Telcos are being asked to fund the next wave of AI and 5G innovation while already burdened by enormous debt and relentless infrastructure costs. By breaking down carrier finances with a simple lemonade-stand analogy and real 2025 operator numbers, Vish Nandlall argues that the problem isn’t a lack of ambition or vision, it’s basic economics.
People ask me why I’m skeptical of the bright shiny toys telecom is supposed to build. Edge AI. Network APIs. Private 5G everywhere. The pitch decks are beautiful.
I’m not negative. I understand infrastructure because I’ve lived through it. I want all of the pitches to succeed. I’m just doing the math.
Let me walk you through it the way I’d explain it to a smart friend who’s never run a P&L.
Start with a lemonade stand
You build a lemonade stand. It cost $1,000, and you borrowed $700 from your uncle. He charges 6% interest, so you owe him $42 per year, no matter what.
After paying for lemons and sugar this year, the stand throws off $100 of operating profit.
What can you actually do with that $100?
– $42 goes to your uncle. You don’t get a vote.
– The blender broke. The sign needs replacing. Keeping the stand competitive costs $55 a year in upkeep and small upgrades.
– That $3 dollars left is yours.
Now a better stand opens across the street. To stay competitive, your upkeep jumps from $55 to $60. Your $3 becomes negative.
You didn’t do anything wrong. The math just got harder.
This is telecom.
The actual numbers, from 2025 10-Ks
For every $100 of operating income, where does it go?
I’m using operating income as the reference point because it’s the cleanest measure of what the business itself earns. Comparing interest expense and capex against it shows the intensity of each. Precisely how much weight the network carries relative to what it produces.
T-Mobile (2025) – operating income $18.3bn
Interest $3.8bn → $21 per $100
Capex $10.0bn → $55 per $100
Verizon (2025) – operating income $29.5bn
Interest $6.6bn → $22 per $100
Capex $17.0bn → $58 per $100
AT&T (2025) – operating income $24.2bn
Interest $6.8bn → $28 per $100
Capex $20.8bn → $86 per $100
Look at AT&T’s row again. The network business, by itself, spends roughly as much rebuilding itself as it earns. The growth in the press release was largely propped up by a one-time DIRECTV gain.
This is not bad management. This is the math of being a network operator in 2025.
The principle
When interest + capex eats most of operating income, every new investment has to clear an unusually high bar. Money isn’t free, and there isn’t much of it left.
Let me make that concrete.
Suppose AT&T’s network team says: “We need $1bn for edge AI.”
The CFO doesn’t ask “is edge AI cool?” She asks: “What does that $1bn have to earn to be worth doing?”
The math, roughly:
– $1bn of new debt at ~6% = $60m per year in interest, forever.
– $1bn of GPU-heavy infrastructure depreciates over ~4 years. That’s $250m per year.
– Power, cooling, maintenance, software ops: call it $100m per year.
The project has to generate about $410m per year in new operating income just to break even. Not revenue. Operating income.
At AT&T’s 19% operating margin, that’s roughly $2.2bn of new annual revenue.
Now ask yourself: what edge AI use case, sold through a telecom go-to-market motion, plausibly does $2.2bn per year of new revenue in the next three years?
I don’t see one. Not yet.
LTE vs 5G
The standard pitch is: “5G is like LTE, just wait, the apps will come.”
This misreads what LTE actually did.
LTE didn’t make the internet faster. LTE made the smartphone into the internet. Before LTE, you used your phone to call people and check email. After LTE, your phone replaced your camera, your map, your wallet, your TV, your taxi dispatcher, and your bank teller. A whole new layer of behavior, and a whole new wave of operating income, landed on top of carriers.
5G is different. It added real capacity. It enabled fixed wireless access, which is genuinely a new revenue stream. It lowered cost-per-bit. Real benefits.
But it did not change what people do with their phones. The promised applications (AR/VR, network slicing for industry, ultra-low-latency consumer cases) either didn’t materialize at scale or didn’t need 5G specifically.
5G made existing traffic faster and cheaper for the carrier to carry. That is an asset-utilization story, not a growth story. It improves the margin on what already exists; it doesn’t add a new layer.
This is why every telecom deck now talks about fiber, convergence, cost reduction, and AI infrastructure. All asset-utilization stories. They’re squeezing more value from the network they already own, because there’s no new behavioral wave to ride.
So what about edge AI?
Same test. Always the same test.
For edge compute to make sense at a carrier, locality has to create more economic value than the cost of distributing the infrastructure to get it close to the user.
Centralized compute is cheaper per FLOP. Always. Hyperscalers get scale economies on power, cooling, real estate, staffing, and GPU utilization. A hyperscaler region runs its GPUs hot, 24/7, across millions of customers. A carrier edge site runs its GPUs for whoever happens to be nearby, whenever they happen to need them.
So edge AI only works if locality is worth a premium that exceeds the utilization penalty.
The use cases where this is true are real but narrow: ultra-low-latency industrial control, certain regulated data-residency workloads, some real-time vision applications. They are not the broad consumer AI market. They are not “ChatGPT but closer to you”. They are vertical, contract-driven, and slow to scale.
That’s fine. Just don’t model it like it’s a smartphone wave.
The lesson, stated plainly
The network can be strategically essential and still financially hard to own.
Every dollar a carrier spends on a new toy has to pass a brutal test: does this generate enough new operating income to cover its own depreciation, its own operating cost, and the interest on the debt that financed it. Specifically a balance sheet already carrying twenty years of spectrum, fiber, towers, and equipment?
If the answer is no, it doesn’t matter how exciting the technology is. The math wins.
When I push back on a shiny telecom pitch, this is what I’m doing. I’m not saying the technology is bad. I’m saying: show me where it lands in the financial table above. Show me the row where it pays for itself. Show me whose $100 it improves.
If you can’t, it’s not a strategy. It’s a slide.
Vish Nandlall is a leading technology strategist focused on the convergence of 6G, AI, and autonomous systems. As Founder and Lead Analyst of Vish Nandlall Consulting, he advises global operators, investors, and policymakers on the economics and architecture of next-generation networks. A former CTO at major telecom and cloud organizations, Vish has shaped industry roadmaps across 5G, edge computing, and AI infrastructure. His current work explores how intelligence, connectivity, and computation will fuse to define the 6G era.
