SpaceX files publicly for the first time. Three business segments — Space, Connectivity, and AI. Eighteen point seven billion dollars in revenue. Four point nine billion in net loss. And one mission that hasn't changed since 2002.
Our mission is to build the systems and technologies necessary to make life multiplanetary, to understand the true nature of the universe, and to extend the light of consciousness to the stars.
To do this, we have formed the most ambitious, vertically integrated innovation engine on (and off) Earth with unmatched capabilities to rapidly manufacture and launch space-based communications that connect the world, to harness the Sun to power a truth-seeking artificial intelligence that advances scientific discovery, and ultimately to build a base on the Moon and cities on other planets.
The S-1 frames the company's entire existence as an engineering response to one fact: for the whole of human history, civilization has lived on one celestial body. The risks are unpredictable on a planetary scale — asteroid impacts, volcanic activity, solar fluctuations, man-made conflict.
The line they keep returning to reads like a software postmortem: existential risk with a probability of one that must be solved.
“We do not want humans to have the same fate as dinosaurs.”
U.S. electricity generation grew at a 0.1% compound annual rate from 2008 to 2023. AI compute demand is doubling. The S-1's solution is to move the data centers up — to a continuously illuminated solar array in orbit.
“The Sun contains approximately 99.8% of the solar system's energy and, as a result, we believe it is the only truly scalable solution to terrestrial energy constraints in the age of AI.”
“The next paradigm shift for humanity is the creation of a resilient, perpetually expanding spacefaring civilization that drives continuous innovation across new frontiers, ultimately propelling us to Kardashev Type II status.”
Half the things you’ll be asked to build shouldn’t exist. Question every constraint — especially the ones that look like physics but are really just convention.
The best part is no part. The best process is no process. Add only after deletion has been tried first.
Optimize what survives. Not before — premature optimization is the larger waste.
Cycle time is the input to learning. The faster you iterate, the more reality teaches you.
Only the proven processes get automated. Automation is amplification; amplifying a flawed process is a flawed process at scale.
A set of core execution principles the filing refers to as “The Algorithm” — a five-step iterative process used as day-to-day guidance. Make the requirements less dumb, delete unnecessary processes or parts (embracing the principle that the best part is no part), only then optimize the necessary processes or parts, accelerate cycle time, and automate only proven processes after the first four steps are completed.
Reusability changed the unit economics; reliability changed the schedule. The S-1 shows the same data SpaceX shows its own employees: annual Falcon launches over time, with the number of new boosters flown each year overlaid on top.
The launches go vertical. The new-booster bars stay flat. The gap between the two is the entire story of modern rocketry.
SpaceX delivered 2,213 metric tons of payload to orbit in 2025 — more than eighty percent of the global total.
Q1 2026 alone: 556 tons. The annualized run rate is approaching three thousand.
“We believe rocket launches and landings should be as routine and commonplace as airplanes taking off and landing.”
Thirteen years after the company was founded, the first orbital-class booster came home and landed under its own power. The S-1 calls it “a 10-year lead.” From that night onward, everything became a balance-sheet question.
“In December 2015, we achieved what many deemed impossible: landing a rocket launched to space back on Earth.”
According to NASA, the historical average launch cost was $18,500 per kilogram. Falcon 9, on its first flight in 2010, brought it to $2,700. Falcon Heavy: $1,400. Starship is targeting a further 99% reduction from the historical average.
The S-1 reports subscriber counts as of each year-end and quarter-end. The Q1 2026 figure of 10.3 million is more than double the same quarter the year prior.
That number includes ~7.4 million monthly unique devices on the satellite-to-mobile constellation — partners with roughly thirty mobile network operators across six continents now route around dead zones via SpaceX.
The trade is visible in the numbers. Annual ARPU has fallen from $99/mo in 2023 to $81/mo in 2025. Q1 2026 ARPU dipped to $66/mo.
The reason: aggressive expansion into lower-income markets and the rollout of the cheaper Starlink Mini and Starlink Mobile tiers. Subscriber growth more than compensates — Connectivity revenue is up +49.8% year-over-year and Segment EBITDA +86.2%.
“From Antarctica's frozen wilderness to vast oceans and towering mountaintops — overcoming barriers posed by traditional terrestrial infrastructure.”
In 2025 Connectivity produced $11.4B of revenue and $4.4B of operating income. The Space segment swung from black to red as it ramped Starship. The AI segment burned six and a half billion.
Starlink is what the rest of the company is paid for with.
“The key constraints in the continued growth of AI are physical — chip manufacturing, data center infrastructure, and power generation. The future of AI will be determined by the control of the physical stack.”
COLOSSUS came online in 122 days. COLOSSUS II in 91 days. The first cluster of COLOSSUS II — 110,000 GB200 chips drawing 210 MW — was operational in roughly the time it takes a normal data center company to write the environmental impact report.
The Q1 2026 nameplate compute draw passed 1.0 GW for the first time. The filing flags a next phase of expansion bringing at least 220,000 additional GB300 processors and over 400 MW on top.
U.S. electricity generation grew at a compound annual rate of 0.1% from 2008 to 2023. AI compute demand is doubling. Something has to give.
