Shotwell Put the Big Three at $600 Billion a Year. Their Revenue Is $352 Billion, and the Traffic Math Says They Survive.

On SpaceX's first earnings call as a public company, on August 4, Gwynne Shotwell was asked about Starlink Mobile and answered by sizing the target:

"The big three in the U.S., AT&T, Verizon, and T-Mobile, roughly between them, $600 billion a year. I expect us to be able to acquire quite a few of their customers. Our service will be better. We will eliminate dead zones, leveraging, basically, satellites in orbit."

AT&T, Verizon and T-Mobile all fell after hours on the quote. We covered the print itself separately; this piece is about the claim, because two of its numbers are checkable and the third — how much of the internet Starlink ends up carrying — is not, but can be bounded.

The $600 billion is not a revenue line

Revenue figures are the companies' reported fiscal 2025 totals; market caps are as of the August 4, 2026 close and will have moved since.

Combined revenue is $352.1B. Shotwell's $600B is 1.70× that. It is, however, within 10% of the three companies' combined market value — so the figure is roughly the size of the equity a successful Starlink Mobile would be attacking, not the size of the revenue pool it could win.

Narrow it to what Starlink Mobile actually competes for and the pool shrinks again. In Q2 2026 the three reported $59.8B of service revenue between them — Verizon's $23.4B of mobility and broadband service revenue, AT&T's $17.41B of wireless service revenue, and T-Mobile's $19.0B of total service revenue. That is a $239B annual run rate. Wireless service revenue, not consolidated revenue, is the line a satellite phone service can take.

What the carriers actually look like going in

This is not a sector being disrupted while it bleeds. Q2 2026, all three:

Verizon's $13.7B was a record. AT&T added 432,000 postpaid phone customers against a 341,000 consensus, with churn at 0.86% and postpaid phone ARPU of $58.01. T-Mobile's postpaid service revenue grew 13%. All three are growing EBITDA faster than revenue, which is the signature of a mature oligopoly with pricing power, not one about to be taken apart.

That matters for the shape of the threat. A business with 0.86% monthly churn and rising ARPU does not get destroyed by a new entrant; it gets its margin structure renegotiated.

Musk's argument is about machines, not people

The more consequential claim on the call was not Shotwell's. Musk argued that the demand curve is about to change kind, not degree:

"A human consuming perhaps about a few hundred bits per second… computers could easily do billions of bits per second continuously over a 24-hour period."

And then the conclusion the whole thesis rests on:

"Really, I think Starlink is the only thing that can actually service that bandwidth."

Paired with a claim about eventual share:

"It's not out of the question that at some point, Starlink will deliver a majority of the world's internet… less than 10 years."

The engineering behind it is disclosed and specific. Musk put the V3 satellite at "about an order of magnitude more capable" than V2, and the constellation at "a roughly two order of magnitude increase in delivered bandwidth" — 10× per satellite times roughly 10× the satellites. He put the critical mass at "about 1,000 V3 satellites around Q2 2027." Shotwell said the Starlink Mobile satellites "start to fly next year" with service "end of next year," using the 65 MHz acquired from EchoStar.

Two things about that 100×. The 10× per satellite is a published hardware figure — V3 is specified at 1 Tbps of downlink against roughly 100 Gbps for V2. The 10× satellite count is not: it implies a constellation on the order of 78,000 satellites against roughly 7,800 today, which is well beyond any current authorization. The capacity ambition is real; the schedule for the second 10× is a regulatory question, not an engineering one.

Sizing the base: what share does Starlink carry today?

There is no published figure for Starlink's share of global internet traffic, so it has to be built. The inputs, and the arithmetic, so you can disagree with a specific step:

• ITU puts 2025 fixed broadband traffic at 7.3 ZB and mobile broadband traffic at 1.5 ZB — 8.8 ZB in total. Fixed has grown ~16% a year since 2021, mobile ~19%.
• There are roughly 1.4 billion fixed broadband subscriptions worldwide. That is 5.2 TB per subscription per year, or about 435 GB a month.
• Starlink ended Q2 2026 with 12 million subscribers, having added "net more than 1.7 million" in the quarter — Shotwell called it "our best quarter of new customers to date."

Twelve million subscriptions at the global per-subscription average is 0.063 ZB a year, which is 0.71% of the 8.8 ZB total. Starlink's mix skews rural and heavy-usage, so the honest range is 0.7% to 1.2%. Call it under one percent.

Now the interesting part. The constellation's aggregate capacity is estimated at roughly 450 Tbps. Run flat out for a year that is 1.78 ZB — 20% of all global internet traffic. At a realistic 20% average utilisation it is 0.36 ZB, or 4%.

So Starlink's capacity today could carry something like four percent of the world's internet, and its subscriber base implies it carries under one. Capacity is not the binding constraint. Customers are. That is the single most important fact for anyone modelling the next five years, and it cuts both ways: it means the 100× buildout does not automatically produce 100× the traffic, and it means the near-term ceiling is demand and ground segment, not satellites.

Three scenarios to 2031

The denominator first. Extending ITU's own growth rates six years from 2025: fixed 7.3 ZB × 1.16⁶ = 17.8 ZB, mobile 1.5 ZB × 1.19⁶ = 4.3 ZB. About 22 ZB of global internet traffic in 2031 on trend.

