Nvidia hits $2T valuation as AI frenzy grips Wall Street
reuters.com
reuters.com
Every provider of "AI as a Service" is spending money like a drunken sailor who just came back after spending six months at sea. None of these providers are charging enough to cover the cost of development, training, or even inference. There's a lot of FOMO: no tech company wants to be left out of the party.
Nvidia, as the dominant software-plus-hardware platform (CUDA is a huge deal), is the main beneficiary of all this FOMO. A lot of unprofitable spending in AI ends up becoming... a juicy profit for Nvidia. We're witnessing large-scale transfer of wealth, from the FOMO crowd and clueless end-users to mainly one company.
The $2T question is: Is all this FOMO sustainable in the long run?
First off, 3D games existed before NVIDIA. NVIDIA wasn't even first to market for making dedicated 3D acceleration hardware.
CUDA existed before Bitcoin and was used for physics and chemistry simulations when it first came out.
> "Compute that does nothing important"
Entertainment is important, so to just toss 3D games aside is asinine. I'll agree that crypto-mining is completely worthless. As far AI, certainly there's a lot of hype around it right now, and we're probably gonna see a lot of AI startups go belly-up in the next couple years, but it's definitely not a fad and isn't going away.
Being highly dismissive doesn't make you cool or edgy, just so you know.
Beyond that, there was no crystal ball, rather a useful physical product and software tools to make it accessible beyond a highly specialized set of users (I am not saying CUDA or anything is that accessible, but it makes GPU's useful to more than just game devs).
I've predicted already the AI hype train will crash into the wall of reality in more or less the same way the crypto one did, and then the grifters will move on to the next thing they can over-hype.
Cisco at its peak in 1999 had a P/E ratio of near 200. Nvidia trailing 12 months is closer to 50. ( Today it is closer to 60 ). So I dont think Cisco is a good comparison.
https://www.cisco.com/c/dam/en_us/about/ac49/ac20/downloads/...
That’s a P/S of 29 (but one quarter in there is forward looking)
NVDA’s year ended Dec 2023 saw $61bn revenue. All 4 of quarters in there are backwards looking. If you shave off the oldest one (7.2bn) and add on this current quarter (22bn forecasted), you get 75.8bn.
Or a P/S of 26.4.
It’s closeish
How much revenue can nvidia realistically have 5 years from now, and at what margins? 100bn in revenue (NVIDIA grossed 25bn last year, so 4x top line growth) at 35% gross margins = 35bn. Slap a 25x multiple on that and you have a 900bn market cap. Discounted at 8% to today that's .65 * 900bn is 600bn. Not 2000bn. Are my numbers too pessimistic? Probably.
But still, it's hard to come up with 2029 numbers that justify a 2000bn market cap today. Even with a 5% discount rate NVDA has to be worth 3000bn 5 years from now, for current investors to get a 0% return.
Q4 revenue was $22B, and guiding $24B for Q1, so likely $110Bish revenue in 2024.
Their net income should be in the same ballpark as Microsoft / Apple / Amazon / Google in 2024.
Meanwhile, Apple / Amazon / Google are growing ~10% YoY, at Nvidia's current growth rate they could surpass Mag7 net income in 2025/2026.
For nvidia to have 100bn in net income they would need at at least 200bn in revenue. Apple has 100bn in net income on 400bn in revenue, for comparison.
Ultimately the question is whether nvidia's moat (chips + cuda) will be durable. My guess is competitors will arrive soon and that nvidia's margins will go down. Right now nvidia seems untouchable, but reality will set in sometime next year.
- total market size will go up - Nvidia's market share will go down - Nvidia's profit margin will go down
The trillion dollar question then is whether the uplift from #1 is bigger than the downdraft from #2 and #3.
I think you are off by a year because NVDA’s fiscal year is 1 year ahead. NVDA’s fiscal 2024 just ended (ended Dec 2023) and was announced this week.
Their latest quarter (ended Dec 2023, fiscal Q4 2024) had 22.1bn in revenue. That’s more revenue than Cisco’s full year in 2000.
