The first G with AI in the room

The core problem: Every network generation decision has been made half on data, half on hope: operators could measure tower capacity, but per-site returns were always a forecast, because deriving the answer requires systems that define the same subscriber differently. AI makes 6G the first G where you can itemize the CapEx decision, with the return on every site known before the money is spent—but only if the operator’s systems are hooked into a powerful ontology. To make an informed 6G decision, you need the Totogi Ontology, not a bigger business-case spreadsheet.

Every G decision in the history of this industry has been made the same way: one giant capital commitment, in advance, on faith. It happened with 3G, with 4G, and with 5G. It doesn’t have to happen again—but it might. Dell’Oro projects the industry will spend more than $500 billion in wireless CapEx during the first six years of the 6G cycle. That’s money spent chasing revenues that, by GSMA’s own account, have barely kept pace with inflation for the entire 5G era.

I predict this cycle will be different because 6G will be the first network generation decided with AI in the room. But there’s a big difference between AI being in the room and AI being able to answer the question the decision actually turns on, which is the ROI of the investment. So before you take comfort in having a copilot for the biggest check of the decade, ask a harder question: can YOUR AI see the money?

For almost every operator on earth, the answer today is no. But there’s one CEO of a Latin American mobile network operator (MNO) who’s using the Totogi Ontology to see the profit of every individual cell.

Here’s what he can see that you can’t.

Unit economics

What is the unit economics of a cell tower? Every tower has two numbers: how busy it is, and how much money it makes. This industry has always been able to see the first number. Capacity is measured site by site, in real time, with data that your network hands you every second of every day. The second number has never been anything but a forecast. Somebody builds a model of what someone, someday, might pay for the new generation. That model becomes the business case, and the business case becomes a decade of CapEx. The unit economics of the network—the actual profit per site—is the one number your systems have never been able to produce.

Half of every 5G business case was data. The other half was hope.

I want to be careful here, because this is a diagnosis, not a criticism. Nobody who signed a 5G business case got it wrong. When the per-site return is structurally unknowable, a portfolio-level bet made on faith is the rational move, and the analysis was as good as the data allowed. Every operator who lived through 5G can defend the call they made. But the excuse won’t stand for any network upgrade in the future.

The view from La Paz

There’s an MNO in Bolivia that can see both numbers for every cell in its network. It can see the traffic, like everyone can, and it can also see the money. It used to have exactly what you have: a few hundred systems that never spoke to each other. Now, because of the Totogi Ontology, it has one model of the business: every site and every cell, ranked by the money it makes, with the cost side netted out. When it evaluates a site upgrade, it knows the return before the first dollar leaves the building.

My favorite part is who uses the ontology. It’s the CEO’s tool. He checks it himself, whenever he wants. There’s no waiting for someone to assemble the quarterly deck. He opens the product, asks any question he wants, and instantly gets the answer about his business. Once you’ve seen an operator run capital allocation that way, a slide with a blended ROI number on it starts to look like what it is: a very confident guess.

Why this is hard

To see what’s missing, try to answer this question about your own network: how much money did one of your towers make last month?

Finding the answer is a genuinely hard task. Revenue doesn’t arrive per tower; subscribers touch dozens of sites in a month, so their revenue has to be attributed across every site they’ve used. The cost side is just as tangled, because fuel and rent are site-specific, but backhaul, spectrum, and core are shared. Getting from “what subscribers paid” to “what this tower earned” is real allocation work, with assumptions someone has to defend. The math is hard but doable; the problem is the inputs, which are scattered across hundreds of systems that don’t agree with each other.

The calculation needs a chain of data: which cells are on the tower, which subscribers used them, what those subscribers paid, and which costs belong to the site. Every link in that chain lives in a different system, and at every link, the words change. The network calls the subscriber an IMSI. Billing calls it an account. The CRM calls it a customer. You have at least three names for the same human being, but nothing in your estate knows it’s the same person. The data exists. The meaning doesn’t.

Historically, attempts at per-site economics have relied on consultants. The consultant team hand-reconciles the chain for a few months, runs the allocation once, and presents a snapshot. The math in those studies is usually fine. But the reconciliation was manual, so it’s too expensive to re-run, and the estate kept moving while the team worked, so the picture was aging before the binder was printed. It’s hard math, done once, on stale inputs. A study like that is an annual snapshot, and you can’t run capital allocation on a snapshot.

AI changed the math. This is exactly what we built the Totogi Ontology to be: a map of your telco, drawn on TM Forum’s own standards, that rebuilds itself every night. AI reads your systems, learns what every field means, and never stops reading, so the system stays correct because it never sleeps. And watch what that does to the hard calculation. Once IMSI, account, and customer resolve to one subscriber and the chain connects end to end, per-site economics stops being a study and becomes a computation. You can argue about the attribution assumptions, change them, and re-run it whenever you want. That’s what the CEO in La Paz is actually looking at: the same hard math everyone else does as a study, but running continuously on inputs that are finally alive. And the network decision is just the first question he’ll ask the system, because the same map can now answer any questions about the business. It can find the at-risk subscribers worth saving, unknown revenue leaking between provisioning and billing, and the promotions that actually made money, with no new project.

What a G decision becomes

This changes what a G decision is. 5G had an alibi: your systems couldn’t see the money, so nobody could have decided it differently. 6G has no excuse, because for the first time in the history of this industry, you can know the return before you spend a dollar on any site.

5G was one lump-sum check, written in advance, on faith. 6G can be thousands of checks, one per site, each with its own receipt, written where the return shows up first. That’s itemized CapEx. Upgrade the sites that earn it this year, let traffic prove out the marginal ones, and skip the ones that will never pay back. Everything else AI promises this industry is efficiency, like running the network you already built cheaper. With the ontology, you can know precisely which parts of the network are worth investing in and target your CapEx spend accordingly. That’s the bigger prize. Spending blind used to be a requirement. Now it’s a choice. So before you write the biggest check of this decade, make sure your AI can see the money.

Because this time, it can.

Recent Posts

Get my FREE insider newsletter, delivered every two weeks, with curated content to help telco execs across the globe move to the public cloud.