Can SpaceX Really Turn Starlink Into a $1 Trillion Business? Ron Baron Has $25 Billion Riding on Yes

Ron Baron has spent his career telling investors to sit still, and his firm’s biggest illiquid conviction is now a public stock. Baron Capital holds roughly $25 billion in SpaceX shares, and Baron himself has staked a headline projection that SpaceX (NASDAQ:SPCX | SPCX Price Prediction) can turn Starlink alone into a trillion-dollar revenue line.

The market is not yet paying for that outcome. SPCX changed hands at $150.88 on September 16, 2026, up 5.15% on the session but still down 6.26% from its post-IPO reference, with a market capitalization near $1.16 trillion.

That gap between price and Baron’s claim frames the entire question for investors. Baron has moved past the consumer broadband story and now argues that agentic AI will consume connectivity and compute at a scale that turns an orbital constellation into general infrastructure, with every number he cites resting on that one assumption.

Original Thesis and What Actually Changed

Baron framed his prior model cleanly. In his words, “SpaceX was planning on going from 10,000 satellites to 20,000 satellites”, and Baron said that against reported revenues of $18 billion last year, Starlink alone could reach $1 trillion in a decade.

Baron has since pivoted, and now says what will overwhelm everything are agentic agents, machine users that run continuously rather than in bursts.

A human broadband subscriber generates traffic in short sessions with long idle gaps. An AI agent booking travel, monitoring a factory line, or negotiating a contract does not sleep, does not close the browser, and multiplies per account rather than per person. Baron treats that as an observation. It is an assumption, because agentic workloads at commercial scale barely exist yet.

SpaceX’s Q2 numbers partially support the assumption. Connectivity revenue reached $4.29 billion, up 66% year over year, while enterprise and government inside that line grew 108%, even as consumer ARPU slipped from $85 to $66.

Constellation Math Is the Load-Bearing Part

Baron’s operational claim is that 100,000 satellites are required to serve the workload he expects, with each generation deorbited every five years and replaced with hardware ten times more capable.

That refresh cycle only closes if launch cadence, manufacturing throughput, and cost per kilogram all move together. Starship V3, which SpaceX says will cut cost to orbit by 99% or more, is the piece the model cannot survive without.

Capital intensity is already visible. SpaceX spent $18.37 billion on capex in Q2 alone, with $15.83 billion going to AI compute, and the company also placed a $25 billion inaugural investment-grade bond issuance to fund the buildout, according to Baron Capital. That kind of spending is why the power, cooling, and networking suppliers behind the AI buildout keep showing up in our free report on seven AI infrastructure names that aren’t chipmakers.

Data Centers in Orbit Deserve Harder Questions

Baron’s cleanest sales line is a cost comparison. He said a terrestrial unit costs about $10 in capital and $2 a year in maintenance, while in space Baron claims it costs $2 one time.

Nothing in space actually runs on a one-time cost basis, however. GAO researchers reviewing space-based data centers in April 2026 flagged that cooling at scale is unproven because heat is not easily dispersed in the near-empty vacuum of space, and that on-orbit servicing is underdeveloped.

The unit in Baron’s comparison is also undefined. A rack, a server, a watt of compute, and a satellite bus are not interchangeable. A ratio quoted this cleanly should invite the question Baron did not answer: what wears out first, and who fixes it.

Bull and Bear Case for SPCX Stock

The bull case is that Baron is right about the demand curve and directionally right about the constellation, and that SpaceX’s 12.0 million Starlink subscribers, $47.5 billion backlog, and 1.4 GW of AI compute already in place give it a running start no competitor can match this decade.

The bear case is that the revenue ceiling was never the binding constraint. Simultaneous execution across constellation expansion, orbital compute, and power generation is, and Baron is a holder talking his book with an unusually large incentive to sound certain about a ten-year forecast that is really a stack of compounding assumptions.

The near-term milestone worth watching is the Starship V3 test-flight cadence, because a launch-economics breakthrough is the single variable that decides whether the 100,000-satellite plan Baron sketched is arithmetic or fiction. Everything else in his thesis waits on that answer.

Contact [email protected] for any questions or corrections.