- Telcos ceded the cloud's value to hyperscalers but AI's distributed architecture could reverse that dynamic
- Inference has to run close to users and data, which favors telco infrastructure
- But the same barriers that cost telcos the cloud — legacy systems, slow culture, dirty data — remain
When cloud computing revolutionized enterprise infrastructure, telcos were snoozing. Operators were too slow, too siloed and too focused on protecting existing revenue to capture the value created. Hyperscalers captured it instead, and carriers were left providing commodity bandwidth between hyperscale data centers and the users they served. It's a low-margin role operators have been trying to climb out of ever since.
"They messed up the whole cloud opportunity," said Sid Nag, president and chief research officer at Tekonyx, in an interview for Fierce Network Research's new report, "AI and the automated network: Designing telco infrastructure for the age of inference."
Why AI changes the math
AI offers a chance for telcos to get it right, because architecture requirements favor telcos, according to Fierce's research. The cloud era rewarded centralization: hyperscalers pooled compute into enormous data centers and won on economies of scale carriers couldn't touch. AI inverts that model.
Training still belongs in centralized hyperscale facilities, but inference — AI doing real work in response to real-world conditions — has to run close to where data is generated and where users are. An estimated 80-85% of AI workloads will be inference within one to two years, according to Jack Gold, founder and principal analyst at J.Gold Associates. Hyperscalers, by design, aren't close to users. Telcos are.
"Our infrastructure being closer to the end user, much closer than the hyperscalers, gives us the license to play and win in this economy," said Salim Kouidri, SVP of field engineering at T-Mobile.
Hyperscalers want to play at the edge too, but they need telco support to do it, which is another opportunity for service providers. Many hyperscalers and second-tier cloud providers are heavily leveraged on telco networks for connectivity, and that dependency will continue, Nag said. The long-feared threat of hyperscalers moving into the WAN market, meanwhile, remains theoretical. "It's never come to fruition," said MetTel CTO Ed Fox.
Barriers remain
The opportunity is real. Will telcos seize it?
None of the industry's structural weaknesses have gone away, according to our research. Telcos remain slow-movers, partly for good reason, since delivering five nines of reliability demands caution. Legacy OSS and BSS systems written in archaic languages constrain agility. And the organizational inertia that caused telcos to miss the cloud transition has not disappeared.
AI infrastructure is hard. Only 28% of AI infrastructure and operations projects fully meet ROI expectations and 20% fail outright, according to Gartner.
Data is another roadblock for AI implementation. Most operators' inventory systems are less than 50-60% accurate, according to Blue Planet's Gabriele Di Piazza. You cannot automate what you cannot see.
Most operators sit between Level 2 and Level 3 on the network autonomy scale, Di Piazza said — a finding TM Forum's recent research confirms — several levels short of the autonomy the AI opportunity demands.
The stakes are stark. Telcos that seize the moment can win a new lease on life; those that don't will continue to operate a commodity business, Nag said. And the opportunity won't last forever. Neoclouds like CoreWeave and Nscale are building AI compute without a network layer. That creates an opening for carriers, but it's time-limited.
Download the full report here: "AI and the automated network: Designing telco infrastructure for the age of inference,", and watch the companion virtual summit on demand: "AI and the Automated Network".