Sebastian’s Substack

Sebastian’s Substack

Share this post

Sebastian’s Substack
Sebastian’s Substack
6G Capex: A Data-Driven Forecast Without the Hype

6G Capex: A Data-Driven Forecast Without the Hype

Three independent models, grounded in public data and two decades of telecom experience, to estimate the real cost of building 6G.

Sebastian Barros's avatar
Sebastian Barros
Aug 13, 2025
∙ Paid
1

Share this post

Sebastian’s Substack
Sebastian’s Substack
6G Capex: A Data-Driven Forecast Without the Hype
1
Share

The hardest part of forecasting the future is that it hasn’t happened yet. We have to make decisions with incomplete data, delayed disclosures, and forecasts that often carry more marketing than mathematics. That’s why this study is built in a way I can defend to a boardroom or a regulator: start with verified numbers for the wireless industry today, isolate the 5G portion as a historical benchmark, and then use that base to model 6G through three independent lenses: history, technology specifications, and traffic demand.

This is not a vendor pitch and not a speculative whiteboard sketch. It’s a data-grounded exercise using the best public datasets we have, tested against what I’ve seen in two decades of telco investment cycles. By the end, you’ll see three different cost curves for 6G, each derived from a separate logic, and you can decide which feels most probable.

The wireless baseline first. GSMA’s latest Mobile Economy report shows global mobile operator Capex at an annual run rate of roughly $180–190 billion through the 2020s, front-loaded in the early 5G build years and tapering toward the end of the decade. Dell’Oro’s wireless capital intensity data tells the same story: a peak near 19% of revenue during 5G’s surge, trending down to 12–13% by 2027.

Now the 5G subset. GSMA estimates that about 92% of mobile Capex from 2023 to 2030 is tied to 5G deployment and evolution, that’s roughly $1.38 trillion over this period. That’s purely network Capex. Spectrum is a separate cost center entirely. For example, the U.S. C-band auction alone cleared $81 billion, about half the annual global mobile network Capex in one market and one transaction.

This is the floor we’ll build on. From here, we’ll scale forward into 6G not by guessing a single number, but by applying three independent forecasting lenses and then triangulating them into credible scenarios.


The Three Lenses of 6G Capex Forecasting

Since 2019, we’ve had a clear view of the global wireless capex envelope, the portion allocated specifically to 5G, and the reality that most operator investment this decade is still anchored in finishing the 5G build. That’s the known baseline. The challenge comes when you try to pin a number on 6G: there is no single “correct” figure. The outcome depends entirely on the lens you choose.

I’m going to run three models, each starting from a different premise and carrying its own bias. One extrapolates directly from the 5G investment curve, another builds from the technical specifications under discussion for 6G, and the third starts with projected traffic demand and works backward to the physical network required. None is definitive on its own, but together they bracket the plausible range of outcomes.

Lens 1: Historical Extrapolation (Momentum Model)

This is the most conservative view, and the one that boards instinctively reach for. I take the cumulative CapEx for 5G from 2018 to 2024, about $500–600 B, excluding spectrum, and project it forward. I adjust for inflation, financing costs, and the likelihood that 6G refresh cycles will be slightly longer than 5G because operators will want more time to amortize assets.

If you simply run this math with an annual mobile CapEx envelope in the $180–200 B range and assign a similar generational share to 6G as 5G had, you land in the below one trillion figure globally over a decade. This is the “history repeats” scenario: it assumes technology ambitions are roughly the same and demand curves don’t surprise us. It’s tidy, but it’s also the least sensitive to the fact that technology roadmaps have a way of outpacing our spreadsheets.

Lens 2: Specification-Driven Model (Technology-Led)

Here, I let the engineers set the ambition, and then I cost it out. The ITU IMT-2030 draft goals are bold: air-interface latency in the 0.1–1 ms range (with some visions pushing below 0.1 ms), ultra-high reliability around 99.9999 %, and native integration of AI and sensing capabilities. While the exact throughput multiplier versus 5G isn’t yet pinned down, the discussion includes a new upper-6 GHz capacity band (6.425–7.125 GHz) as a prime candidate for wide channels with macro-class propagation, very much the 6G analogue of 5G’s mid-band. That choice alone has enormous capex implications, because like 5G’s mid-band, it would drive a large-scale radio and transport upgrade cycle.