Nvidia’s fiscal 2024 just ended (ended Jan 2024) and was reported this week.
They are currently in Q1 2025.
The Company that Broke Canada - BobbyBroccoli https://www.youtube.com/watch?v=I6xwMIUPHss
But their marketing reports know EXACTLY what to do. Just add a blurb in any press release about "AI enabled x, y, z for our a, b, and c product lines" and the board and other execs are happy.
Meanwhile 9 times out of 10 any feature that gets rolled out will be a collection of shell scripts that simply analyzes some data and spits out an answer, any answer, and "we AI enabled."
Transformers and diffusion models have already completely changed the game and the architectures are still very primitive. They will only get better from here.
The crazy part is that for the foreseeable future, the limiting factor for improvement is compute. Train the models for more epochs is all that is needed, for now. Moores law seems to be coming back for another round.
I have no doubt that society will benefit from all that spending...
but I'm much less certain about the fate of all the companies incurring all that spending!
What I said is that they're spending money, for a variety of AI and AI-related services provided by third parties.
Those third parties, in turn, are buying GPU's.
Anyone who's worked in tech for a few decades knows that what the OP describes happens regularly.
There's nothing patronizing about sensing the pattern forming again here, and recognizing that the only sure thing is that the doe-eyed forty-niners racing to California all need pick axes and camp supplies and that the axemaker is going to get rich on their dreams no matter how things turn out.
It doesn't suggest that there's no gold out that or that the game theory calculus that sends everybody chasing that gold is unsound. No individual is being called stupid.
It just emphasises that a lot of people are going to make the wrong bets on AI, using more money than they really could afford to lose, and that NVidia is the only guaranteed (absurd) winner so far. There's nothing patronizing or even contentious to that.
What's not clear is if they'll get 20% better or 500% better.
How did it go with bitcoin and cryptocurrency in general? That seems the apt comparison.
There's so much money being thrown at anything with "AI" in it at this point - definitely FOMO fever. While I think Nvidia will do just fine (as the primary maker of picks in a gold rush) it seems like they're going to need to go after yet other markets - the trajectory so far has been gaming -> crypto -> AI. What's coming after that?
It is not going to be a differentiator.
Every worker will need to be equipped with an AI assistant or the company is pedalling a bike on the freeway.
AI is a tool, like a laptop. Does having a laptop make you special? Nope. But you MUST have it.
The same would go for many other companies, especially in the tech sector.
Almost 20 years later, almost none of it has come true.
Maybe it is irrational, but at this point, it feels like the irrationality is likely to outlive anyone here.
After 2008, AAPL was trading for something like 7x earnings, GOOG same, MSFT same.
The funniest part is that people genuinely don't understand that these examples disprove the point they are making.
(And yes, NVDA is obviously overvalued...)
All I know is that it's been one of the terms being thrown around during the past 20 years as a reason why Apple's stock price will plummet to the ground and why I'm stupid for investing a large portion of net worth in it.
At this point, my upside is immense so unless something absolutely catastrophic happens, odds are it seems I'll be having the last laugh.
EDIT: I guess I don't understand the "gives you $1" part if we're not talking about dividends.
But in the real world, company valuation is an entirely subjective matter that prices in expected future growth or losses.
> it earns (or rather pays out to in that period) the shareholder
Comment I replied to was calling dividends, and only dividends, within a given year we're calculating P/E for no less, the company Earnings. That just isn't correct, whatever your views on valuation, shareholder ownership, and market efficiency.
Earnings thus either become dividends or become re-invested or are used to buy back shares.
P/E is the central metric by which to judge stocks on a fundamental level.
Profits are not the only thing either. Some assets give political power or power to shape the world. If NVidia made 0 it would still be very valuable because of that.
Others would simply tell you that it is:
(share_price / earnings_of_that_year) == price_to_earnings_ratio
Whether one should or shouldn't trade based on a particular P/E ratio has always been a matter of opinion.
It's based on market cap, not share price. Market cap is a function of share price (It's just share price * number of outstanding shares), of course, but to just say it's share price leads to misunderstandings.
