I think you're right because yes they make the hardware behind the products, but they're not making the products, which is where you'd think the actual money should be.
On the other hand, it will be interesting to see how much money there is in "intelligence" the more abundant it becomes.
Who is competing with Nvidia right now? if nobody is then they can basically dictate their own price with the kind of demand there is right now for GPUs.
This is dot-com rush all over again, but it's worth keeping in mind the dot-com rush (with some poetic license) gave us Google, Amazon, Meta, and much of the value of modern Apple+Microsoft.
Those are five of the top ten companies in the world by market cap.
Not many more dot-coms on the list of top 100 though:
It was a winner-takes-all market with a few big winners and a lot of losers.
There's a lot of product to be made. AI is already transforming a lot of businesses. It's too early to tell who the big winners will be, how the market will be structured, or what will happen.
I suspect, in the metaphor, NVidia is the Cisco or the Sun Microsystems. They power the gold rush. Cisco is #60 on biggest companies, and Sun died 14 years ago. I suspect NVidia is now overvalued, but I suspect it should still be highly-valued.
I suspect things will self-correct if AMD, Intel, or one of the startups gets software in shape. The problem is that as much as NVidia can maintain competitive advantage, competition is likely to cap profit margin. If Intel is 30% less efficient than NVidia in compute-per-dollar, than NVidia's profit margin can be at most 30% more than Intel.
The way I see it right now, the situation with NVIDIA and AI companies is like shovel sellers and gold miners. Picking which gold miner is going to strike gold and which ones will go bust is difficult to tell, but knowing that all of them will need shovels/mining equipment means that shovel sellers are a safe but extremely rewarding bet for the next little while.
maturity of the CUDA environment, continued good engineering and execution, a world-class organization (probably the best hardware team in the world) with a business savant at the helm and a founder-mindset and a truly agile and flat team structure, etc
what's the price for apple after introducing the iphone? visionary leader in charge of a company that just delivered a money-faucet technology that - while yes, competition will emerge - they will still inevitably end up owning a large and highly desirable stake of in the market?
cuda is literally 15 years ago, and people are just realizing they have to actually start by catching up to that. The "NVIDIA is an AI company now" was like 2015, "mid-2010s", it was a realization jensen had during alexnet's rise, in the early/mid 2010s. It truly will take a long time to build that kind of ecosystem, Apple's support is great with Metal and it still is fairly well a potemkin village until there's more traction on apple gpu support. And a lot of competitors don't even have that, AMD is only finally just realizing this and Intel is also spending like crazy to build out OneAPI/SyCL.
Best time to plant the tree is 15 years ago, and Jensen has repeatedly planted trees where they needed them to be successful in the long term, and moved the engineering and business sides to make it happen.
Programmable shaders, CUDA, tessellation, GSync, and tensor/ML (and RT at the same time) have all been big big bets, and many people criticized some of them real harshly at the time, or backseat-drove that NVIDIA should have given it away, etc.
Also, just in general there's really only a handful of companies that can build something like H100/H200 or MI300X or Ponte Vecchio and get an actual competitive product out of it. Let alone the rest of the integration in the system (multi-node, networking, etc) which is all increasingly necessary to build HPC systems at scale etc. NVIDIA paid for mellanox, and amd paid for xilinx, etc. "just" integrate all that successfully into something that clients can start building products on, when can you deliver a from-scratch competitor? lol
(shoutout to oxide though, that actually does look pretty sick, bryan cantrill is in charge so you know it's gonna be good)
There's definitely an enormous amount of key-man risk, but it's also certainly the best cross-disciplinary team (hardware, software, business) in the business today. Nobody else can identify what's gonna be big, lay the groundwork, execute successfully, and roll with the punches (and manage inventory) 90s-tech-business style.
What if you could invest in cray, but it was a business where cray had repeatedly executed successfully and managed to pioneer several large commercial applications for their systems, and expected to continue pivoting and innovating into new areas going forward? There is nobody else in the business like Jensen right now, he is a savant.
that earnings remain same as this quarter for a few quarters (higher revenues, lower margins ofc). forward P/E is 30-40 range, typical for tech companies.
Apple has been making a bank on iPhones, with no major leap frog in 10 years.
Nvidia could do that too, hard but not unimaginable.
Over the last year they've seen a 3x jump in gross profit up to 44b on 16b cost of revenue with no signs of going the other way any time soon. That's where I'm guessing the price is coming from
Supply can be fully constrained and future expected revenue can still go up if demand or capital availability increases, because higher demand and access to more capital mean that NVIDIA can charge more per card and still sell out their entire supply. The hypothesis would be that AI research precedes GPU demand that increases NVIDIA's profit margins, so if you're seeing AI research that you're judging will increase future demand, you should expect NVIDIA's value to go up.
