Five months later, Nvidia’s $100 billion OpenAI investment plan has fizzled out
- Thread starter JournalBot
- Start date
Upvote
466
(476
/
-10)
Upvote
305
(307
/
-2)
Just for context, noted liar Elon Musk claimed yesterday that he’d be launching 100 GW worth of compute per year into orbit Real Soon Now. Where 10x nVidia's production rate would come from was carefully left unstated.
Upvote
428
(430
/
-2)
People talking in GW of compute like GW grows on trees. Well, it falls from the sky half the day and blows around but there's one small orang problem there.
Upvote
297
(299
/
-2)
It all returns to nothing
It all comes tumbling down, tumbling down, tumbling down
It all returns to nothing
AI just keeps letting me down, letting me down, letting me down
Upvote
64
(69
/
-5)
A letter of intent isn't a deal.
It's barely a concept of a deal. Maybe 12% of a deal.
It's barely a concept of a deal. Maybe 12% of a deal.
Upvote
237
(237
/
0)
"Guys, $100 billion was the most we were going to invest, and we've invested $0 so we didn't lie!"
Upvote
250
(250
/
0)
So, how many stocks were sold by people in charge of the letter of intent and how much is this just a pump-and-dump with empty promises?
Upvote
198
(200
/
-2)
anonymouschicken
Ars Scholae Palatinae
I'm still scratching my head about why anyone thinks this is a good idea, even if it was feasible. What AI problems become easier in space? Space is a vacuum so you'd need huge radiators to cool the things. The distance between individual satellite nodes, as well as between satellites and the ground stations, would add meaningful latency vs a data center where you've got thousands of servers physically wired together.
I guess they're thinking it's about energy? They can put big solar panels on the satellites and avoid having to rely on terrestrial power grids? It doesn't seem to me like the cost of producing and launching all those satellites and solar panels could possibly be worth it, especially when existing AI companies are still struggling to figure out how to grow demand and become profitable. If you're going to build that many solar panels, why not just put them in a desert on Earth?
I get that Elon is a fraud and the plan isn't even supposed to make sense, but I'm amazed so many finance people are apparently willing to entertain it like it's a serious idea.
Upvote
387
(392
/
-5)
GW is a measure of power consumption, not of computing capacity though, so as a metric it can be met with any number of ways up to and including hundreds of old vacuum tube computers the size of buildings so long as they are computing devices that consume power.
Upvote
117
(122
/
-5)
OpenAI had a big headstart but their competitors caught up pretty quick. Not a lot of compelling reasons to make a huge bet on them being the ones to come out of this on top.
Upvote
72
(72
/
0)
EXACTLY my thought.
The rush to get in on it has hit full stop on their part. That was a HUGE announcement, especially with its impact on prices and the other fallout since then.
But these things are predicated on full disclosure of the financials, and any objective inspection of any AI company's financials is going to reveal a VERY unhealthy picture. OpenAI is less than 18 months away from bankruptcy, and THAT was predicted WITH Nvidia being on board.
This collapse of that deal only accelerates the point of no return for OpenAI's financial viability. I don't know by how much, but they had $16 billion in revenue that was being burned at a substantial rate. Without Nvidia there to migrate the operational costs, they have less time now. POSSIBLY substantially much less time, if you consider that the 18 months limit INCLUDED Nvidia's involvement.
The weasel is about to go and the monkey's are gonna have fun.
Upvote
110
(116
/
-6)
Yeah. As I understand it an LoI can document steps towards a contract without any kind of obligation.
Getting an LoI from a big player and waving it around to get investment is not a new trick, but borderline ethically.
How many times can OpenAI and NVIDIA do this without losing face seems to be an ongoing experiment, no?
Upvote
26
(28
/
-2)
At this point, everyone is just held hostage by their bad investments, and have to keep throwing good* money after bad just to keep the bubble from popping. It applies to the market as a whole, but especially to cultish meme stocks like anything related to Musk.
It's a nightmare version of the gambler's fallacy: they can't stop gambling or the cumulative hangover will kill them.