Otherwise, that would imply that a stock split creates a multiplier of the P/E, ie, a company with a P/E of 20 does a 1:5 split ends up with a P/E of 100 post-split, and that's certainly not what happens.
P/E is defined usually defined as `share price / earnings per share`. Market cap is `share price * number of shares`. Via simple substitution, this means P/E is equal to `market cap / earnings`.
But calling P/E `share price / earnings` is simply incorrect.
In short: net income is total, "earnings" is "per share"
2005 it was a gamble I have no idea, and if they hadn’t invented the iPhone they would have been DOA.
To some degree, they will need another revolutionary product as iPhones seem pretty mature, and I think they are still a little unsteady on that front judging from Vision reviews and sales.
There is no rational universe where a company can be worth almost $3T, and at the same time, investors know there is a ~% chance every year that Taiwan gets attacked or blockaded, and that company won't be selling anything for half a decade, minimum. It's like betting on the Death Star fully knowing the exhaust port issue exists. Complete with "but surely our fighters (er, US government) will prevent anyone from actually hitting that port."
At this point, I just believe the entire stock market is irrational and in the stratosphere; simply because, where else are you going to invest? Too many investors, not enough companies. But what goes up...
(Edit: I'm primarily referring to Apple here. I am aware NVIDIA primarily uses South Korea - but at the same time, if something were to happen to Taiwan, even NVIDIA will need to prioritize fab capacity.)
another million dollar question: is the likelihood of another magnitude ~8 SF earthquake priced into nasdaq?
Just because some 1% chance that a new Jedi comes along actually happens doesn’t make it a bad bet.
In the market timing is both 1. everything and 2. impossible to predict without cheating.
Short term, though? Nah, you should absolutely trade on margin.
I started investing in 2020 immediately after markets hit rock bottom from COVID. Margin multiplied my gains as I bet on the market bouncing back.
You could certainly argue that I was just lucky. I could have easily been wrong and it's possible markets could have taken a longer time to recover than I expected. But tbh, the market recovered significantly FASTER than I expected.
Risking getting wiped out may be a palatable prospect if you're early in life and have the time to rebuild your fortune, but it won't be if you're risking 20 years of savings.
There were a ton of promising companies back then that no longer exist today. But you don't hear about then in HN.
I only had to sell AAPL/META for a downpayment.
What is the lesson you take from this, though?
To me, the lesson is that "Predicting the future is hard." No matter how well you're doing, you still need to make some sort of prediction of the future to stay on top.
If you get $2 billion and want to stay a billionaire, you can bury it underground and predict that inflation will be minimal, or you can invest it and predict that investments will be profitable. Either prediction could be wrong. (But they're not equally likely.)
In that same vein, even though Cisco was printing money, the world changed. The world will always change. It's not necessarily a failure that they couldn't predict and invent the next money printer.
While Cisco was THE brand at the right moment, they didn't have any secret sauce.
Nvidia though (sadly), the industry has been playing catch-up with them forever.
And this will happen to NVIDIA as well, but it might be a slower pace as NVIDIA tech is more expensive to develop than Cisco's
the very occurrence of speculative bubbles, (to be pedantic; where asset prices soar well beyond their intrinsic values and eventually crash) is a strong argument against market efficiency.
It's more useful to notice that, while everyone and everything is irrational, some are far more irrational than others. That is, it's much more useful to not consider it a binary.
Apple: 2.84T (P/E 28.60)
Microsoft: 3.07T (P/E 37.47)
Nvidia: 2.02T (P/E 67.63)
Quite astonishing.
HFV for Hypothetical Future Value: get ready for the quazillion dollar!
"hypo" means under or less than.
OTH Especially Apple is not doing that great at all (and if above assumptions hold they are much more overvalued than Nvidia is). Their revenue this year is estimated to be lower than in 2022 while Nvidia's more than doubled and Microsoft is growing 5% or so per year.
Yeah but they've tried. Year after year for the last decade. They keep failing. Maybe they're more motivated now? But CUDA isn't a moat because it's unassailable, it's a moat because nobody trying to cross it knows how to operate a shovel
Google is trying for some time to use its TPU, so is Amazon. It makes so much more sense to pay few billions to Nvidia and don't risk the trillions in valuations.
