> Do you also have a calculation for ORCL ? 

For ORCL, the historical numbers might not tell you very much, because they are planning to significantly ramp up both capex and borrowing over the coming years.  So the concern is about the [economics of the business in the] future.  

Let's look at the street's forecasts for FY2028.  ORCL has a May year end, so that would be 2027 for some people.  The average forecast for net debt and EBITDA gives us a net debt:EBITDA of about 2.4x.  At first glance it could be manageable, putting aside for a moment the motivations of sell-side analysts, which, let's say tends to influence their estimates.   

But one wants to consider, among other factors, what kind of assets are on the B/S, and their ability to generate cash flow.   In this case, they are data centers filled with GPUs.  What is the productive life of a GPU?  Well, I think that it was earlier this year that during an Nvidia AI conference, Jensen quipped that with the launch of the  newer and more powerful Blackwell GPUs, “when Blackwell starts shipping  in volume, you couldn't give Hoppers away,” suggesting jokingly that the older Hopper GPUs might have so little value they'd be difficult to  dispose of even for free.  (I saw that as a very rare foot-in-the-mouth moment for Jensen.)  Perhaps they could be used for inference, but I would guess that not many hyperscalers would still use them for training.  That would imply that their ability to generate cash flow could be significantly impaired.  And that is over just 2 years.  

OTOH also relevant for consideration is that most hyperscalers still have a depreciation period for their servers of 5-6 years.   Although if their auditors are listening to Jensen perhaps that will have to change.