The Orbital Bid

In six months, the orbital data center went from conference slide to line item in the largest IPO in history. If you want to know what capital actually believes about the terrestrial power curve, that repricing is the tell.

The sequence matters. In February, Musk folded xAI — Grok, the X platform, and the Colossus campus — into SpaceX in an all-stock deal valuing the combination at $1.25 trillion, the largest private merger ever. His stated rationale was not social media synergy. It was, in his own words, orbital data centers. In June, the combined company listed at $135 a share — a $1.77 trillion valuation, roughly $75 billion raised — and crossed $2 trillion on its first trading day.

What the S-1 actually commits to

Strip away the roadshow and the prospectus describes a real program with real physics: AI compute satellites in sun-synchronous orbit "as early as 2028," building toward a stated long-term goal of 100 gigawatts of orbital compute per year — a figure that would require, by SpaceX's own math, thousands of launches annually and on the order of a million metric tons to orbit. The unit design is a 70-meter-wingspan satellite — wider than a 747 — sustaining roughly 120 kW of compute payload. A February FCC application reserves room for up to one million satellites in the "SpaceX Orbital Data Center system" at 500–2,000 km.

The case is the one orbital advocates have always made, now with a balance sheet attached: near-continuous solar flux, radiative cooling instead of water fights, no interconnection queue, no neighbors, and a launch monopoly that keeps cutting its own costs.

The prospectus versus the pitch

At Davos in January, Musk said space would be the lowest-cost place to run AI "within two years, maybe three at the latest." The S-1, filed with liability attached, says something different: orbital data centers may never be commercially viable. Both statements come from the same company in the same quarter. The gap between them is the gap between a pitch and a prospectus, and only one of them is signed.

The skeptics' math has not moved: radiation degrades terrestrial silicon, radiators — not compute — dominate the mass budget, a failed rack in orbit is dead inventory forever, and SoftBank's Son spent June publicly picking at the two soft spots, latency and launch cost.

The terrestrial irony

Here is the detail the orbital narrative skips: the cash engine underneath the space bet is aggressively terrestrial. Per the IPO filings, Anthropic is paying $1.25 billion a month for exclusive run of Colossus 1, and Google signed on at ~$920 million a month — roughly $30 billion through mid-2029 — for capacity at Colossus 2. That is on the order of $26 billion a year in annualized compute leasing, flowing from two frontier labs to the company that says compute belongs in orbit. Meanwhile xAI's own operation lost $2.4 billion in a single quarter, and the ~2 GW, 555,000-GPU Colossus campus it built is the largest single-site AI installation on Earth.

The orbital bid is being bankrolled by renting out exactly the terrestrial scarcity it claims to obsolete. That is either elegant hedging or a confession, depending on your priors.

How to read it

As a price signal, the message is unambiguous: the market just assigned trillion-dollar weight to the proposition that terrestrial power constraints are durable enough to justify leaving the planet. As an engineering claim, the S-1's own risk factors price the doubt. Both can be true — and the pure-plays like Starcloud, flying demonstration GPUs without a launch monopoly to lean on, are where the thesis gets tested without the cushion.

File the orbital bid where it now belongs: no longer with 1970s space solar power as a paper dream, but not yet with fiber as inevitable infrastructure. It is a publicly traded hedge against the grid — and the grid, for now, is still collecting the rent.