Comcast Analysts Are Afraid of the Wrong Thing

Yesterday Comcast announced it will split in two: a connectivity company built around Xfinity broadband and wireless, and a separate NBCUniversal holding the studios, Peacock, and Sky. It follows the Versant spinoff from January, which carved off the cable channels. Strip away the media businesses and what remains is close to a pure-play broadband operator, with a new CEO whose entire mandate is to defend and grow that pipe.

Several commentators read the move as a recognition of something specific: that Comcast's shareholders want a focused broadband company, and that the threat it most needs to focus on is Starlink. The cable business, the reasoning goes, is under siege from satellite internet, and a stand-alone connectivity company is better positioned to fight that war.

I think the diagnosis is half right and the named enemy is wrong. Comcast has real threats. Starlink, for its core urban and suburban business, is not the one that should keep the new CEO awake. And I want to walk through why, because the reason is not opinion. It is arithmetic, and I have watched this particular arithmetic break a company before.

A dream I watched meet its math

In the late 1990s I stood up Dotcast's Seattle office around a Boeing skunkworks team that had been working the original satellite-broadband problem. The program they came from was Teledesic, the Craig McCaw and Bill Gates venture that promised fiber-like internet from a constellation of hundreds of low-orbiting satellites. It raised the ambition, and a good deal of the money, that Starlink would later make real. It never flew a commercial network, and suspended its satellite construction in 2002.

Popular history says Teledesic died because satellites and launches were too expensive, and since modern launch costs have since collapsed, the old objection is dead. That was true in that moment, but it was just the first operand in a series of multiples that doomed the project. The launch cost factor is not the lesson the engineers I hired carried with them, and their lesson is the one that matters today.

Their telling went like this. The system worked in the base case. You could close the link budget for a single satellite over a single cell and show a beautiful demo. The trouble appeared when you stacked the business case on top of the engineering. To hit the company's financial projections you needed a specific routing capacity in each satellite, times a specific number of successful launches, times a specific uptime across the fleet, times a specific uplink capacity, and that is only the front of the list. Ground-station coverage, spectrum, terminal cost, install rate, churn: every one of them another operand. Each number was a stretch but plausible on its own. Multiplied together, against the revenue the model required, the product became infeasible. The base case held. The financial case, the only one that pays for the constellation, fell apart under its own compounding.

That is a different and more durable lesson than “space is expensive.” It says the killer was never any single factor that a cheaper rocket fixes. It was the multiplication of many stretch factors against the demand the business needed to serve. Cut the cost of one term and the structure still has to clear all the others, simultaneously, at scale.

Starlink genuinely solved some of those terms. This is not a “history repeats” story, and I want to be fair to my former engineers, who went on to help prove that pieces of the dream were buildable. Reusable rockets crushed launch cost. Mass production crushed per-satellite cost. Cadence and uptime are real now in a way they never were in 1999. Three or four terms in that product have genuinely come down.

But one term went the other way — and it is the one that decides whether Starlink can take Comcast's customers.

The term that got worse

Teledesic was modeling demand in the dial-up era. A 720-megabit downlink shared across a region looked like an ocean of capacity when the customer on the other end was pulling email and web pages over a modem. Today's broadband household streams multiple simultaneous 4K video feeds, backs up to the cloud, runs video calls, and increasingly trains the family's gadgets on the network all day. Per-user demand has grown by orders of magnitude, and it keeps growing.

To put the shift in scale: a dial-up modem in 2000 topped out around 56 kilobits per second, and a typical connected household consumed on the order of tens of megabytes a month. A single 4K stream today runs at roughly 25 megabits per second sustained, and a connected home routinely moves hundreds of gigabytes to several terabytes a month. That is a jump of four to five orders of magnitude in per-household appetite over the same twenty-five years in which satellites got cheaper. The capacity models of Teledesic's era were built for a hunger that no longer exists.

So while the supply side of the satellite equation improved dramatically, the demand side ran just as hard in the opposite direction. The tech scaled. The appetite scaled with it. That is the quiet reason satellite broadband keeps looking feasible in the base case and keeps struggling against the financial case: the very progress that made satellites cheaper also made the customers hungrier.

And there is a second problem the cheaper rocket does nothing about, because it was never a cost problem in the first place.

Where the people are

A satellite constellation spreads its capacity over geography. A Starlink satellite passing over a region delivers a fixed pool of bandwidth to everyone in its beams, and that pool is shared. The capacity is distributed roughly evenly across the Earth's surface, over ocean and desert and farmland and city alike.

People are not distributed evenly. They cluster in cities. Roughly 80 percent of Americans live on about 3 percent of the land. Comcast's subscribers (and infrastructure!) are concentrated in exactly those dense metros and inner suburbs, which is precisely where a constellation's per-area capacity is most overwhelmed and least able to help. This is the detail the Starlink-fear commentary skips: the satellite threat is real at the edges of Comcast's map and weakest at its center.

You can see this in Starlink's own behavior. In dense and growing markets across the US, the company has repeatedly hit capacity, paused new residential signups, and reimposed waitlists. Congestion in busy cells already drops evening speeds. This is not a temporary hiccup that more launches smooth away. It is the architecture revealing itself: capacity sits where the satellites are, demand sits where the people are, and the two do not line up.

The clearest statement of this came from inside SpaceX. At Mobile World Congress this year, the head of Starlink told the audience that satellite “cannot provide the data density that terrestrial networks” deliver, and positioned the service as complementary to terrestrial systems rather than a replacement. When the people building the constellation tell you it is not designed to win the dense market, that is worth believing.