What’s an AI product that’s actually making a lot of money (profit)? I cannot find reliable stats on how much profit OpenAI actually generates.
There’s an awful lot of FUD on HN about this being like crypto. In a decade of trying I never found a meaningful and practical use case for crypto beyond “HODL.” For generative AI we are finding immediately useful applications under nearly every rock we turn over, with increasingly powerful results. These aren’t speculative, they’re allowing us to do things at scales we couldn’t imagine a few years ago. I’m seeing this with most my professional network involved in high end tech. The tool chains suck, and we are limited by GPU capacity as well as inference costs, but these are short term constraints. As patterns solidify the tool chains will too, GPU capacity is ramping up rapidly, and there are plenty of optimizations on the horizon.
I for one see this much more akin to the gold rush of the internet over the 1994-2020 time frame in super fast motion. Will we overshoot? Absolutely. But the effect is real.
I also would challenge your "year and a half" timeline... Gen ai has been around.
This is what hype looks like.
Be the change.
Oh it doesn't? odd that.
ChatGPT cannot make judgement calls like what you're trying to imply it can.
ChatGPT can do some really cool things, but it's not magic.
maybe stop doing that.
The PDF example being thrown around in this thread. There's a magical step in the middle that no one is acknowledging.
You also realize that OpenAI has a dedicated Document Assistant which will literally extract information from a document you upload using prompts? Are you just unable to get that to work? I just don't know what you arguing at this point, like you're watching people walk backwards and then yelling that it's impossible for humans to walk backwards.
When I'm frustrated, I talk to ChatGPT like that.
It works as well for the LLM as it does for the humans in this thread.
What's worse is, I'd been writing some SciFi set in 2030 since well before Transformer models were invented, and predicted in early drafts of my fiction that you'd get better results from AI if you treated them with the same courtesies that you'd use for a human just because they learn by mimicking us (which turned out to be true), and yet I'm still making this mistake IRL when I talk to the AI…
You understand that If I want to extract the Date a letter was sent, who the recipient was, what amount is due on an invoice, etc that I have to send a specific prompt asking that to GPT with the PDF in the context? Do you just literally not know how this works?
With LLM at GPT4 level ability and a modest investment in scraping, context, building an agent based tool you can automate the entire thing with a pretty high precision and recall (such as a 0.9 precision with a 0.8 recall, vs 0.4 / 0.5 for humans). Using classical NLP, schema matching, and classification techniques you could triage and improve human performance by as much as 20%. But that requires a lot of effort and provides important but modest improvements. The LLM approach however is almost effortless from an engineering effort (note all of the infrastructure is the same as the NLP/schema matching/classification/etc effort).
This may not seem earth shatteringly important to you, but for an enterprise this is monumental. It dramatically reduces risk, complexity, cost, and all the while improves outcomes by an enormous amount.
Another example is incident management. Having LLM agents monitoring incidents on slack as reduced our MTRR dramatically. They provide summaries for new joiners, broad cast management updates, etc, and take care of all the glue out response teams typically did but with a high degree of quality, low degree of variability, and extremely low latencies compared to humans.
Another example is policy management. Many enterprises are saddled with thousands of pages of external requirements from regulators, and those in turn create tens of thousands of pages of policy, and hundreds of thousands of LoC of control implementations. By indexing them into a Context+RAG LLM you can create a policy oracle, allowing you to create coverage and effectiveness testing, analyze for policy or control gaps, draft responses to auditors and regulators with total policy knowledge, answer legals questions, advise management on the implications of a business driven exception, etc. The state of the art before was a search engine dumping phrase based matches into a grid of thousands of rows of results, which was effectively useless for all of these use cases.
followed by
> GPT4 level ability and a modest investment in scraping
oh yeah, these speech patterns are those of someone who isn't unnecessarily hyped while trying to downplay other approaches.
no bias here at all folks...
It’s readily apparent from the pessimistic HN threads they come from people who don’t actually do these things for a living and haven’t had much real world experience. These speech patterns as you call them are the way people who “do things” differentiate from those who can’t or haven’t done things. It’s called “experience.”
