Big Tech's AI Ambitions Require Billions Of Dollars Of New Fiber, But No One Wants To Pay For It

The AI data center boom is driving an urgent need for massive investments in fiber infrastructure. But while investors are tripping over each other to fund new data centers, fiber providers are having a hard time accessing capital.

Networks of fiber-optic cables and other telecommunications infrastructure represent the information highways carrying data across the world, but they weren't designed for the demands of artificial intelligence. The AI models, products and services that tech giants are aggressively rolling out require data to flow at faster speeds and to more places than ever before. 

Insufficient network infrastructure has already become a pinch point, slowing AI development in the U.S. But industry executives speaking last month at Bisnow’s National DICE Connectivity, Edge and Telecom event said the problem is only getting worse, thanks to growing AI adoption and new use cases existing networks can’t support. 

Preventing network constraints from limiting the AI sector’s growth will require massive investments in deploying new fiber and evolving existing networks to anticipate the needs of the coming AI economy. Yet despite the urgency of the problem and the enormous scale of anticipated demand, executives said the capital needed for these forward-looking infrastructure build-outs remains frustratingly scarce. 

Amid rising fears of an AI bubble — and with memories of the role speculative fiber build-out played in the dot-com crash — investors and lenders remain unwilling to fund future-focused fiber projects that don’t already have a customer attached. 

The result is that despite the hype around AI, the economics of AI fiber deployments are harder than ever, ASG Chief Technology Officer Daniel Golding said. He said this problem may be “intractable,” with insufficient funding for fiber infrastructure continuing to slow Big Tech’s AI ambitions. 

“It's kind of miserable right now. It's very hard. It's bleak,” Golding said at the event, held at the Hyatt Regency Reston in Virginia.

“I hope we can figure out a solution for it, because we need a lot more capital to build a lot more fiber real fast.”

Fiber network traffic was already growing before the AI boom, but the technology is causing the volume of data flowing through data centers to skyrocket as more users engage with high-performance processing and as generative AI models produce synthetic data.

“We need more fiber,” Golding said. “We put fiber in the ground 25 years ago, and it's gone. You try to buy fiber along major routes, and you can't buy it anymore.”

AI is also changing the routes along which end users need this flood of data to travel. 

Power constraints facing the industry have pushed large-scale development for AI training outside of traditional data center hubs into rural areas that require network build-out. And hyperscalers are now training AI models across multiple data center sites instead of in a single cluster, requiring new high-speed fiber between those facilities that allows them to effectively act as a single data center.

Rapid corporate adoption of AI means that networks must also serve a growing number of low-latency inference use cases, applications like real-time fraud detection, where data must be processed instantly, no matter where the user is located. The edge network infrastructure needed to support these use cases will require both new fiber routes and significant upgrades to existing infrastructure to increase data transport speeds. 

Meeting the imminent needs of the AI economy will require nothing less than a systematic transformation of the entire U.S. telecommunications network, BIG Fiber Chief Commercial Officer Patton Lochridge said. 

“Everything has changed in terms of what's actually being demanded by the market,” Lochridge said. “It's a complete overhaul, physically, of the fiber networks needed to support all this.”

Avoiding network constraints will require a proactive approach to these infrastructure investments, Lochridge said. Projects need to be built to support demand and use cases that are likely to emerge in the coming months and years, even if they don't exist today.

While there may be risk attached to this speculative approach, panelists said the cost of running out of network capacity is far greater than the extra capital deployed to overbuild a fiber project beyond demand. 

“People have to be thinking about what's next, not what's now,” said Neila Wilson, Visa director for data center engineering and operations. “It’s going to take commitment to the future and investment.” 

But this investment isn't happening anywhere near the scale that is needed, Lochridge said. 

Between the major telecom giants like Verizon and AT&T and fiber infrastructure firms like Lumen, companies have taken several different approaches to addressing AI’s fiber requirements, but Lochridge said no firms are addressing this issue on a national scale or with the kind of capital that is needed. 

At the heart of the issue, executives said, is the fact that lenders and investors are showing almost no interest in funding fiber projects to support AI use cases beyond those driving demand today. 

“The biggest bottleneck is getting forward-thinking investors to release that capital,” Edgeology founder Adam Noll said. 

This desire for forward-looking investment in network build-out is running into a digital infrastructure financing ecosystem with a very low tolerance for speculative risk.

The fiber ecosystem faces a financing Catch-22, similar to the challenge for data center developers that need preleased capacity from a credit-grade tenant to secure project financing. Fiber firms struggle to obtain financing for build-outs unless they have a hyperscale customer under contract, yet no hyperscaler will commit without the firm already having financing in place.

“We're in this sort of very wretched phase of fundraising,” Golding said.

The reluctance to invest in network expansion stems from the dot-com crash more than two decades ago. That crash led to the bankruptcies of numerous fiber providers that had taken on substantial debt for fiber build-outs, betting on demand that materialized much later than anticipated.

As concerns mount about the magnitude of Big Tech's AI spending, tech leaders have refuted the bubble narrative by drawing a distinction between their investments and the late-1990s fiber build-out. They say their data center spending is driven by existing, demonstrable demand. 

While speakers at Bisnow’s event acknowledged that investors need to be judicious in evaluating future demand behind any fiber deployment, they said this aversion to risk has gone too far in shutting off capital for much-needed fiber. 

“People are still upset about the fact that everyone went bankrupt 25 years ago from doing this, but right now everything is sold,” Golding said. “I don't know how to get more money to build more fiber. It’s a mystery to me.”