Is There a Fiber Crunch?

There have been a number of articles in the industry press predicting a major shortage of fiber in 2026. Fiber manufacturers have already been working at full capacity due to the large amounts of fiber networks being built. Telcos like AT&T, Frontier, Brightspeed, Windstream, Consolidated, and many others have been busy building fiber. The big cable companies like Comcast, Charter, and Cox have been building fiber. There are numerous fiber overbuilders like Lumos and Metronet, which were purchased by T-Mobile and numerous other companies funded by venture capital. By my math, there was also over $13 billion spent in 2025 to build fiber, funded by grants and subsidies like ARPA, Capital Project Funds, RDOF, ReConnect, EA-CAM, etc.

2026 should also be a busy year for fiber construction. The telcos, cable companies, and fiber overbuilders are all planning a lot of fiber construction. There is still a little over $10 billion in planned fiber construction funded by the same existing subsidy and grant programs, plus there will start to be orders for fiber from BEAD grants as the year goes by.

Fierce Network talked to the major fiber manufacturers like CommScope, Clearfield, Corning, and STL, and was told that the companies are seeing unprecedented demand to provide fiber for AI data centers. This demand comes from both inside new data centers and also for the networks that tie data centers together.

I think people will be surprised to hear the amount of fiber wiring needed inside an AI data center. The Fierce Network article quoted Rahul Puri, the CEO of the Optical Networking Business STL, as saying that an AI data center needs 36 times as much fiber wiring as a normal data center. Anybody who’s ever been in a traditional data center will be floored by that assertion since there are typically large amounts of fiber wiring either under the floors or overhead of racks in a traditional data center. A Fierce Network article in December said that the giant 1 million processor data center being built in Louisiana by Meta will require 8 million miles of individual strands of fiber. Most of these strands will be part of fiber bundles of hundreds to a thousand fibers. The data needed to connect processors is gigantic.

Corning and other vendors are working on new technologies that will provide the needed connectivity within a data center, such as co-packaged optics that place optics and electronics closer together. Other vendors like STL are investigating hollow-core fiber to increase density and decrease latency.

There is also a huge demand for middle-mile fiber to connect AI data centers. Research firm RVA LLC predicts that 92,000 new route miles of fiber will be needed to connect data centers over the next five years. These are also big fiber bundles. My firm worked on a data center proposal last year that didn’t come to fruition, where the data center builder wanted a 512-fiber backhaul network.

One of the biggest challenges for the vendors is that there are different kinds of fiber for different uses, like inside a data center, in middle-mile networks, in last-mile networks, for drops, and inside buildings. The challenge for vendors will be to match manufacturing output with demand.

Vendors and industry experts are predicting that some kinds of fiber could experience ordering backlogs as long as a year. Vendors are likely going to satisfy their largest customers first, so smaller projects might find themselves in a bind.

It will be ironic after all of the hurry up and wait for BEAD if grant projects are badly delayed due to a fiber supply chain problem. But all of the industry predictions are based upon demand staying firm. There  are a number of credible predictions that there will be an AI market contraction in the coming year since data center supply seems to have outstripped the ability to generate the revenues needed to make the industry viable.

The Fiber Broadband Association says it is not expecting big backlogs in the fiber needed to build last-mile networks. I guess none of us will know for sure until we start seeing smaller ISPs place orders for fiber later this year.

The Growth of Backhaul Data

Zayo recently released a report that talks about the boom in backbone data usage in the country. For any readers who don’t know Zayo, it provides fiber connections for large data users and is one of the major companies that carry data between cities and across the country.

One component of backbone data is the accumulated usage of residential and small business broadband customers. We learned recently from OpenVault that residential and small business data usage grew 18% from 2023 to 2024. The average total usage per customer per month grew from 606 gigabytes at the end of the first quarter of 2023 to 663 gigabytes at the end of the first quarter of 2024. Zayo reports that data usage on fiber backbones grew far faster than the 18% that came from individual broadband users.

Zayo cited the following statistics:

  • Long-haul dark fiber sales were up 52% from 2023 to 2024. Carriers buy dark fiber when they want to send a large amount of data.
  • Wavelength capacity grew by 280% from 2020 to 2024. A wavelength represents the full bandwidth available from a band of light on a fiber. Zayo says that sales of 400 GB wavelengths accounted for more bandwidth than all sales from 10 GB and 100 GB connections in 2024.
  • Large buyers are buying the majority of new bandwidth. Zayo says that hyperscalers and carriers purchased 91% of dark fiber sales and 67% of all wavelength sales since 2020. Hyperscalers are large cloud users like Amazon AWS, Google Cloud, and Microsoft Azure.

Zayo says that AI traffic accounts for a lot of the new backbone data usage. The report discusses how AI data center usage in new markets is. For example, data sales in Memphis increased by 4,300% in 2024 and data sales in Salt Lake City grow 348%, both due to new AI data centers.

Zayo describes an interesting history of backhaul bandwidth. The report says that backbone bandwidth started growing faster than the bandwidth used by homes around 2007 due to the migration of data to the cloud. Large companies began storing data in data centers instead of locally as a way to protect data. Over time, a lot of the functions used by homes and businesses also migrated to the cloud as the common applications we all use moved to data centers instead of being stored only on people’s computers.

Zayo uses the term “distributed era” to describe the period that began in 2020 when the pandemic suddenly forced companies to expand networks to include people working from home. The decentralized workforce forced employers to find solutions to safely distribute access to their network to myriad locations.

Zayo coined the term “the intelligence era” for the period starting in 2024 and is characterized by modifying networks to handle the massive increases in data created by AI data centers. The changes are not just from larger bandwidth and include lowering latency and increasing real-time responsiveness to end-users.

The report digs a lot deeper than this blog and includes a  lot of interesting graphs of the detail of bandwidth growth since 2020.