When claude shannon first came out with his information theory everyone started trying to apply it to everything, discuss everything in terms of his theories.
He himself thought it was ridiculous and basically told everyone to stop it.
This is no different.
Right now, I have to copy+paste my input and outputs, but my next task will be to write a Tampermonkey script to automate it.
I equate their foundational models such as GPT to the iPhone. You can build on top of their ecosystem of models.
Everyone else are just app makers using their foundational models as the base.
It wouldn’t surprise me if they want to build an App Store on top of ChatGPT and take 30% eventually.
I suspect we're going to end up with a wildly different architecture before it's all over.
This is my naïveté showing but aren't GPUs (relatively) general purpose? Do "AI" workflows require that flexibility or would they be better served by special purpose hardware? To use the obvious analogy, what happened with cryptocurrency mining moving from GPU to ASIC.
Nvidia single-handedly carried the United States of America's 2023 GDP from recession territory (0% growth) to unprecedented massive economic boom (6% growth)!
--
https://en.wikipedia.org/wiki/Film_industry
https://www.bea.gov/news/2024/gross-domestic-product-fourth-....
But in all honesty, the box office numbers aren't as impressive as the longtail licensing and merchandising.
That's where the real money is made for the movie franchises.
Also, to the extent that the chips are built in Taiwan, I'm not sure that they count in the US's GDP at all.
People were shouting "P/E" and "Dividends" back then for a few years as well thinking it's crazy AMD approaches Intel's market cap.
I was aware that ChatGPT was a gamechanger when it was released. I was also aware that Nvidia was the only one making the hardware.
But I'm just not the type to immediately jump to "How can I best profit from this?" Rather than researching stocks, I researched transformers more.
The fact that OpenAI valuation raises was very understandable (but not easy to invest in practice + the corporate structure is shady).
However, Nvidia, it's a very optimistic hype, as the cards are very likely to get replaced with more specialized AI chips in the near-term.
Probably by Apple first, then whoever wakes up.
Also, once you built your big datacenter with all these cards, then you don't renew them every year.
market reacts rationally - just not your brand of rational.
You could honestly make a hedge fund by listening to me pontificate things and then betting opposite of me. I have been bearish on MS, twitter, dropbox, bitcoin, the list goes on. It's admirable how I am so consistently wrong.
Same thing with BTC. I tried to buy 20,000 in 2010. Was too much hassle back then so I gave up. But even if I went through with it. I would have sold the moment I turned $20 in $200. Much less hold past $500 or $1000.
I am no expert but it sure seems like Groq might be on to something with their LPU (language processing unit).
From what I understand, if someone made an ASIC for LLM inference, such a device would effectively have the specific model locked in. You wouldn't be able to update the ASIC for newer versions of the model that alter the architecture. At best, you might be able to make one that allows you to update the weights.
I'm also not sure how cheaply you could produce such an ASIC. One that runs a 7B parameter model probably wouldn't be too expensive, but something that could rival GPT-4 would be.
Yeah it's the exact same scenario, except for... you know... the gold part. Everything is a gold mine when you print money like there is no tomorrow and your entire economy is based on debt
They would never dare doing that again, right? Right?!
* transformer architecture can continue to scale to become good enough and fast enough for real use-cases
* useful real-life applications (or even operating systems) will be built on this infrastructure
* alignment, data governance, privacy, or hallucination issues are solvable (or at least mitigatable).
* supportive regulatory environment, limited legal liability for users.
* China won't take advantage of the current geo-political clusterfuck to move on its ambitions for Taiwan.
I'm sure there is more, basically there is a lot that needs to happen or not happen for Nvidia to continue to be the perceived winner here.
I remember this famous AI professor Andrew Ng in an online lecture, talking about his "AI dream" and his enthusiasm to reach general intelligence.
Even John Carmack boarded that ship, and I don't see he is going to deliver, and honestly I don't really trust him to deliver some real skepticism about AI if he doesn't deliver something good.
