> When Microsoft buys an Nvidia GPU for $100, it will claim that it lasts 5 years, so writes down $20 in the expense column for this year. Meanwhile Nvidia takes all the profit at time of sale, $100

Nvidia would log $64 of profit against a $100 sale [1]. (The R&D cost to develop the chip have already been booked.)

> for each GPU bought, SP500 logs $80 of profit, even though no money entered or left the SP500

In your example the S&P would log ~$40 of earnings. (About because taxes.)

[1] https://www.macrotrends.net/stocks/charts/NVDA/nvidia/profit...

▲

bagacrap 2 hours ago | parent [-]

Ok great, thanks for clarification. But still, on year 3 of the AI buildout, assuming the pace doesn't accelerate (it's already pretty fast!) then each GPU bought logs $0 profit since it's offset by amortization of the last two years. Compared to $40 of today, profit dropped by $40, meaning earnings contraction is around the corner...? Unless of course Microsoft manages to actually earn $20 or more from each GPU each year.

I suppose the uncertainty here would be the reason for financial market volatility.

▲

JumpCrisscross 2 hours ago | parent | next [-]

> Compared to $40 of today, profit dropped by $40, meaning earnings contraction is around the corner...? Unless of course Microsoft manages to actually earn $20 or more from each GPU each year

Correct. If Microsoft fails to earn more profit from a capital asset, over said asset's useful life, than was paid for it, Microsoft's earnings would be reduced. Put simply, it would have made a bad investment. Presumably, that would affect whether Nvidia gets another order.

If you've ever wondered why subscription mania took hold over the last decade, you've backed into the answer.

(Note that Microsoft depreciates "servers and network equipment" on a straight-line basis over "two to six years" [1].)

[1] https://www.sec.gov/Archives/edgar/data/789019/0001193125263... page 61

	▲	Frieren an hour ago \| parent [-]
		> why subscription mania took hold over the last decade, I don't see business to business subscriptions as a bad thing. It it a way of sharing risks. The horrifying thing is that they do that to consumers. Consumers don't have the legal teams to analize such contracts. Consumers do not have people dedicated to evaluate prices and change providers (Even for companies is costly, so individuals get stuck in products they do not use anymore). Subscriptions change prices and conditions at will, what makes it impossible for a normal consumer to follow up on everything that changes over time. The fact that even when you theoretically buy, like a phone or even a TV, also becomes a subscription because the software it runs needs updates. Subscription models that may make sense for corporations are killing the working class ability to own things or even choose which business they want to buy from. It is just about capturing customers and not letting them leave as prices are pushed up.

▲

protocolture 42 minutes ago | parent | prev [-]

>Ok great, thanks for clarification. But still, on year 3 of the AI buildout, assuming the pace doesn't accelerate (it's already pretty fast!) then each GPU bought logs $0 profit since it's offset by amortization of the last two years.

Probably worth noting that Azure has had GPU resources for a while, I think it was already a profitable line for them, and since people have been chasing GPU compute on cloud platforms it can be difficult in some regions getting a reservation. Microsoft is probably looking at AI as Another profitable use case for our assets, let's expand it rather than a greenfields GPU buildout from zero. And where capacity doesn't get used by AI their customers can use it for other stuff across most not all SKUs