Mike Dano, Editorial Director, 5G & Mobile Strategies
April 23, 2025
4 Min Read
There are growing rumblings of a retrenchment in hyperscaler spending on data centers for AI. And that could have impacts – both short and long term – on the telecom market.
"Over the weekend, we heard from several industry sources that AWS [Amazon Web Services] has paused a portion of its leasing discussions on the colocation side (particularly international ones)," the financial analysts at Wells Fargo wrote in a note to investors, first reported by NBC News. The analysts warned that AWS delayed several new data center leases.
That warning comes just weeks after the financial analysts at TD Cowen reported that Microsoft canceled data center projects in the US and Europe, to the tune of 2 gigawatts – 2GW worth of data center power capacity is enough to power about 1.5 million homes. The rating is crucial for data centers because it determines their ability to support massive computing loads, like AI training or cloud services.
Moreover, the AI-powered data center boom cannot be overstated. Hyperscalers like Microsoft, AWS, Google, Meta and others are funneling historic amounts of cash into the construction of new data centers that can support AI. Here's just one statistic that puts the situation into context: The International Energy Agency (IEA) found that the 460 terawatt-hours (TWh) consumed by data centers in 2022 represented 2% of all global electricity usage.
Related:AI demand sends profit soaring for China optical vendor
'Routine capacity management'
However, it's difficult to determine the exact scope of the recent changes at Microsoft and AWS, and what the impact might be on the telecom industry, which often provides connectivity between data centers.
An AWS official took to social media to push against the notion that the company is reducing its spending on data centers in general.
"This is routine capacity management, and there haven't been any recent fundamental changes in our expansion plans. Fortunately for our customers, they're able to focus on their business and leave these details to us," wrote Kevin Miller, a top executive in AWS' data center business, on social media.
"First and foremost, we continue to see strong demand for both Generative AI and foundational workloads on AWS," Miller added.
Indeed, according to a recent note from the financial analyst at TD Cowen, "hyperscale data center demand remained strong" in the first quarter of 2025. The analysts wrote that Oracle and AWS led demand in data center leasing.
However: "It is worth noting that during our checks, we observed a moderation in the exuberance around the outlook for hyperscale demand which characterized the market this time last year," wrote the TD Cowen analysts.
Related:AWS builds AI agent for the DOCSIS 4.0 era
Further, the analysts wrote that AWS is shifting its data center business away from colocation deals in order to focus on its self-owned data center buildout projects. Colocation deals involve companies renting space, power and cooling in third-party data centers to house their own computing equipment, rather than building and maintaining their own facilities.
The TD Cowen analysts also noted that President Trump's contemplated tariffs could increase the cost to build new data centers by 5% to 15%.
Potential impact on telecom
Why should the telecom industry care? First and most importantly, telecom providers like Verizon and Lumen Technologies often provide the connectivity to transport AI data out of data centers and to the users who want it.
Indeed, Verizon recently claimed $1 billion in sales from shunting AI services over its existing networking infrastructure. Separately, Lumen late last year reported a total of $8 billion in private custom fiber agreements with the likes of Microsoft, AWS, Google and Meta.
But that's just one part of telecom's story in AI. For example, both AT&T and Verizon have also hinted at their interest in using their own networking infrastructure to run AI applications. Meaning, the operators would route AI requests to computers running inside their networks, thereby earning revenues from hosting those computers.
Related:Huawei pitches new AI cloud kit as Nvidia alternative
That particular business model remains in its early stages, but companies like Nvidia continue to argue that telecom network operators could score significant revenues from this kind of AI "inferencing."
Thus, any reductions in data center demand – from Microsoft, AWS or other companies – could temper the revenues telecom providers might make from AI services either now or in the future.