We really need more skeptical professors and experts to talk more on how AI is not as great as it seems. The inventor of SIRI did a great talk about this.
Apparently people seems to confuse and conflate "making quality deepfakes" and "AI is useful".
It's weird when I am a developer, and I really want LESS technology.
Excellent point, as the broad output of most of this so far is just more fake garbage that clutters everything up with generated spam (except to be fair the code tools like copilot). Reminds me of 3D printing, seemingly under-delivering on the massive hype in terms of "what can it actually produce". Recent VR hype feels the same in relation to expectations and cost of the products vs usefulness.
IF you really can't see any utility for current gen GenAI beyond deepfakes there's really no hope for you.
Even ignoring 'general AI' (whatever that is... no one even knows what that means), that's already enough utility. Just the searching ability makes it easily more valuable than Google and Bing in their original incarnations.
The massive demand is priced in as Jenson indicated, it’s now a bet on how much more past the next quarter.
Revenue of its competitors.
Intel ~60B
AMD ~20B
Broadcom ~35B
Qualcomm ~35B
Mediatek ~13B
On the other hand: I feel like this is a bubble that’s going to pop. Everyone is promising AI futures and I just see a skirmish to bring a product to market that will leave a lot of folks crashing.
Either way, NVIDIA wins.
I am amazed how much the rest of the silicon industry has slept on this.
I know AMD have their competition, but their GPU software division keeps tripping over itself.
Intel could have done more but got complacent for a solid few years.
Qualcomm feels like a competitor in the making with Nuvia but is late to the dance.
Apple is honestly the biggest competitor in terms of hardware and software ecosystem , but refuses to get into the data center.
And of course I’m leaving out the Chinese companies like Huawei who would be a force to be reckoned with if it weren’t for sanctions. But China is also a hotspot for AI/ML.
How well Nvidia can hold its current valuation depends on how useful the new AI/LLM ecosystem will become. Just chat apps and summarizing documents aren't that IMO.
Pixel phones extend their services.
Selling TPU units reduces the differentiator for their services.
I think AI will burst though. So few companies have a novel use, but all of them are promising a future they can’t yet reliably deliver. How much capital runway is there on this?
I’d bet (figuratively of course, as if I had any money) that AI will see a lot of casualties by end of year.
It's hard to imagine the future but the imagination of the masses never ceases to amaze.
That's not to say democratizing abilities is not a worthy goal. I just don't think many of the companies rushing into this space really have a product vision that will end up being mass adopted.
The real key will be client side inference. And as more chips go the route of the Apple ones with onboard inference, I think it’ll be more appealing to the mass market.
Running LLMs locally on high end GPUs is going to always be niche.
Are they? It's not that obvious that they could necessarily scale up their low-power integrated GPUs (as fast as they are) to be competitive in the datacentre. AMD and Intel at least are trying and have/are developing some products.
And I'm not sure about their software? What do they really have that Intel/AMD don't?
Given their power efficiency, they could just go wider and come out on top.
And in terms of software, metal compute is the only real competitor to CUDA backends. Between their PyTorch backends, MPX, and large memory, a lot of ML engineers are going with a combo of Macs locally and NVIDIA in the cloud these days.
Intel and AMD don’t factor in at all comparatively for ML use because Intel doesn’t have a GPU backend for most things people want, and AMD has repeatedly made a mess of rocm
They are actively stepping on every rake there is. Eg they just stopped supporting the drop-in-cuda project everyone is waiting for, due to there being "no business-case for CUDA on AMD GPUs" [0].
If the US wants to win in Ukraine it needs to print like a mad man, look at the SPR; it tells you the truth (albeit with 3 months lag for mere mortals).
This is the end game, no matter what the eternal growth crowd says.
How money (and empires) die.
Better ?
On the other side the US gets free weapon testing range, weakens russia like never before, keep the military industry afloat, use all their soon to expire/expired stocks, sign military and construction contracts for the eventual post war rebuilding, &c.
https://www.forbes.com/sites/niallmccarthy/2019/09/12/the-an...
[0] https://en.wikipedia.org/wiki/The_Geographical_Pivot_of_Hist...