Nvidia overtakes Apple as No. 2 most valuable company | Hacker News Reader

Nvidia overtakes Apple as No. 2 most valuable company

reuters.com

46 pointstestrun28 comments

If anything, this tells me that NVIDIA is overvalued and should face a correction soon-ish once/if the AI hype flattens down a bit

But then again, I don’t know anything about market dynamics. Just judging based on what each company does in the world

It’s not and it won’t. This tech bubble narrative tends to be more wrong than right and nvidia has so much tech so sell.

Sure heaps of money will be burnt on hype but the hype exists because of the truth, AI is capable of creating more value than all of humanity put together if done right.

Really? I think we're already in the "two more weeks" zone with generative AI. Still waiting to see what concrete value it adds beyond blogspam. Where is the money going to come from?

AI is both further and closer than you think.

I have spoken with dozens of other academics who have been using it to iterate both quickly and effectively in both their research and teaching.

AI hasn’t seen positive ROI yet but it is producing work of value already.

We're talking about a 3 trillion market cap here. If it makes every academic in the world 50% more effective, that's still like nothing, and it still doesn't answer the question: where will the money come from?

A $20/month subscription to ChatGPT will fuel trillions of dollars of new value? Is every household going to start spending 1000s of dollars per year on AI like they do on Apple products? Laying off a few hundred thousand call center workers in India?

I don’t think I have purported to have the answers to those questions. Just that AI has the potential to change every job on the planet and there is a huge amount of money to be made.

Unless your job requires you to think or act physically, then yes, you are in danger

there are at least 20 companies, maybe 50, who have received founding rounds of $300M+ with the stated purpose of those funds to buy nvidia cards and run them in their own colo space. there’s even more established companies who can do this without funding rounds. as long as the cards are getting better every 2/3 years then if you want to be a player in this space you must buy a new $300M+ cluster every 3 years or get left behind.

there’s enough vc money to ensure some don’t get left behind. and there is one ai card company in the world

this only slows down when someone else can compete with nvidia and i don’t see any thing close

Nvidia is definitely overvalued, but the AI revolution is real. Nvidia will see a huge correction in their stock market price soon-ish, as it’s overvalued even if one barely factors into account AMD and Intel GPUs being decent competition.

Once we start to see diminishing returns to model size and model performance begins to plateau (we are already nearing the plateau) we will have a set of models all pretty close to one another in performance which will be made available for downstream workloads on demand.

The consequences of this will be twofold. First, there will be less incentive for new entrants to train proprietary models as performance across SOTA models will be uniform and thus less possibility of differentiation and gaining market share. Second, there will be less incentive for SOTA model developers to invest heavily in GPUs because of diminishing returns.

Now the beauty of LLMs is that inference requires orders of magnitude less compute than training. So once we have a stable of performance plateaued models, compute demand will drop off a cliff.

Once we get over this mad rush to train train train, GPU orders are going to crater.

The two critical assumptions I'm making are 1) we are rapidly approaching a performance plateau and 2) models other than LLM's are much less widely useful than LLM's. Image generation I see as a fad, I played around with it a ton last year but haven't touched it since. Other niche models will have very little general applicability compared to LLM's.

I doubt there’ll ever be a time within the next 15-20 years at least where we’ll see a point of diminishing returns of increasing the model size. The reason is that we are still just scratching the surface on the input size to the model, 128K tokens of text is nothing compared to few frames of a single-stream HD video.

Discussion [0] (67 points, 6 hours ago, 105 comments)

[0]: https://news.ycombinator.com/item?id=40589585

We are more than due for an AI tech correction. NVIDIA is increasingly getting overvalued if not it already is due to the unsustainable AI hype with low competition.

All hanging on the geo-political risks and US tensions with China around Taiwan, risking disruption to the semiconductor industry and over relying on Taiwan as the top supplier.

If it were me, I would start selling. History has taught us that the same happened with Intel in the 1990s when it peaked with the PC revolution and the same happened with Cisco when it peaked with the growth of the internet.

Very concerning.

[dupe]

Some more discussion: https://news.ycombinator.com/item?id=40589585

I'm wondering what will happen when competition finally shows up. While I'm not an expert, I'm sure cloud providers would love to stop paying high prices for hardware. CUDA seemed like the roadblock,, but that can only last so long right?

Disclaimer: It's been a while since I've worked on an implementation involving GPUs.

Nvidia has released Enterprise AI suite in 2022 as the first company in the world.

While everyone thinks that CSP will catch up with Nvidia in chips, Nvidia is actually working in displacing CSP in their domain which is SaaS. Nvidia has already NIMs which go into the category of AI development/deployment as a service.

Here's the theory: We are only at the beginning of the AI revolution. Every single job that requires a brain will have that brain replaced with NVDA chips. As of now there are still far too many wetware brains still working. That's addressable market for NVDA.

Every job that does not require a brain will have <..>

It might be overvalued, but I am very glad I bought a few shares of Nvidia in March of last year. I'm wishing I bought more in hindsight, but a win is a win, and hey if they do that 10-to-1 split I'll have a lot of shares and I'll get to feel important.

Nutty, considering they have not one but two competitors (Intel and AMD) with strong competing products. The assumption that they’ll continue cranking out such margins in decades to come is comically wrong.

And Apple.

Listening to bits of the keynotes and he likens the "NIMS" to be prepackaged software (ala Microsoft in the 90s) that can run on any PC. It sounds to me like he is striving to be a monopolist.

Any reason why Nvidia doesn't expand into building their own data centers and cloud platform? Seems like the logical next step once you own the chips everyone wants, giving them more moat. Of course, that's not their core business and maybe not their expertise, but I'd expect they can acquire it.

Saved you a click:

"Nvidia's stock rose 5.2% to end the day at $1,224.40, valuing the company at $3.012 trillion. Apple's market capitalization was last at $3.003 trillion after its stock climbed 0.8%.

Microsoft (MSFT.O), opens new tab, based in Redmond, Washington, remained the world's most valuable company at $3.15 trillion after its shares climbed 1.9%"

One makes the largest computation infrastructure to run language models.

The other makes overpriced GPUs.

Which one is living in a bubble?

Who are you talking about ?

Reply on news.ycombinator.com