Background: SpaceX is Elon Musk's rocket company, which listed in June this year in a US$1.8 trillion IPO. But these days, it's not just launching rockets. SpaceX is also running some of the world's biggest AI data centres, which means it has found itself in the same familiar AI infrastructure race: get as many Nvidia chips as you can.  

What happened: SpaceX is reportedly in talks with asset manager Apollo and a group of banks to borrow US$40 billion to buy Nvidia chips. The chips would be used as collateral for the loan. So basically, if SpaceX defaults, lenders could potentially take control of the chips.

What else: The borrowing comes after SpaceX raised another US$25 billion in bonds... just two weeks after its June IPO. And get this: the company already had nearly US$40 billion in debt in its latest results. So the race for AI infrastructure isn't just about finding enough chips anymore. It's also becoming a race to find enough bank debt to buy them.

What's the key learning?

💡 The AI boom runs on chips, but those chips are running more and more on borrowed money. Building AI is ridiculously expensive. We’re talking about the cost of AI chips, the cost of running data centres and the power to actually run the AI compute.

💡 AI spending is getting too big for companies to fund alone. Big Tech used to cover these costs, but Morgan Stanley expects global AI-related debt issuance to hit nearly US$570 billion this year. That's more than double what it was last year.

💡 Banks want higher interest rates to keep funding the boom. JPMorgan recently offered a US$5 billion AI data centre loan at around 11% interest, putting SpaceX's proposed US$40 billion chip financing among the biggest examples of how expensive AI infrastructure is becoming.