SpaceX is asking bond buyers for at least $20 billion only days after the largest IPO on record, and the real story isn't rocket launches. It's whether Elon Musk can turn compute, Starlink, and debt markets into one machine.
SpaceX's first trip to the bond market looks strange only if you still think this is mainly a rocket company. Business Insider reported that SpaceX had $100.8 billion in cash and equivalents as of June 19, after raising $85.7 billion in its June 12 IPO. Then, on June 22, the company moved to raise at least $20 billion in senior unsecured notes while its stock fell hard. If you're trying to understand the market's reaction, start there: investors aren't just pricing launches anymore. They're trying to price Musk's AI infrastructure plan.
The offering is current, and the timing is the point. The Wall Street Journal reported Monday that SpaceX plans to use the proceeds primarily to repay a bridge loan tied to its earlier xAI deal, with any remaining money going to general corporate purposes. Moody's assigned SpaceX a Baa1 rating and S&P Global gave it a BBB rating, according to the Journal. Those are investment-grade ratings, which changes who can buy the debt and how cheaply SpaceX can fund itself. Pension funds and insurers that wouldn't touch venture-style risk suddenly have a rated bond to look at.
Don't confuse that with financial weakness. A company with $100.8 billion in cash doesn't need to borrow because the drawer is empty. It borrows because long-term debt lets it preserve equity, stretch maturities, and keep cash available for the expensive parts of Musk's plan: Starship, Starlink, AI data centers, and the proposed space-based compute infrastructure that still sits closer to ambition than operation.
The stock told you how uncomfortable that mix still makes public investors. Business Insider said SpaceX shares closed Monday at $154.60, down 17% for the day and 27% below their June 16 peak of $211.39, though still above the $135 IPO price. MarketWatch reported that the drop wiped roughly $400.8 billion from SpaceX's market value and pushed the company below its IPO-day close. That's not a small wobble. That's the public market deciding it needs more proof before it treats SpaceX like a cloud company with rockets attached.
There is proof starting to arrive, but it isn't clean yet. The Journal reported that Reflection AI agreed to pay SpaceX $150 million a month from July 1 through the end of 2029 for data-center capacity, a contract worth $6.3 billion if it runs the full term. The same report noted that either side can terminate after the first three months with 90 days' notice. Keep that caveat in the story. It matters more than the headline number because AI compute contracts sound permanent right up until supply, model economics, or customer funding changes.
Still, $150 million a month is not a vanity metric. It tells you what SpaceX is trying to become. The Colossus data-center operation, built around Nvidia hardware for xAI's Grok models, is now being sold to outside AI customers as capacity. Reflection wants open-source model training. Other reported customers include Anthropic and Google. Cursor, the AI coding company, is tied to a separate $60 billion all-stock acquisition agreement reported by the Journal. This is no longer a side project sitting beside rockets. It is moving into the center of the valuation.
Frankly, that is why the old SpaceX story is now too small. Rockets got the company here. Starlink gave it recurring revenue and a global network. AI compute is the attempt to turn both into infrastructure that other labs have to rent. Amazon Web Services, Microsoft Azure, and Google Cloud built their models from terrestrial data centers and enterprise software. SpaceX is coming from satellites, launch cadence, and Musk's willingness to finance gigantic physical bets before the spreadsheet looks tidy.
The risk is just as concrete. S&P analysts, according to the Journal, described SpaceX's rocket operations as solid and Starlink as capable, but called the AI division the riskiest part because it needs heavy upfront spending and revenue is still uncertain. Business Insider reported that SpaceX lost $4.9 billion in 2025 on $18.7 billion in revenue. The company is trying to sell the bond market a future in which those losses are the cost of building a new infrastructure layer. Bond buyers can accept that story for a coupon. Equity buyers have to believe the upside is still worth a valuation above $2 trillion.
The orbital data-center idea remains the hardest part to value. Space-based compute could, in theory, use constant solar power, radiative cooling, and tight integration with satellite communications. But the Reflection deal, the Anthropic capacity, and the Google work reported so far are ground-based contracts. You shouldn't give SpaceX full credit for orbital infrastructure until there is something more specific than ambition and capital spending language.
That's the real issue with this bond sale. It isn't shocking that SpaceX is raising debt after an IPO. It is shocking how quickly the company has forced public investors to decide what kind of business they own. If you bought a rocket-and-satellite company, the $20 billion bond raise looks aggressive. If you bought an AI infrastructure company with its own launch system and global broadband network, it looks like the opening move.
For now, the bond market is giving SpaceX money at investment-grade terms while the stock market is taking some heat out of the IPO frenzy. That split is useful. Debt investors are being paid to wait. Shareholders are being asked to believe that Musk can turn cash, compute, and orbit into one balance sheet. Watch the contracts that survive their cancellation windows. Watch where the next data centers actually get built.
Also read: Chevron and Microsoft sign a 20-year gas deal that turns an oil major into an AI infrastructure company • AI Customer Acquisition Automation Lets One Founder Do the Work of Three • Amazon is recruiting Hindi speakers to beta test Alexa+ as it eyes India's 600 million-person voice market
