- Telcos could turn enterprise AI’s token-cost crunch into a new infrastructure revenue stream
- Red Hat’s Stephen Watt argues operators are uniquely suited to tackle AI model routing and hosting
- The AI infrastructure window is big but risky, with Gartner warning telcos must plan for what the market looks like years from now — not just the next 12 months.
Telcos could reap the rewards of the emerging enterprise AI cost crunch, with Red Hat executive Stephen Watt arguing they’re well positioned to host and route AI workloads for companies that do not want to build their own infrastructure to manage spiraling token costs.
With the tokenmaxxing era behind them, enterprises are starting to zero in on AI spending with an eye toward controlling both consumption and costs. Hyperscalers have begun giving them more visibility and tooling, but there’s another lever enterprises can pull that cuts cloud titans out of the equation.
Watt is a distinguished engineer and VP in Red Hat’s Office of the CTO. He told Fierce Network that enterprise AI adoption is really a three-step process. The first step is ensuring AI can run securely. The second is managing costs. This can be done through quota management and model routing, with the latter focused on using less expensive, open models for certain tasks (see also: AT&T’s “tokenomics” strategy).
The third, self-managing, is less trivial, Watt said. That is, enterprises can take the models and host and route them on their own infrastructure to reap cost benefits. The catch is that retailers, healthcare providers and a wide array of other businesses don’t specialize in technology and probably aren’t super comfortable understanding all the AI infrastructure they need to purchase.
But there is a subset of companies – who aren’t hyperscalers or neoclouds – who do have a better understanding and who already have existing enterprise relationships: Telcos.
“I think there's an opportunity for telcos to step in and address the space for the companies that don't necessarily want to get into hosting their own infrastructure to solve this problem. There’s demand, and that demand can be addressed by telcos,” Watt told Fierce.
He continued: “Who knows routing better than telcos? I think they're the classic experts at network routing. And this is inherently a network routing problem.”
Sizing up the AI opportunity
A new forecast from Gartner indicates global AI spending will jump 49.5% year on year in 2026 to $2.7 trillion. AI infrastructure is expected to make up more than half ($1.5 trillion) of that total.
Spending is expected to jump again to $3.6 trillion in 2027, with AI infrastructure accounting for nearly $2 trillion of the total. Telcos building AI-optimized servers would fall into the AI infrastructure category.
“If a telco decided to build an AI data center and then start offering out that capacity to enterprises to run their self-trained models or their trimmed down models…[that’s] certainly a viable business,” Gartner Distinguished VP Analyst John-David Lovelock told Fierce.
He acknowledged that operators have the access to capital, ability, influence and partnerships required to make such a play but argued they lack the skills and workforce required to run data centers at scale. That’s a problem that could be solved via M&A, he added.
“There’s no reason that the telcos couldn’t buy the neoclouds,” he said. “If I wanted to get into the AI infrastructure race quickly, have revenue and reduce my risk and hire all the people that I wanted to hire, why wouldn’t I just buy somebody already doing what I want to do? Pay the premium, get my name in the ring earlier, get the revenue stream earlier.”
Lovelock pointed out that “it’s a good season to buy for mergers and acquisitions,” noting a lot of 2023-era generative AI startups are struggling or failing.
The four coming AI inflection points
But that doesn’t mean a move into the AI infrastructure space would be without risk.
“If you think you know anything about what AI is going to look like in the next two years based on what we’ve gone through in the last 12 months, you are wrong,” Lovelock warned.
He predicted four major turning points are coming up in the industry.
- Like generative AI, AI agents will soon move into what Gartner calls the “trough of disillusionment.”
- There will be an inversion of supply and demand where demand will exceed supply ...
- ... causing free AI experiences will disappear.
- “And then the market is going to collapse,” Lovelock said.
By this, Lovelock said he means it will be a replay of the early days of the cloud market, where eventually two or three players were designated winners. Those winners cornered the lion’s share of growth while the losers were left fighting for scraps.
“Getting into the market now based on what you think the next 12 months is going to look like is bad business,” he concluded. “If you’re starting to build a data center now, you have to be thinking about what does the market look like in 48 months, because that’s more reasonably when you’re going to be able to sell your capacity and the market you’ll be operating in.”
Read more about AI cost management on Fierce Network
Enterprises risk AI sticker shock as token costs pile up
Open models are driving AT&T’s AI ‘tokenomics’ strategy
AI enters cost crunch era as hyperscalers zero in on affordability, controls
Enterprises now write $100M checks for AI infrastructure, says VAST Data VP