Amazing how high expectations were that a 56% jump in sales still led to a 3% drop in the stock.
It's very clear that the ramp for GPUs continues.
It’s the top valued company so its floor is being #1 ($1 ahead of #2). That’s what the market is saying. No one is able to tell its real value because comparatively there is nothing beyond it. The market is only saying that it’s at the very least $1 more than #2.
As unreal as it sounds, it could truly be worth trillions more, it’s just that the market is unwilling to say so atm. There’s nothing normal going on because almost any stock you ever looked at is surrounded by hundreds of other stocks to compare to.
People missed this stock pre-split, then at 90, 120, 140 … And they will miss it again.
NVDA had a surprise earnings on May 24, 2023, closing at $305 and opening at $385 (before the 10 for 1 split). Pretty much every earnings day since then has been the same - the numbers come out, they beat what they estimated, the stock goes down a little. People have been doom and glooming it every earnings call - you can read the threads here from May 2023, people were saying it was a bubble then, and it's over four times what it was then.
[dead]
Long ago stopped trying to keep on top of GPU releases at home as the gains were too incremental and releases too frequent for my wallet. That trillions of dollars may be spent on this is great news for Nvidia, but seems like a source of incredible quantities of parked inventory and landfill if not managed carefully.
The idea that the world will ever have enough compute is ridiculous. I'm sure all of this sudden growth will need to be digested at some point, but that won't mean the industry will permanently lose a big chunk of its market cap.
Same argument can be somewhat applied to CPUs circa 90s. The growth did stop/stalled in the end.
I think the expectation of ever growing compute is not totally crazy. It will come with lower margins eventually though, and more players in the market. It also might get much more moderate, including from hardware limitations. Efficiency wise h200->b200 isn’t as crazy as a100->h100.
Same argument can be somewhat applied to CPUs circa 90s. The growth did stop/stalled in the end.
Would this work once the funds start to dry up and the business has to self sustain, at some point?
Assumption is that business become profitable, and not just break even profitable but profitable at their expected multiples, isn't it?
Not sure if it's profitable but I don't think intelligence can be compared to just CPUs in dotcom.
Given enough compute, you can a very smart LLM to work on any number of things by itself.
With all due respect, do you folks actually think deep and hard about what you write before you do?
I get the impression that you don't. This isn't meant to offend you, since this stuff is really hard to do. But I notice this a lot - wishy washy comments, that lack deep thought.
This place would be better with less posts but more thinking.
When dot com crashed it was similar - compute use didn’t reduce, far from it, it’s just capex crashed to the ground and took 90% of Nasdaq value with it, for a few years.
Stop comparing to dotcom. There’s a big difference between the internet then and the internet now. Ignoring the astounding nature of the AI tech (dotcom isn’t even in the same ballpark), you also have to consider these two things:
1) We didn’t make the world digital native in that era
2) We didn’t connect the whole world to the internet
The desire for it may be without limit, but the desire to spend money on it definitely has some limitations.
[dead]
I can’t wait for how cheap compute will be during the inevitable AI winter.
It's all but inevitable that compute gets cheaper.
AI winter -> Datacenters have too much compute to sell -> Compute gets cheaper
AI summer -> Hardware companies are incentivized to make better chips-> Compute gets cheaper
The second actually is more powerful in the long run. The AI boom has actually made compute very cheap per flop.
Will it work out the second time around?
As long as these data centers and their cloud patrons still keep providing those compute options, unlike, the 2000 crash where commodity hardware was widely available and the business use case was comparatively decentralized.
unfortunately mostly useless 4 bit floating point
How is it useless?
If you try to do any sort of scientific calculation (ex. physical simulation), at least 32-bit floats are needed to ensure adequate numerical precision. Ideally 64-bit floats are the best, but FP64 performance in recent GPUs are bad up to the point that you can say they're not supported anymore...
It will slow down, wait for it.
Wouldn’t this be a lagging indicator?
There seems to be no limit to the size of this bubble.
i really only started hearing about AI bubble this year. with the housing bubble, people were talking about it in 2005 and it took years for it to finally break. maybe AI bubble is the same
Except the housing bubble wasn't actually a bubble. Housing prices today are much higher than in 2007. Based on realized results, the mispricing wasn't the peak of the housing bubble but the trough of the crash. With the benefit of hindsight, the peak actually undervalued how expensive housing would get. (Speaking about the US market in aggregate. Not other countries, nor localities like Las Vegas where the case for a bubble is stronger.)
If you adjust your chart for inflation you see a big peak in 2025 and another in 2022. The trough of the crash matches the historical trend.
Yes, the price of housing is crazy. But so is the price of eggs.
Eh.. "peak in 2005 and another in 2022"
The price of housing is ultimately a political choice. If voters choose to restrict supply of housing through zoning and other laws, prices can be pushed up hugely.
I doubt zoning has any real effect. Henry George realized in the 19th century that if you increase the money supply most of that money goes into driving up the cost of real estate. And the government has been printing money like crazy for decades now.
Where I live there aren't too many restrictions on housing, but a developer friend tells me he can barely make a profit at current prices because of 1) the cost of the land and 2) modern building codes make housing a lot more expensive. You can have cheaper houses if you're willing to give up things like the insulation that forces builders to use 2x6s instead of 2x4s for exterior framing.
That NIMBY stuff is mostly a problem for pricing if you want to live in a big city.
Boom and bust capitalism is completely orthogonal to AI as a technology. Capitalists are sociopathic opportunists that exploit every aspect of life, that doesn’t have anything to do with AI.
Housing bubbles existed, we still use houses. The dot com bubble existed, we are still on a dot com.
What makes you think the capitalists won't train a capitalist AI?
Stock is down 3% after hours.
Taking it all the way back to the price a week ago.