In this analysis, I’m not assigning much weight to the higher-frequency sub-THz ranges with very short propagation, since they would require dense urban site builds and are likely to follow the 5G mmWave path: deployed only for niche, ultra-capacity scenarios. THz bands will not materially shift the global investment curve. By contrast, the new upper-6 GHz band stands out as the probable “6G mid-band,” with a cost profile similar to 5G’s 3.5 GHz layer and the potential to drive a major RAN and transport upgrade cycle. Beyond spectrum, the architecture calls for AI-capable edge nodes to process workloads locally, ultra-low-latency transport infrastructure, and, if satellite integration becomes a standard element, orbital gateways and interconnects. In financial terms, these translate into tangible units: radios, antennas, edge servers, and upgraded optical backhaul.

Vendor roadmaps today put a fully kitted urban 6G site well above the per-site cost of a mid-band 5G macro, even allowing for cost declines. Multiply by the number of new sites implied by the coverage model and you start to see a total that can quickly overshoot the historical extrapolation, especially if non-terrestrial components are included.

Lens 3: Traffic-Demand Model (Demand-Led)

This one starts with the user, not the network. Ericsson’s Mobility Report and Cisco’s historical VNI data give us a long-term view of mobile data growth. If we extend those curves to 2040 and overlay the bandwidth demands of XR, fully autonomous systems, industrial automation, and immersive media, the per-user consumption assumptions get extreme in the later years.

From there, I back-calculate the capacity required in bps per square kilometre, then derive the spectrum, site density, and transport needs to carry it. I apply cost-per-bit improvement factors, because radio and optical tech will get more efficient — but not infinitely so. This model is highly sensitive to adoption rates: if industrial and machine traffic ramps faster than human demand, it drives Capex harder than any consumer app ever did.

These three models are deliberately different. The momentum model is conservative, rooted in financial history. The spec-driven model is ambitious, driven by what engineers and standards bodies put on paper. The demand-led model is volatile, bending upward or downward based on how society uses the network.

In Part 3, I’ll triangulate them, side-by-side, to show just how wide the plausible 6G Capex range can be, and what that means for funding models.


Pause here: Capex is Physics

You can love or hate 5G or 6G. You can argue about business models and vendor marketing. None of that changes the constraint that drives spend. Networks move bits through space and glass, which means radios in the air and fiber in the ground. Traffic keeps growing, and physics does not negotiate.

And the anchor is both Traffic and Hardware modernization. Total monthly global mobile network data traffic reached about 172 exabytes in Q1 2025, up 19% year over year. Video alone is roughly three-quarters of that load. At the end of 2024, 5G already carried about 35% of mobile traffic. These are not opinions, but come from the latest published measurements from the Ericsson mobility report, which I consider a reputable source.

What follows is simple engineering economics. Area capacity is the product of three things: bandwidth, spectral efficiency, and cell density. If traffic rises, you must increase one or more of those terms. More bandwidth means more spectrum and wider channels. Higher spectral efficiency means better radios, smarter antennas, and tighter scheduling, which in practice often needs cleaner spectrum and denser grids. More density means more sites, more backhaul, more power, and more civil work. None of these comes for free.

This is why mobile CapEx continues regardless of branding. Even if you erased the label “6G,” operators would still need to add spectrum where available, extend fiber backhaul, and multiply sites in hot zones to keep ahead of load. The slope has moderated, which helps, but the rule holds. More cars, more lanes. You can debate the name of the highway. You cannot avoid building it.

With that physics baseline in place, we can now put numbers on 6G using the three lenses and see where a rational Capex range lands.


Keep reading with a 7-day free trial

Subscribe to Sebastian’s Substack to keep reading this post and get 7 days of free access to the full post archives.

Already a paid subscriber? Sign in
© 2025 Sebastian Barros
Privacy ∙ Terms ∙ Collection notice
Start writingGet the app
Substack is the home for great culture

Share