[vanuatu]

can someone explain why this is actually bad?

nvidia spends X amount to invest in data centres or investments on the agreement that the counterparty spends Y amount back, the net delta is the actual amount of value being transferred aka Nvidia sells chips as usual despite the high numbers of X and Y?

The frontier labs do not have enough chips to meet demand, and AI demand is ferocious and climbing, so I'm not sure what the story is here

[nickff]

The problem is that when a vendor finances their customers, they can create the illusion of 'real' demand for their product, when most of the the end-users are only actually using something because it's cheap. When the vendor runs low on cash and starts requiring payment, the customer may not be able to afford it, taking both vendor and customer down, and leaving the end-users who have a real need, and were willing to pay sustainable prices without any options.

[vanuatu]

that makes sense to me in a conceptual sense

however inference is very profitable and plummeting in cost for a given point on the intelligence curve, and nvidia gpus can serve different models so they are protected post-buildout

[nickff]

You are describing the justification that NVDA is using to explain their behavior; they see it as something of a 'bridge-loan' until the LLM business model reaches steady-state. The problem is that this explanation has been used for many bubbles, where companies mis-categorize ongoing costs as one-time expenses.

[javier2]

> however inference is very profitable

Is it? OpenAI and Anthropic are burning cash faster than anyone has ever shoveled cash into a furnace.

[polski-g]

This was true about 18 months ago. It is no longer true.

[surgical_fire]

> however inference is very profitable

Is there any actual evidence of that?

[CoolestBeans]

Its bad if the expected demand is an illusion. For example, when a company builds out a data center they don't build it for demand today, they build it for the demand they expect when the data center is running and for how much they expect demand to grow over the lifetime of the data center (this is a simplification, they build a financial model of how they can grow capacity as demand increases over the lifetime of the data center). If the demand is lower than expected then the counterparty cannot spend that Y amount back. In other words, Nvidia now holds bad debt (really worthless equity since these aren't loans on paper). Furthermore, Nvidia has been making the same bet with multiple companies. That Y amount the counterparty can't pay back is probably correlated with all of the counterparties Nvidia lent X amount to. Suddenly this circular flywheel begins operating in reverse. Now Nvidia has no X amounts to lend to AI companies which makes their ability to pay back Nvidia worse which means Nvidia has less money to lend out and on and on.

There's other problems too, why do we think AI demand is ferocious right now? Nvidia's revenue is one of the biggest signals we use to determine that. Why is Nvidia's revenue so large? They're spending their revenue on more revenue. This process overinflates what AI demand might actually be.

The issue really boils down to that this is a risk that gets reported in a way that makes it look less risky than it really is and therefore actors make investment decisions that they might not otherwise make. Sure, it might work out. But if it doesn't, the pain could be way more painful than it looks on paper.

[javier2]

its not so bad if the funds arent used to borrow 10x, and then spent entirely on nvidia chips.