Nvidia is no longer selling just expansion cards. They are selling full cluster solutions with Infiniband networking (Mellanox acquisition) and even CPUs (Grace). I wouldn't be surprised if they put most of their GPGPU chip production in these.
And they are much better than the competition, too. AFAIK, only Google and Microsoft have anything competitive in terms of full architecture (but internal).
Nvidia is heavily overpriced at the moment. My personal prediction is that in the long term, Nvidia can expect a similar situation to what happened to Tesla, which involves a relatively slow price correction after the AI hype diminishes. The EV hype slowly died out, and we can see it reflected in the prices of all EV companies.
Tesla still seems to be over-valued. I don't see why they would be worth 10x other car manufacturers, especially ones that have quality manufacturing already down.
It's not. Shorting bubbles or frauds is not a good idea, unless you find an instrument through which you can do it for a long time, without paying too much premium. This is very rare.
Not based on the fundamentals. Tesla maintains a high profit margin and doesn’t carry immense debt loads like the traditional manufacturers. Additionally Tesla has some degree of self-driving success priced in. Traditional auto also returns most profit as dividends and the stocks do not grow.
They still sit at roughly 50x PE Ratio though. You'd have to expect such high growth (way more than even they forecast) to make sense of buying into that if you're not reliant on making money on hype and brand awareness, which is brutally fickle.
Is there a reason why you didn't include their current battery production operations and future plans in this regard, along with solar, along with the TeslaBot that once far enough along will basically create a feedback loop that will make practically everything we currently do "free" to do?
The TeslaBot with even a dumb level of "AI" for learning and mimicking actions the Bot is shown to perform and can mimic based on its non-meat stick appendages - whole systems and funnels from farming to getting a cooked meal on the dinner table could be (eventually) fully automated with little to no human oversight, where the only initial inputs are raw-natural resources; the problem then that 100% of humanity needs to focus on is then heavy oversight and real-time in-person witnessing that such systems aren't being corrupted-captured by bad actors.
There are other synergies with companies Elon is involved with that people seem to fail to see the synergies of, including but not limited to his effort towards colonizing Mars - of which there will be plenty of ongoing
Elon also controlling Starlink, assuming every Tesla vehicle will have capacity in future for their vehicles to be connected, they'll be able to undercut competitors - and will also act as downward pressure on competition to not overcharge; I believe Elon stated his goal is $5 billion/month from Starlink subscriptions to pay for his Mars efforts, with the first organization to get space mining operations into play - able to extract and send home refined product - has then tapped into the unlimited abundance of the universe, the profits of which reinvested fuels progress along an exponential path; once these systems are in play and self-sustainable, reaching that point by being a better offering, value-price wise, then that frees up time to focus on launching the next puzzle piece - each puzzle piece compounding with the others; his "free" time, fortunately for us, most recently leading to him somewhat impulsively - but clearly his impulses are existential crisis-importance driven - he bought Twitter to protect free speech and unveiled-unmasked the corruption and treason via releasing the Twitter Files.
Elon's arguably 20 years ahead of everyone else in regards to the holistic possibilities that his ecosystem of companies will allow for, and arguably every year there is the possibility he leapfrogs another 10 years ahead.
> TeslaBot that once far enough along will basically create a feedback loop that will make practically everything we currently do "free" to do?
How can this happen? This sounds like some serious hype.
What I find really funny about the TeslaBot is that it looks as capable (if not less) than Honda’s ASIMO which is a 2005 era product. What makes you think the TeslaBot won’t hit the same wall that Honda’s bot did?
It's not 2005 anymore and LLMs have far surpassed Markov Chains from the 80's thanks to advancements in computing technology, so it doesn't seem unreasonable to think there's a chance they'll get past ASIMO level of performance.
TeslaBot will be using the same AI chip and system that Tesla's Autopilot uses to understand the vector space of the world around it.
They've already demoed teaching a robot to repeat-mimic the movements that a human dose with its arms-hands-fingers, and so training will be as simple as that. You could 1 single robot who could clean your whole house as well, say during a 10-hour period while you're out, getting every nook and cranny - and where speed isn't really an issue in that situation; other tasks like a delivery comes in, and the bot can immediately take that delivery and put the items in their proper places - whether that's fridge, freezer, cupboard, or left out on counter for cooking a meal later.
