> ## Content Index
> Fetch the complete content index at: https://broadbandbreakfast.com/llms.txt
> Use this file to discover other available public pages before exploring further.

# Glenn Lurie: There Is No AI Without a Pipe
- URL: https://broadbandbreakfast.com/glenn-lurie-there-is-no-ai-without-a-pipe/
- Published: 2026-08-25T20:50:25.000Z
- Updated: 2026-08-25T20:50:33.000Z
- Description: AI's infrastructure isn't just about chips and data centers. It's about the fiber, wireless, satellite, edge and power infrastructure required to move intelligence everywhere.
- Author: Glenn Lurie
- Tags: Expert Opinion, #with-siderail, AT&T, AI, NVIDIA, Nokia, Data Center, SpaceX, IEA

I spent 36 years in wireless, including years as AT&T’s Mobility CEO and through the launch of the original iPhone and iPad, and I've watched a handful of significant moments when whole industries and society overall had to rethink itself all at once. 

AI is definitely the latest significant shift, but most coverage and discussions are treating it as a story about models and chips when it should be a story about pipes and networks.

[![CTA Image](https://storage.ghost.io/c/29/97/29976d9d-89ea-4c77-8c0b-0e8a0eafed35/content/images/2026/08/250150_course_graphic_YT_wordmark_title-1.png)](https://broadbandbreakfast.com/course/#bbb-offer) 

[***Sign Up for FREE for Section A**](https://broadbandbreakfast.com/course/#bbb-offer)

[Sign Up for FREE for Section A ](https://broadbandbreakfast.com/course/#bbb-offer) 

Every step of AI (training a model, running inference, an agent executing a task in real time) depends on physically moving data from a sensor, a phone, or a factory floor to a processor and back, fast enough that the answer still matters when it arrives. 

The model gets the attention, but whether the network can deliver that round trip with the speed and latency is the constraint most people outside this industry aren't watching closely enough.

### *The investment case is changing*

For most of the 5G era, the return on network investment was hard to defend. [PwC's analysis of 5G economics](https://www.pwc.com/us/en/tech-effect/emerging-tech/5g-monetization.html?ref=broadbandbreakfast.com) found operator return on assets running 1.5% to 4.5%, against 17% to 20%. Carriers funded infrastructure while other companies monetized. 

AI changes that math. For the first time in a decade, there's a workload that genuinely cannot run without dense, low-latency connectivity everywhere, not just at a handful of hyperscale campuses, and capital is starting to follow that logic. Nvidia's $1 billion investment in Nokia, paired with a partnership to build AI directly into radio access networks, signals that AI infrastructure money is moving into the network itself, not only the data centers behind it.

### *Fiber, wireless, satellite, and power have to move together*

Broadband isn't one network. It's fiber, terrestrial wireless, satellite, and the power and data center infrastructure behind it, and none of them solve the AI problem alone. 

Fiber and 5G handle density while satellite closes the geographic gaps neither can reach economically. SpaceX's IPO filing disclosed a direct-to-cell service already reaching 7.4 million devices a month through roughly 30 carrier partnerships. This is a capability that barely existed three years ago. 

Edge computing exists because the round trip to a distant data center is too slow for virtually real-time AI. But none of these systems matter without the infrastructure to power them.The [International Energy Agency](https://www.iea.org/reports/energy-and-ai/energy-demand-from-ai?ref=broadbandbreakfast.com) projects data center electricity demand will roughly double by 2030, with the U.S. alone adding enough new demand to rival the electricity used by all of heavy industry combined. A network with more bandwidth than its power supply can support is not more capable.

### *What 5G should have taught us*

Having watched the industry move through multiple generations of wireless technology, I don't think 5G's real lesson is that the technology underdelivered. The sequencing was the problem. The industry built capacity years ahead of a monetization plan and backfilled the business case afterward. 

AI is different in one respect that matters: the demand is already proven and paying. That removes the excuse for building blind, but it doesn't remove the risk of building the wrong thing. 

The 3GPP standards process is already shaping 6G, and the choices being locked in now about how AI gets embedded at the radio layer will be just as hard to unwind a decade from now.

### *Networks are starting to run themselves, faster than expected*

The most underappreciated shift is inside network operations. A network engineer today typically manages 50 cell sites. Agentic AI systems already in field trials are pushing that toward several hundred by handling fault detection and root-cause diagnosis before a human is looped in. 

[Google Cloud estimates agentic AI represents a $60 billion opportunity](https://appledoreresearch.com/wp-content/uploads/Reports/Appledore-Agentic-AI-in-Telecom-WP-June-2025-Revision.pdf?ref=broadbandbreakfast.com) for telecom operators, largely through the self-healing networks this shift points toward. The same math shows up in customer care, where the average inbound support call still costs providers upward of $7 for problems that increasingly don't require a person at all.

Some of the more interesting applications sit outside the core network entirely. AI-driven sensor networks combining cellular, satellite, and edge inference can now detect a wildfire within a minute of ignition, before it would register as smoke to a human observer. This is possible only because wireless, satellite, and edge processing were designed to work together rather than as separate systems solving separate problems.

### *Where the next decade of value sits*

The biggest opportunities aren't inside any single layer, they're in the seams between them. The orchestration software that routes traffic intelligently across fiber, wireless, and satellite, the edge inference that makes real-time decisions physically possible, and the power infrastructure that is telecom's tightest constraint. 

Whoever builds the intelligence that makes convergence invisible to the end user will capture more value than anyone building a single layer in isolation.

The industry already has every piece it needs, but what’s been missing is treating them as one system instead of four separate ones. Getting that right, before the next generation of standards locks in the wrong assumptions, is the real opportunity in front of broadband right now in the age of AI.

***Glenn Lurie joined Stormbreaker Ventures as a venture partner in August 2021, bringing with him over 30 years of experience in scaling and leading multi-billion-dollar businesses. His extensive background includes serving as president, CEO, and Board Member of Synchronoss Technologies and as president and CEO of AT&T’s mobility and consumer operations. This Expert Opinion is exclusive to Broadband Breakfast.***

***Broadband Breakfast accepts commentary from informed observers of the broadband scene. Please send pieces to*** [***commentary@breakfast.media***](mailto:commentary@breakfast.media)***. The views expressed in Expert Opinion pieces do not necessarily reflect the views of Broadband Breakfast and Breakfast Media LLC.***