These are our estimates, not forecasts from anyone's guidance.

Hypothesis 1 — Subscriber grind (Starlink ≈ 1.5% of traffic)

Starlink compounds subscribers at roughly 27% a year, 12M to 40M by 2031. V3 lifts average consumption from 5.2 to 8 TB per subscriber per year. That is 0.32 ZB against a 22 ZB market: 1.5%.

Starlink Mobile works, but as a coverage layer — dead zones, messaging, low-bandwidth fallback. The per-beam physics is the constraint: measured direct-to-cell throughput today is around 3–4 Mbps per beam shared across roughly 30–40 concurrent users, against terrestrial LTE managing ~37.5 Mbps peak on a single 5 MHz channel.

Carrier impact: the rural coverage premium disappears and satellite fallback becomes table stakes. Churn rises 10–20bps, ARPU growth stalls. Combined service revenue growth goes from mid-single-digit to roughly flat; EBITDA margin gives up 100–200bps. Nobody exits.

Hypothesis 2 — V3 lands and Starlink becomes the default for everything outside the metros (≈ 4–5%)

The Q2 2027 critical mass arrives roughly on time. By 2031 Starlink has 40M consumer subscriptions plus ~20M enterprise, government, backhaul, maritime and aviation connections at much higher average draw — call the blend 15 TB a year. That is 0.90 ZB, or 4.1% of the 22 ZB market.

The shift here is not consumer phones. It is tower backhaul, enterprise WAN, and the ~40% of the planet's landmass where terrestrial economics never worked. SpaceX's Q2 numbers already point this way: enterprise and government connectivity revenue was $1,806M, up 108% year over year, growing far faster than the $2,485M consumer line.

Carrier impact: they lose the wholesale and rural-premium business, which is disproportionately high margin. Combined service revenue flat to slightly down; EBITDA margin gives up 300–500bps. Capex relief from not building out marginal rural coverage offsets part of it. Still nobody exits — dense urban traffic is a spectrum-and-cell-density problem that satellites are the wrong instrument for.

Hypothesis 3 — The Musk case: machine traffic breaks the trend line (≈ 12–14% of a much bigger number)

If AI inference, robots and autonomous vehicles do what Musk describes, the denominator itself inflates. Suppose machine traffic pushes 2031 total traffic to 35 ZB rather than 22 ZB — 60% above trend. Suppose Starlink, with the full 100× buildout, carries 4.5 ZB. That is 12.9% of global internet traffic, roughly eighteen times today's share.

The capacity supports it easily: 45 Pbps of nameplate at 30% utilisation is 53 ZB a year. Carrying 4.5 ZB needs about 2.5% utilisation of the full buildout. In this scenario capacity is so far from binding that the constraint becomes landing rights, ground stations and national regulators.

Carrier impact — and this is the part the after-hours selling got wrong: even here the Big Three survive. Starlink's 65 MHz nationwide sits against AT&T's 375 MHz of population-weighted spectrum and Verizon's 279 MHz, with T-Mobile holding more than either. More fundamentally, a terrestrial network reuses its spectrum thousands of times across a metro's cell grid; a satellite beam covers tens of kilometres and reuses nothing within it. Dense urban capacity is a physics problem satellites do not solve at any constellation size.

What the carriers lose in this scenario is the growth option, not the base. They become utilities carrying the traffic that has to be carried locally, on flat revenue and structurally lower margin, while the incremental machine traffic — which is disproportionately remote, mobile and unlocated — routes to Starlink. Margin shrinks; the companies stay.

What this means for the Rule of 40

SpaceX does not have a Rule of 40 score on this site, and cannot yet. The score needs a quarterly free-cash-flow figure, and SpaceX publishes operating cash flow only year to date — with no pre-IPO quarterly report to difference against. Six-month free cash flow on our definition was −$25,010M, on $28,476M of half-year capex.

That is the honest frame for every number above. Connectivity is a genuinely excellent business inside SpaceX — $4,291M of revenue in Q2, up 66%, throwing off $1,656M of operating income on $1,367M of segment capex, a 38.6% operating margin. It is funding an AI segment that consumed $15,828M of capex in a single quarter and lost $1,257M.

The carriers, meanwhile, convert. Verizon's record $13.7B quarterly EBITDA is cash that exists.

So the trade being priced on August 4 was: how much of a $239B annual service-revenue pool moves, and how fast. Our three scenarios say the answer over five years is some of the margin, very little of the base — and that the more dramatic Starlink traffic scenarios are the ones where the carriers get squeezed rather than replaced.

A 5% price concession across that $239B pool costs roughly $12B a year, most of it straight out of EBITDA since network cost is largely fixed. Against the $26.0B of adjusted EBITDA that Verizon and AT&T alone produced in a single quarter, that is a real wound and a survivable one. It is also, on our reading, the most likely outcome.

The number to watch is not Starlink's subscriber count. It is the gap between the 4% of world traffic its capacity could already carry and the under-1% it does.

Sources: SpaceX Q2 2026 shareholder letter and earnings call, August 4, 2026; company Q2 2026 results for Verizon, AT&T and T-Mobile; ITU Facts and Figures internet traffic statistics; FCC spectrum filings. Scenario figures are our estimates and are labelled as such.