to where the data is being generated.
Softbank’s new data centers, for example, “will be more evenly distributed across its footprint than those used in the past, and handle both AI and 5G workloads. This will allow them to better operate at peak capacity with low latency and at substantially lower overall energy costs,” according to Nvidia.
Virtual competition
A primary question, however, has to be whether telcos can compete with the deep pockets and established business of hyperscalers, or data center specialists such as Equinix.
AWS, Microsoft Azure, Google Cloud, Alibaba, and Oracle, for example, already hold a strong competitive advantage, as IDC points out, even if some pockets of opportunity remain. “Where global players have limited footprint, such as in Eastern Europe or the Middle East and Africa, there is greater scope for national or regional players like telcos to make the infrastructure landgrab to establish a leading position in these markets,” notes Chris Silberberg, Research Manager, EMEA Telco Insights, IDC.
Nonetheless national – or even international – scale is not a guarantee of profits in the short- to mid-term. “Google Cloud which includes GCP and Google’s workspace initiative lost nearly $3bn in 2022, despite growing revenues to over $26bn, up $7bn YoY,” states Silberberg. “There is the risk that national players in many markets simply won’t have the scale to square this as a competitive commercial offering without significant policy maker support and funding.”