*In as much as ill-gotten trickle-up lucre can ever be considered good.
Upvote
108
(109
/
-1)
I wonder if I can persuade anybody to take an interest in my scheme to build very rad-hard computers out of 6L6 beam tetrodes and carbon resistors?
At around 50 watts per gate you'd only need 20 million gates to get to a gigawatt. And if you put them in space, getting rid of the heat isn't a problem since it would be done by straight radiation, no need for heat pumps.
Upvote
59
(60
/
-1)
I am not.
I've encountered too many software people who just don't understand (fairly) basic physics.
And I think it doesn't help that what in the public mind is a "radiator" is 99 times out of a hundred a liquid to gas heat exchanger, and they don't think through the implications of there being no air in space.
If a car radiator runs at around 100C, radiating heat away from a data centre can't be that difficult, can it?
Upvote
146
(146
/
0)
See also 2008, 2001, 1989, 1929, and whenever the hell the Dutch decided tulips were the most important thing in the world.
Upvote
110
(111
/
-1)
We had a huge discussion in the comments about this when it was first proposed a while ago. I’m actually disappointed that Ars hasn’t done an in-depth article on this. They only make sense to people who believe that “cloud computing” literally happens in clouds.
Upvote
143
(143
/
0)
Nvidia's share price dropped at the news? The news that they could now sell the chips they make at full price instead of giving them to OpenAI at cost?
Huh.
Huh.
Upvote
155
(155
/
0)
Oh, is the money-go-round stopping?
Well that's a shame for those who were pretending that it went somewhere.
Anyway, here in the real world, let's have a look at what's happening...
Well that's a shame for those who were pretending that it went somewhere.
Anyway, here in the real world, let's have a look at what's happening...
Upvote
28
(30
/
-2)
I’m relieved rather than gleeful, the sooner the bubble pops the less damage caused
Upvote
67
(68
/
-1)
DC hardware lifecycles are short. I don't see how it's economically viable to put all this hardware into orbit that you'll need to refresh in 5-7 years because all the latest chips have lapped it. It seems like just the replacement/refresh cycle alone would not only be continual but utterly unfeasible financially.
It's not like you can just load up a delivery truck and haul it 150 miles to the DC. Rockets are still pretty expensive to launch.
Upvote
60
(63
/
-3)
Oh, that's easy. Tesla is going to fab them. It'll just take 3 or 4 years instead of the usual Musk "2 or 3" years. I mean yeah it's hard to build a fab, but he can just build a million Optimus robots and send them walking out into the Nevada desert and they should get it build lickity-split.
Copied from the latest earnings call writeup on Shacknews:
- I think Tesla needs to build a terafab
- In order to remove the constraint in 3-4 years, we are gonna have to build a very big fab that includes logic, memory, and packaging domestically
- and that is to insure we are secure from any geopolitical risks
- geopolitical risks are going to be a major factor
- that's crazy, fabs are really hard
- We do a lot of hard things
Upvote
84
(84
/
0)
Giant floating balloon data centers! You could use all the hot air generated to keep them afloat!
I guess Google already tried this once for internet access.
Upvote
64
(64
/
0)
This bullshit stock-market-only economy is the only reason there isn't chaos in the streets over this disgusting fucking administration.
Upvote
132
(136
/
-4)
I will pay eleventy squijillion dollars for your entire production capacity indefinitely.
Unrelatedly, I have elevent squijillion dollars worth of 6L6 tetrodes and carbon resistors.
Upvote
54
(54
/
0)
Its such a strange metric to use, unless you're in the energy business, in which case it seems like a perfect way to view datacenters.
Did you know Subaru sold between 28.5 and 31.0 million horsepower of Outbacks in the US last year?
Upvote
237
(237
/
0)
Your mistake was assuming that finbros are intelligent or aware of science or engineering concerns. Most would consider actual expertise in either field a drawback for the big financial movers. They deal in raw numbers and let plebes like you and I figure out how to make it work, if at all. And if they lose their whole bet, the government will cover the loss to prevent total economic collapse.
Upvote
130
(131
/
-1)