The system is overhyped. The numbers do not close. When a CEO with a controlling stake in a satellite network claims his constellation will carry 50% of global internet traffic, and a potential investor projects $1 trillion in revenue, the prudent response is not to extrapolate growth. It is to run the capacity calculations and examine the unit economics. Based on my audit experience—where every claim must be attached to a verifiable dependency—I have dissected the Starlink thesis. The conclusion is unambiguous: the forecast is a design flaw, not a technical roadmap. Silence before the breach.
Context: The Claim and the Conflict of Interest
In August of this year, a podcast conversation between David Friedberg and Elon Musk produced a flurry of headlines. Friedberg, a potential investor, suggested Starlink could reach $400 billion in annual revenue. Musk responded that there are no obvious obstacles to carrying 50% of all internet traffic, driven by a "few orders of magnitude" increase in data transmission from AI and robotics. These statements were filtered through an industry news brief, stripped of technical detail.
The source set is problematic. Musk is the principal of SpaceX. Friedberg has an incentive to talk up an investment. No independent engineering review accompanied the forecast. In my line of work—auditing DeFi protocols where a single unchecked loop can drain a vault—we treat such claims as hypotheses to be stress-tested, not statements to be repeated. Code is law, until it isn't. The same holds for physics.
Core: Capacity, Unit Economics, and Growth Ceilings
The first test is physical. Global internet traffic is projected to reach 396 exabytes per month by 2027, with a peak throughput of approximately 1.1 PB/s. If Starlink were to carry 50% of that peak, it would need to sustain 550 TB/s of downlink and uplink capacity. Current V2 Mini satellites deliver between 60 and 100 Gbps per satellite. To match 550 TB/s, the constellation would require between 44,000 and 73,000 satellites. That upper bound exceeds the total constellation of 42,000 that SpaceX has filed with the ITU. The lower bound assumes perfect efficiency, which does not exist in radio frequency transmission.
This is not a marginal gap. It is a physical scale barrier. The company has launched approximately 7,000 satellites to date, with a dispatch rate that has taken six years to reach this level. Reaching even the lower bound of 44,000 would require building and launching more than 37,000 additional satellites, at a time when spectrum coordination and orbital debris mitigation are already binding constraints. The claim of "no obvious obstacles" deliberately ignores the frequency coordination problem. Every orbital shell requires international licensing. Every ground station needs backhaul capacity. These are not engineering nuisances; they are regulatory and physical ceilings.
The second test is economic. Friedberg's $400 billion revenue figure implies a free cash flow of $30 billion, which suggests a 75% FCF margin. The telecom industry, where Starlink operates, has median FCF margins of 10–20%. A 75% margin assumes that the constellation is fully depreciated and requires no further expansion. But the 50% traffic target demands the opposite: continuous capital expenditure to replace satellites every five to seven years and to build out ground infrastructure. This is not a one-time cost; it is an annual sink. The satellites themselves are the most significant line item. I have run the model: at a fleet of 44,000 satellites, each costing approximately $250,000 to manufacture and $500,000 to launch, the replacement cycle alone would consume $165 billion per year. That is nearly the entire revenue figure. The FCF projection is not optimistic—it is nonsensical.
Let me be precise with the unit economics. At the current average revenue per user of roughly $100–120 per month, Starlink would need between 30 million and 35 million subscribers to hit $400 billion in annual revenue. Today, the company has around 6 million. To reach 30 million, Starlink must grow fivefold while sustaining ARPU. That growth is possible in theory, but it collides with the user base needed for the $1 trillion figure. At $1 trillion, Starlink would require 400 to 600 million users, including high-ARPU enterprise contracts. That is not a satellite internet user pool; that is the entire global telecom market. The global telecom services market is $2–2.5 trillion in total revenue. A $1 trillion Starlink would mean seizing 40–50% of all global communication services revenue. No company in any infrastructure sector has ever achieved that share from a new entrant position, especially when the incumbent fiber and 5G networks are expanding coverage.
The third test is user segmentation. Starlink's growth story is not uniform. The service is sticky for users with no other option: remote communities, ships, aircraft, disaster zones. These users have a strong incentive to stay, and they generate word-of-mouth advocacy. But the tail of the market—urban and suburban users with fiber or 5G alternatives—views Starlink as a premium-priced compromise. The high-ARPU customers that would be needed to support the revenue model are concentrated in maritime and aviation. There are approximately 100,000 commercial ships and 25,000 commercial aircraft globally. At conservative ARPU levels, that yields a few billion dollars in revenue, not hundreds of billions. The bulk of growth must therefore come from the consumer segment, which is exactly where competition from terrestrial networks is most intense.