The S-1's answer is to move the data centers into Sun-synchronous orbit, where solar arrays produce more than five times the energy per unit area as on the ground. SpaceX already builds the satellites. Already manages a 9,600-vehicle constellation. Already owns the launch capacity. The orbital data center is the same product, with a different payload.
“Our goal over time is to launch 100 gigawatts of compute to space each year. If operated continuously, that could generate approximately one-fifth of the annual power production in the United States.”
AI segment capex for 2025: $12,727M. Q1 2026 alone: $7,723M. That makes AI by far the largest cash sink in the company — and the deepest losses too.
The S-1 admits this is “an earlier stage of development” with “continued investments to support long-term growth opportunities in AI.”
Until 2024 the company looked like a launch business with a side hustle. Then Starlink crossed the line. Connectivity is now the largest, fastest-growing, and only profitable segment.
The AI segment was reclassified onto the income statement in February 2026. The 2023 and 2024 figures shown here include xAI and X retrospectively, reported as under-common-control.
2024 was the first year of operating profit in company history: +$466M. 2025 sent it back into the red on the AI buildout: ($2.6B).
The S-1 wants you to read the path, not the snapshot.
Connectivity throws off cash. Space is investing in Starship — three billion of R&D in 2025 alone — and is back to a small operating loss. AI is the deepest hole on the balance sheet: a six and a half billion operating loss against three billion of revenue.
“We believe we have identified the largest actionable total addressable market in human history. $28.5 trillion.”
Source: S-1, “Our Market Opportunity,” p. 11. Excludes China and Russia per the filing's methodology.
The filing's Risk Factors section runs sixty pages. Below are the ones most likely to keep an underwriter awake at night, ranked by severity and category. Click any row to expand the summary and the verbatim source quote.
If Starship fails to achieve full reusability and the required launch cadence, the V3 Starlink rollout, satellite-to-mobile expansion, and the entire orbital AI compute plan delay with it.
“AI compute satellites at scale need full Starship reusability to be economically compelling.”
Incorporated as Space Exploration Technologies Corp. in Delaware, March 14.
First private company to put a liquid-fueled rocket into Earth orbit.
Launch cost ≈ $2,700/kg — 85% below the historical average.
First commercial spacecraft to dock with the International Space Station.
December 2015 — Falcon 9 first stage lands back on Earth. Establishes a 10-year lead.
First reflight of an orbital-class booster. Reusability becomes routine.
First Falcon Heavy lifts a Tesla Roadster into orbit around the Sun. Space segment EBITDA-positive.
First operational Starlink satellites enter LEO.
First private crewed orbital launch — restores U.S. ability to launch astronauts.
First company to manufacture consumer-grade phased-array user terminals at scale.
Starship integrated flight test. Connectivity segment EBITDA-positive on a sustained basis.
Launch tower “chopstick” arms catch a returning Super Heavy booster mid-air.
First large-scale LEO satellite-to-mobile constellation.
xAI merged into SpaceX. COLOSSUS + COLOSSUS II reach 1.0 GW. Form S-1 filed May 20.
Expected first payload delivery to orbit, including V3 Starlink satellites — up to 60 per launch.
$19.6B AWS-3, AWS-4, and H-block spectrum acquisition expected to close ~November.
Initial deployment of orbital data-center satellites in Sun-synchronous orbit.
Cargo to the Moon, surface industrialization, factory pilot for AI compute satellites.
Starship-class crew transport to Mars in a transfer window.
Harnessing the full energy output of the Sun — the stated north star of the filing.
“We believe rocket launches and landings should be as routine and commonplace as airplanes taking off and landing.”
“We expect to begin deploying our orbital AI compute satellites as early as 2028.”
“Our plans to deploy large-scale orbital infrastructure, including orbital AI compute systems, will require the operation of very large satellite constellations, potentially numbering up to one million satellites.”
“Our goal over time is to launch 100 gigawatts of compute to space each year.”
“Our initial efforts will prioritize lunar cargo landings and returning Americans to the Moon, followed by expanded crewed missions that we believe can establish a continuous flow of cargo and humans between Earth and the lunar surface.”
“Our plans to deploy large-scale orbital infrastructure will require the operation of very large satellite constellations, potentially numbering up to one million satellites.”
If Starship works at scale, the launch industry is a footnote and SpaceX runs the new physics layer of the economy. If Starlink keeps compounding, it's a trillion-dollar ISP within the decade. If orbital AI compute happens at all, it happens here.
And if any of those doesn't happen, you own equity in an extraordinarily cool aerospace contractor.
Four hundred and seven pages. Read once, by a machine.
Here's how the film was made.
Three sample pages from the filing — pick a tab. Each parsed item is overlaid on the rendered page with a colored bounding box and listed on the right with the actual extracted value. Hover either side; the highlight pairs across.
The Market Opportunity page — heading, prose, a stacked bar chart, and a complex multi-row table with rowspans.
Every page of the parsed document. Hover any tile to see its index. The raw output for the full 407-page filing fits in a 4.5MB JSON. We wrote zero layout code.
LlamaParse is the parsing layer underneath this site. It turns long, layout-heavy documents — S-1s, 10-Ks, research papers, contracts, decks — into the structured data you can actually build with.
