SpaceX Lost $37 Billion. Now It Wants Your Pension

SpaceX filed its S-1 prospectus two days ago, and it immediately stood out as unlike any IPO document in securities regulation history. The filing opens with 14 pages of rocket and satellite photographs, reminiscent of WeWork's ill-fated prospectus that was later described as one of the worst IPO filings ever. The document is filled with phrases never before seen in SEC filings, including the repeated claim that SpaceX's primary objective is «to extend the light of consciousness to the stars» — a phrase that appears 10 separate times.

The prospectus outlines future business lines that read like science fiction: long-haul point-to-point terrestrial travel using rockets instead of airplanes, in-orbit manufacturing, passenger and cargo transportation to the Moon and Mars, manufacturing and energy production on celestial bodies, and asteroid mining. The acronym AI appears over 200 times in the document, the word consciousness appears 10 times, and the word profit appears far less frequently than either. Investors are warned that «we do not want humans to have the same fate as dinosaurs,» which sets the tone for what follows.

What emerges is quite possibly the most extraordinary document in American securities regulation history. Every claim, no matter how absurd it sounds, comes directly from the official filing. The company is asking investors to believe in a vision that spans from GPU rental to Mars colonization, all while burning billions of dollars annually.

According to the prospectus, 93% of SpaceX's claimed $28.5 trillion addressable market is attributed to AI, with enterprise applications accounting for 80% alone. This means the company's valuation rests primarily on businesses paying for Musk's AI tools to improve productivity. The problem is that SpaceX's AI product, Grock, holds roughly 3.4% market share in a market dominated by OpenAI, Anthropic, and Google. The primary documented use cases for Grock are fact-checking tweets and generating non-consensual nude images that have resulted in regulatory investigations in multiple countries.

According to Bloomberg, SpaceX's own engineers have been slow to adopt Grock for technical work because it underperforms rival tools. Musk himself admitted last month that the AI code would need to be rebuilt from scratch, shortly after selling XAI to SpaceX for $250 billion. His solution was to attempt to acquire Cursor, the coding tool his own engineers actually use, for $60 billion with a $10 billion break fee. The company whose AI product represents 93% of its claimed addressable market has a flagship product its own engineers won't use and is spending $60 billion trying to buy a competitor.

The near-term AI revenue story rests substantially on a $15 billion annual deal to rent compute capacity to Anthropic, one of the very competitors whose existence proves Grock is losing the AI race. This contract, representing roughly 40% of projected near-term revenues, can be cancelled by Anthropic with just 90 days notice. SpaceX is essentially functioning as a GPU landlord rather than an AI innovator.

Musk is receiving 1 billion performance-based restricted shares that only vest when SpaceX establishes a permanent human colony on Mars and hits a $7.5 trillion market cap. SpaceX explicitly describes achieving these milestones as «improbable» in the prospectus. This isn't sudden candor — it's an accounting trick. Under accounting rules, if the company admitted these goals were probable, they would have to expense the full value of the award, potentially hundreds of billions of dollars, on their income statement today. By categorizing the grant as improbable, it gets booked at zero cost.

But as law professor Ann Lipton points out, Musk doesn't actually need to achieve those goals to benefit. He gets the utility of these shares immediately: he can vote them today, collect dividends on them today, and with board approval pledge them as collateral for loans to access tax-free cash today. He receives all the benefits of owning the shares based on a Mars colony the company admits he will probably never build. This is one of the most extraordinary compensation structures ever designed, allowing immediate control and financial benefit while deferring the actual performance requirements to an indefinite future event the company itself calls improbable.