OpenAI Growth Stalls at $10B+ Burn: The Structural Cracks in the AI Infrastructure Trade

For me the question is whether OpenAI can keep paying for the compute it has already promised to buy, that suppliers are relying on for their own debt repayments.

Oracle is the one Im most concerned for. It's borrowing against a very durable database business with renewals around 97%, and gross margins near 90% to build this GPU capacity. However, net debt is about $125B, roughly 3.7x EBITDA, with around $40B more in planned raises. Return on the new money has been about 7% against a 10% hurdle. Last quarter it spent $28.5B on capex against $19.3B of revenue. Not earnings, revenue! The stock is already down about 50% over a year, but cheaper isn't the same as safe when the customer concentration sits in one lab that's still raising money privately as it burns tens of billions per year with no sign or path to things changing direction. Meanwhile their bonds sitting one notch above junk is extremely worrying as that in itself could make things messy if they face another downgrade.

CoreWeave is the same structure with more leverage. Interest alone is heading toward $3.5-3.8B a year, which is more than the owner earnings I can estimate. If cap ex was to slow due to insufficient power supply to bring on new capacity then how do these companies survive in an inflationary environment where higher rates for longer is likely? To me it seems like a complete gamble to take that risk as an investor.

Nvidia is the opposite on the balance sheet (basically no net debt, huge cash generation). What I'm watching is the credit side: around $50B of equity in labs, up to $105B of lease guarantees, and receivable days going from 45 to 60 in one quarter. Funding your own demand through equity in startups does not seem a sustainable path for growth in comparison to what the other hyperscalers are capable of. CNBC love to talk about how much net income has grown in the recent Q, while ignoring the fact cash generated was 3x less than that figure, at around 20b. Customers are on paper buying 3 times more gpu’s than they are willing to pay for, something about this just doesn’t sit right with me, especially as I watch that gap between actual cash flow and earnings grow wider and wider. Artificially boosting demand/growth figures when the real amount getting brought in is 3 times less than the figure everyone talks about is interesting.

What do people think about this? Ps I love Ai and Claude’s api key/claude code has changed my life in ways I can’t even put into words or would have ever thought was possible. Allowing me to do things otherwise would have been impossible. Therefore there are two sides to this whole Ai bubble debate ! I don’t doubt for a second the growing demand for the ever expanding use cases of these models, which I also believe will diverse across many different customer types.