What'd it be worth to have 1 extra human worth of productivity at your command for 24/7? How much more value it creates compared to cost will depend on your circumstances of course. Having a small apartment it may not be that useful, but maybe if you have multiple properties and at once property you've cut down a tree for firewood - the bot can move the logs into a splitter and then stack them for you, etc.
Once you realize that a single robot that can switch between very different tasks in a chain of tasks is possible, the size of production required to reach the same efficiencies of economies of scale will be far lower than currently required for mass production ,e.g. where maybe the equipment needed beforehand required needing to be able to produce say at least 100,000 of something per month but now maybe only 10,000 of those units need to be created for the same margins; which makes the environment more competitive, the barriers to entry lower, derisking and distributing power-profits by allowing more decentralization, a reduction in how much large industrial complexes can capture production and sales.
The battle will be on 2-3 fronts: 1) making sure industrial complexes and bad actors don't try to try to prevent the general population from having access to these technologies, and 2) making sure bad actors don't try to capture and corrupt-takeover these systems to then weaponize them against society, and a possible 3) preventing the companies that produce these technologies, perhaps including AI, from trying to extra value from what "their" bots are capable of doing. E.g. they start trying to take a % off of every business type, depending on the added value the bots create for them, so instead of those values and gains being distributed to all of society - they try to capture and hoard as much of that value creation for themselves, reminding me of the rent-seeking behaviour of what I call the Landlord-Rental industrial complex; "leasing" the technology rather than selling it.
How about addressing and countering any of my specific points, ideally each one in sequence so you can't cherrypick to avoid the ones where I'm certainly correct - rather than what's essentially ad hominem as your response?
How many variations out driving across the US and the world do you think there are compared to a relatively static environment that a Bot can be trained in-tailored for?
TSMC doesn't seem to use their market power to push price. Their take of the final consumer sale price is quite low... they're selling chips to NVDA for ~2k which NVDA turns around and sells for ~60k.
I haven't researched it in-depth, but my take is that they don't have the culture to really be aggressive on pricing.
Doesn't that assume the market is perfectly able to assess value?
As an example the market decided Netflix was a terrible value when they announced they were a video streaming company in like 2008ish losing around half their value. Except we all know now that was not a priced in moment.
The market is clearly not efficient, otherwise there would be no opportunity for alpha.
You can go onto the market today and buy like-kind stocks at double digit discounts to valuation just because they're smaller cap/lesser known. e.g. Retail REITs, where the business and risks are almost identical.
EMH as commonly interpreted is clearly wrong in my view.
What is true is that the majority of people can't beat the market picking stocks, but that doesn't mean there aren't observable inefficiencies. Just that most people don't care or know how to observe them
You could have said the same thing about netscape, yahoo, enron, &c.
"it's too big to fail", "xyz is now part of life and will never ever disappear nor change in any way shape or form",... People really lack imagination, they take the last 2 years and extrapole it to the next decades as if there were no variables whatsoever.
AI still has to bring a profit to anyone other than the company providing them. And even if their wet dreams somehow manifest (aka replacing all human labor) you're opening another Pandora's box and all bets are off when it comes to stock valuation, global finance, & c.
True but various companies will start produce more hardware in house in the future too. For now Nvidia is the sole king in that but other companies will catch up sooner or later.
It's still the sole king for cloud consumers, but for internal use, which is bigger than their cloud, Google mostly uses its own stuff, and with the amount the other big companies have to give Nvidia it is worth it for them to do the same.
There doesn't seem to be a moat we'll see how high it can go but in the long run they're a ridiculously overvalued commodity (eventually) producer.
If you were willing to risk $2,100 on a "buy-to-open" long 2-days-to-expiration (2023-02-23 expirations bought 2023-02-01 right before close) 3.5% out of the money call ($700 strike), you would've had a 300% return today.
How many people (hundreds? thousands? millions?) tried to guess the right combination of earnings up/down (long/short, call/put) risking money they might not necessarily should be, high premium / theta decay / implied volatility crush be damned?
I'm not saying the stock market is "as much like a casino as it has ever been" but... like... at 50-50 coin flip... it is better odds than a casino? lol
IDK but I bought some NVDA throughout last year in my self directed IRAs and it’s really been single-handedly keeping me above market return. E.g. without it i’d be worse than S&P over the same period, lol.
61 comments
[ 4.3 ms ] story [ 126 ms ] threadSome leapfrogging in energy efficiency?
Just more CUDA monopoly?
On the other hand, it will be interesting to see how much money there is in "intelligence" the more abundant it becomes.
Those are five of the top ten companies in the world by market cap.
