A bigger, more granular model
Saigau explained that this is the fourth full generation of the model, refined across hundreds of iterations. Where earlier versions deliberately kept to 20 or 30 adjustable variables, this one runs to roughly 500 rows. The trade-off was intentional: less user-friendly, but far more customizable, so the team can tune assumptions year by year as SpaceX progresses. At its core sits a cash allocation engine that splits the company into three business lines (Starlink, customer launch, and AI compute), nets out corporate overhead and an arbitrary carve-out for the Moon and Mars, then ranks each module by profitability and allocates capital accordingly.
The TerraFab surprise
The biggest takeaway, by both hosts' reckoning, concerns TerraFab, the chip fab joint venture with Tesla. The model suggests the full facility does not need to be built until past 2040 to supply enough chips for the compute revenue they project. Roughly one tenth of the aspirational 1 terawatt output, on the order of 100,000 wafer starts per month, would cover orbital data center needs through 2040. Because TerraFab is shared with Tesla, which will consume a large portion of the chips for cars and robots, not all of that capex even lands on SpaceX's books.
Starlink, D2C, and customer ownership
Starlink remains the revenue majority until around 2030, when the model has V3 deployment reaching demand saturation and the return on launching more satellites falling. The hosts spent time on direct-to-consumer, where their thesis runs against the consensus. Rather than viewing D2C as a coverage-gap filler, as a player like AST Space Mobile might, they believe SpaceX will push to own the customer relationship outright, making Starlink Mobile a subscriber's primary connection and wholesaling from carriers only where needed. Owning the relationship flips the economics from capturing 1 to 10 percent of customer value to capturing 90 to 99 percent. Notably, D2C accounts for less than 1 percent of total bandwidth, with broadband carrying the rest, which underscores that the value of D2C is about customer ownership, not capacity. The model projects roughly 1 billion subscribers by 2040 across both lines, though heavy overlap means unique customers likely land closer to 500 to 600 million.
AI compute takes over
AI compute starts small and grows to match Starlink by 2035 before accelerating. Saigau was candid that AI demand in 2040 is essentially unknowable, so the team imposed a semi-arbitrary demand curve to keep projections sane. Even reined in, company-wide revenue reaches $1.8 trillion in 2040. Orbital compute economics improve as satellite power density climbs from about 50 kilowatts per ton toward Elon Musk's 100 kilowatt per ton target, which the model does not hit until 2040. AI satellites cost an estimated $5 to $6 million each early on, versus about $1.2 million for a Starlink V3.
On the debate over whether space compute is too expensive, the hosts pushed back on a recent Semi Analysis report. They argued terrestrial compute is getting more expensive (Jensen Huang reportedly sees 1 gigawatt costing $100 billion soon) while no one can drive down space payload costs like SpaceX. A useful external check: Goldman pegs SpaceX AI 2030 revenue at $322 billion, close to the model's $305 billion.
Valuation: pick your methodology
The model outputs valuations three ways. A revenue multiple makes sense early, EV/EBITDA fits the 2030 to 2040 window, and a Gordon perpetuity-growth DCF is the most conservative. For 2026, those methods give $3.7 trillion, $5.6 trillion, and $2.4 trillion respectively. By 2040 the spread widens to $10 trillion, $18 trillion, and $6 trillion. The conservative perpetuity figure of about $2.5 trillion sits roughly where the IPO is expected to land.
The hosts closed by noting they will keep updating through the roadshow, with an initial write-up due later that day and a thermal engineer who worked on Starlink V1 lined up for a future episode.