The growth curve is already decelerating. Starlink took five to six years to reach 6 million subscribers, and the annual growth rate has slowed to a range of 30–50%. To support the revenue projection, Starlink would need to accelerate growth severalfold without reducing price. This is a classic extrapolation error. The growth curve is a physical integration of user acquisition, which depends on how quickly the company can manufacture terminals, install them, and provision satellite capacity. The bottleneck is not demand; it is supply chain and manufacturing. SpaceX has stretched its manufacturing capacity to its limits, and scaling to tens of millions of terminals per year would require a level of industrialization comparable to the smartphone industry. No company in the telecom hardware space has achieved that from a near-zero base in less than a decade.
The fourth dimension is the Direct-to-Device pivot. This is the least discussed but most structurally significant shift in Starlink's business model. Through strategic partnerships with T-Mobile, KDDI, Rogers, and other mobile operators, Starlink is pivoting from a consumer-branded ISP to a wholesale infrastructure layer. Mobile users will be able to connect to Starlink satellites when outside terrestrial coverage, with Starlink selling capacity to operators as a roaming service. This is a B2B2C model that avoids hardware subsidies and marketing costs. It places Starlink in a similar position to a bandwidth wholesaler. The catch is that wholesale margins are significantly lower than direct-to-consumer margins. The operator, not Starlink, owns the customer relationship. This transition would erode the very ARPU assumptions that support the revenue forecast.
Moreover, the B2B2C model introduces a dependency on the goodwill and pricing power of the carriers. In the consumer broadband market, Starlink controls the hardware and the monthly fee. In the direct-to-device market, Starlink is a supplier to that carrier's network. The operator can demand lower prices or design competitive alternatives. This is the same trap I have seen in DeFi protocols that rely on a single price oracle; the architecture looks elegant until the external dependency is compromised.
The Silent Blind Spot: Data Center Traffic
The most contrarian observation is that the AI narrative at the heart of Musk's claim is technically misdirected. AI computations generate enormous amounts of data, but the vast majority of that data never leaves the data center. Training clusters shuffle terabytes between GPU racks through high-speed internal networks. Inference queries travel from the user to a cloud provider over conventional fiber backbones, not over satellite links. Satellites are not and will not be the optimal medium for intra-datacenter traffic. They are useful for connecting edge devices that lack terrestrial service—a self-driving car in a remote desert, a robot operating on a lonely offshore oil platform—but the number of such devices is finite. There are only a few hundred million vehicles and robots that might need continuous connectivity, and each requires a fraction of a consumer broadband connection. The aggregate capacity requirement is not orders of magnitude larger than current demand; it is an order of magnitude smaller than the 50% internet traffic claim.
The blind spot is a governance one, not just technical. If Starlink were to carry even a tenth of global internet traffic, it would be a single point of failure for international communications. A single company, a single country (the United States), and a single individual (Elon Musk) would control the flow of data for a substantial portion of humanity. This is not a business plan; it is a geopolitical vulnerability. The article from which this analysis draws its source material mentions no such risk. In the forensic world, we say verification must precede reputation. The reputation of the man does not alter the physics of the bottleneck.
Takeaway: The Forecast Will Break
Within twelve to eighteen months, the revenue figures will be tested. Starlink may well grow to 10 million subscribers and reach $15–20 billion in annualized revenue. That is credible. The $400 billion and $1 trillion figures will be revised or quietly abandoned. The direct-to-device wholesale business will generate real cash flow, but it will place Starlink in competition with the very operators that control its distribution. The capital expenditure treadmill will persist, and the FCF margins will remain in the telecom band, not the software band.
The real question is not whether Starlink can grow. It is whether the investment thesis is priced for a future that cannot exist. As a protocol auditor, I learned to check the maximum allowable borrow against the collateral. Here, the collateral is a constellation of 7,000 satellites and a physical ceiling of 44,000. Vaults drain when the loop is unverified. The Starlink revenue model is an unverified loop. Code is law, until it isn't. The ledger never forgets—and neither will the balance sheet when the replacement cycle matures.