The number that would have to be true

Let me put rough figures on it, and I will keep them deliberately directional rather than falsely precise, because the order of magnitude is the whole point.

Starlink today carries something like 450 to 600 terabits per second of total network capacity, serving around 10 million subscribers across more than 150 countries. That is the entire planet's worth of capacity.

Now let's define a “material” threat to the new Comcast as capturing 10 percent of its roughly 32 million broadband subscribers. I pick 10 percent deliberately. A pure-play broadband CEO who is handed the company and then loses a tenth of the base to a satellite is not a CEO who survives the next proxy season. That is a resignation-level event, and it is the scale at which “Starlink fear” would actually mean something. So it is the right number to test.

Serving 3.2 million subscribers at a genuinely cable-competitive level, on the order of 100 megabits per second available in the busy evening hours, requires roughly 300 terabits per second of capacity dedicated to that slice alone. That is most of Starlink's entire global network, redirected to one tenth of one American cable company's customers.

But the capacity cannot be redirected, because it is spread over geography, not pointed at demand. The fraction of Starlink's capacity physically passing over America's urban land at any moment, the few percent of the country where those customers actually live, is on the order of a single terabit per second. To serve the target at parity you would need that urban-concentrated capacity to grow by something like two orders of magnitude. Not 10 percent more satellites. Not double. Roughly a hundredfold increase in capacity concentrated over cities — in a system whose capacity inherently refuses to concentrate.

And a hundredfold is the floor, not the ceiling, because it assumes today's demand frozen in place. It is not. Per-household broadband consumption has grown on the order of 20 to 30 percent a year for a decade, and shows no sign of stopping. Any buildout large enough to close a hundredfold gap takes years, and over those years the parity target itself keeps climbing. Close the gap to today's demand in a decade, and demand has moved six to fourteen times higher in the meantime. The honest number is not a hundredfold; it is a hundredfold chasing a target that keeps receding.

The next-generation V3 satellites raise per-satellite capacity by about tenfold, which is a real and impressive leap. It still leaves you needing many times the entire current fleet to close the urban gap, and it does nothing about the underlying geometry: every satellite-second spent over the Pacific, the Rockies, or the wheat belt is capacity that cannot be sold to a household in Chicago. A uniform constellation cannot aim itself at Comcast's map.

There are engineering answers on the horizon, and I want to credit them rather than wave them away. Spot beams can concentrate more capacity over high-demand cells. Tighter frequency reuse, more beams per satellite, denser orbital shells over populated latitudes, and the move to inter-satellite laser mesh all push in the right direction, and SpaceX is pursuing all of them. The honest engineering statement is that the urban-density problem is not forbidden by physics. But look at the multiple again. Closing a hundredfold capacity gap over cities is not a roadmap item; it is a sequence of breakthroughs stacked on top of each other, each one a stretch, multiplied together against a financial target. There is science to be created before there is a business risk to Comcast, and the distance between “conceivable in a lab” and “a threat a broadband CEO plans around this decade” is the entire point. A competitor you might face after three or four compounding research wins is not a competitor you reorganize your company around today.

This is the Teledesic stack again, with the demand term turned up. The base case is gorgeous and an amazing business: a Starlink dish in a Montana valley genuinely outperforms anything the local telco will ever string up a mountain. The financial case against Comcast, the one where you take a tenth of a dense urban base at parity speeds, multiplies out to a constellation that does not exist and that the architecture resists building.

What Comcast should actually fear

None of this means the new Comcast is safe. It means the danger is wearing a different uniform, and the commentary fixated on satellites is looking past it.

The genuine threat to cable's dense-market base is fixed wireless: T-Mobile and Verizon delivering home internet over 5G, and Dish pushing hard into the same space. Fixed wireless lives exactly where Comcast lives, in the cities and suburbs, riding terrestrial spectrum and towers that put capacity where the people are. It is already taking subscribers at the margin, and it does not have a geographic mismatch problem, because it is terrestrial. That is the competitor that can plausibly reach 10 percent.

There is an irony worth naming. Comcast itself, and Charter, both have their own fixed-wireless plays, and there is recurring talk of the two combining. I know both companies and their executives well enough to be skeptical that a merger would be the clean fit the speculation imagines, but the strategic logic they are all chasing is sound: the fight that matters is terrestrial, for the dense markets, on spectrum and towers, not in orbit.

So the new CEO's instinct to build a focused connectivity company may well be right. The restructuring could be exactly the correct move. But if the threat board in his first all-hands has Starlink at the top, the company is bracing for the wrong war, and the analysts cheering the satellite narrative are helping it misaim. Starlink will keep winning the customers Comcast was never going to serve profitably anyway, the rural and remote households at the end of the line. The customers Comcast actually depends on live in precisely the places a constellation serves worst.

I learned to distrust the beautiful base case from a team of engineers who had already watched it betray them once, twenty-five years ago. The rockets got cheap. The satellites got good. The demand got hungrier and the people stayed in the cities. Three of those four changed. The one that decides this fight did not.

There is a second half to this worth writing: the fixed-wireless threat itself, how far 5G home internet can actually scale into Comcast's base before it hits its own capacity walls, and whether the Comcast-Charter merger talk survives contact with the engineering and the antitrust math. If there's interest, I'll take that one next.