I do not read the whitepaper; I read the bytecode. But Starlink has no bytecode. It has telemetry. And the telemetry screams a different story than the narrative.
A few weeks ago, David Friedberg sat with Elon Musk and painted a vision: Starlink, already at 600,000 subscribers, could one day carry 50% of global internet traffic. The implied revenue? $1 trillion annually. Free cash flow? $300 billion. The market yawned. Then it cheered. Then it forgot to check the physics.
I spent the last 96 hours reverse-engineering the claim. Not the whitepaper—there is none. The public satellite data. The capital expenditure models. The spectrum allocation tables. The result is a cold, quantitative teardown of a vision that smells more like a Ponzi than a protocol.
Context: The Hype Cycle
Starlink is SpaceX’s low-earth-orbit (LEO) satellite constellation. It currently operates ~7,000 satellites. Each V2 Mini unit delivers about 60-100 Gbps of capacity. The service is a last-mile broadband solution for rural, maritime, aviation, and government clients. The unit economics are decent: $100-120/month ARPU, high churn in competitive areas, but a sticky base where no fiber exists.
Friedberg’s thesis rests on two pillars: (1) AI and robotics will explode bandwidth demand, and (2) Starlink will capture that demand because it’s the only global, low-latency network. The numbers are seductive. But they ignore the physical constraints that make the 50% claim mathematically impossible without a 10x improvement in satellite capacity and a 5x increase in ground station density.
Core: The Systemic Teardown
Let’s start with the traffic target. Global internet traffic in 2025 is estimated at 1.1 PB/s peak (Cisco forecast). 50% of that is 550 TB/s. A single V2 Mini satellite, at best, handles 100 Gbps. To meet 550 TB/s, you need 44,000 satellites operating at full capacity simultaneously. SpaceX has filed for 42,000. But current launch cadence is about 200 satellites per month. At that rate, it takes 18 years to build the constellation—assuming no failures, no orbital decay, and no spectrum interference.
Orbital decay is not an assumption. It’s a law. Starlink satellites have a 5-7 year lifespan. To maintain 44,000 birds, you need to replace 6,000-8,000 per year. That’s 600-700 launches annually. SpaceX launched 96 times in 2024. The Falcon 9 can lift 60 satellites per launch. The Starship, still in testing, promises 400. But even with Starship, you need 15-20 dedicated launches per year just for replacement. That’s a CapEx of $5-10 billion annually, not including the satellites themselves (costing ~$500,000 each).
Now, the free cash flow claim. Friedberg said $300 billion FCF. At 75% margin, that implies $400 billion revenue. Telecom operators average 10-20% FCF margins. Starlink’s vertical integration gives it an edge, but 75% is absurd. The satellite replacement cycle alone eats 30% of revenue. Ground station maintenance, spectrum licensing, and customer acquisition costs add another 20%. Realistic FCF margin: 15-25%. That gives $60-100 billion FCF, not $300 billion.
But the bigger flaw is the assumption that Starlink can capture 50% of traffic. The majority of AI and robot bandwidth is generated inside data centers. Datacenter-to-datacenter traffic never touches a satellite. Edge AI inference may produce some, but that’s a fraction of the total. The 50% claim implicitly assumes that all new demand will be routed through LEO. That’s like assuming all new car traffic will use a single bridge.
Contrarian: What the Bulls Got Right
I must admit: Starlink is the only LEO constellation that works at scale. OneWeb is dead. Kuiper hasn’t launched. The technology is real. The Direct-to-Device partnership with T-Mobile, KDDI, and Rogers opens a B2B2C channel that bypasses hardware costs. That could generate $10-20 billion in wholesale revenue by 2030 without adding a single consumer subscriber.
And the military contracts are a goldmine. The U.S. Department of Defense alone spends $50 billion annually on connectivity. Starlink already has a Pentagon contract for Ukraine. If it captures 10% of that, that’s $5 billion in high-margin recurring revenue. The maritime and aviation segments are similarly sticky.
So the bulls are right that Starlink will be a successful business. But not a $1 trillion business. The addressable market for satellite broadband is at most $200 billion globally (current telecom market is $2 trillion, but satellite share is 2-3%). Even if Starlink dominates that, it’s still a $50-100 billion revenue company. That’s a 10x from current estimates, not a 100x.
Takeaway: The Accountability Call
I have audited many DeFi projects with similar narratives. "We will capture 50% of the market." They all hit the same wall: the physical layer. Starlink is no different. The difference is that Musk’s rockets are real, but the numbers are just as fictional as a yield farm promising 100,000% APY.
When the satellites burn up and the replacement launches are delayed, the free cash flow will evaporate. The market will wake up, and the narrative will reset. I’ll be watching the launch manifest. Not the timeline.
I do not read the whitepaper; I read the bytecode. In Starlink’s case, I read the telemetry. And the telemetry says: 50% is a dream, not a forecast.