Not many more dot-coms on the list of top 100 though:
https://companiesmarketcap.com/
It was a winner-takes-all market with a few big winners and a lot of losers.
There's a lot of product to be made. AI is already transforming a lot of businesses. It's too early to tell who the big winners will be, how the market will be structured, or what will happen.
I suspect, in the metaphor, NVidia is the Cisco or the Sun Microsystems. They power the gold rush. Cisco is #60 on biggest companies, and Sun died 14 years ago. I suspect NVidia is now overvalued, but I suspect it should still be highly-valued.
I suspect things will self-correct if AMD, Intel, or one of the startups gets software in shape. The problem is that as much as NVidia can maintain competitive advantage, competition is likely to cap profit margin. If Intel is 30% less efficient than NVidia in compute-per-dollar, than NVidia's profit margin can be at most 30% more than Intel.
what's the price for apple after introducing the iphone? visionary leader in charge of a company that just delivered a money-faucet technology that - while yes, competition will emerge - they will still inevitably end up owning a large and highly desirable stake of in the market?
cuda is literally 15 years ago, and people are just realizing they have to actually start by catching up to that. The "NVIDIA is an AI company now" was like 2015, "mid-2010s", it was a realization jensen had during alexnet's rise, in the early/mid 2010s. It truly will take a long time to build that kind of ecosystem, Apple's support is great with Metal and it still is fairly well a potemkin village until there's more traction on apple gpu support. And a lot of competitors don't even have that, AMD is only finally just realizing this and Intel is also spending like crazy to build out OneAPI/SyCL.
Best time to plant the tree is 15 years ago, and Jensen has repeatedly planted trees where they needed them to be successful in the long term, and moved the engineering and business sides to make it happen.
Programmable shaders, CUDA, tessellation, GSync, and tensor/ML (and RT at the same time) have all been big big bets, and many people criticized some of them real harshly at the time, or backseat-drove that NVIDIA should have given it away, etc.
https://youtu.be/Xn1EsFe7snQ?t=1086
https://youtu.be/tu7pxJXBBn8?t=273
Also, just in general there's really only a handful of companies that can build something like H100/H200 or MI300X or Ponte Vecchio and get an actual competitive product out of it. Let alone the rest of the integration in the system (multi-node, networking, etc) which is all increasingly necessary to build HPC systems at scale etc. NVIDIA paid for mellanox, and amd paid for xilinx, etc. "just" integrate all that successfully into something that clients can start building products on, when can you deliver a from-scratch competitor? lol
(shoutout to oxide though, that actually does look pretty sick, bryan cantrill is in charge so you know it's gonna be good)
There's definitely an enormous amount of key-man risk, but it's also certainly the best cross-disciplinary team (hardware, software, business) in the business today. Nobody else can identify what's gonna be big, lay the groundwork, execute successfully, and roll with the punches (and manage inventory) 90s-tech-business style.
What if you could invest in cray, but it was a business where cray had repeatedly executed successfully and managed to pioneer several large commercial applications for their systems, and expected to continue pivoting and innovating into new areas going forward? There is nobody else in the business like Jensen right now, he is a savant.
Apple has been making a bank on iPhones, with no major leap frog in 10 years. Nvidia could do that too, hard but not unimaginable.
Just like teslas became fully autonomous in 2014 and replaced all human drivers as saint Elon Musk prophecised
And they are much better than the competition, too. AFAIK, only Google and Microsoft have anything competitive in terms of full architecture (but internal).
The TeslaBot with even a dumb level of "AI" for learning and mimicking actions the Bot is shown to perform and can mimic based on its non-meat stick appendages - whole systems and funnels from farming to getting a cooked meal on the dinner table could be (eventually) fully automated with little to no human oversight, where the only initial inputs are raw-natural resources; the problem then that 100% of humanity needs to focus on is then heavy oversight and real-time in-person witnessing that such systems aren't being corrupted-captured by bad actors.
There are other synergies with companies Elon is involved with that people seem to fail to see the synergies of, including but not limited to his effort towards colonizing Mars - of which there will be plenty of ongoing
Elon also controlling Starlink, assuming every Tesla vehicle will have capacity in future for their vehicles to be connected, they'll be able to undercut competitors - and will also act as downward pressure on competition to not overcharge; I believe Elon stated his goal is $5 billion/month from Starlink subscriptions to pay for his Mars efforts, with the first organization to get space mining operations into play - able to extract and send home refined product - has then tapped into the unlimited abundance of the universe, the profits of which reinvested fuels progress along an exponential path; once these systems are in play and self-sustainable, reaching that point by being a better offering, value-price wise, then that frees up time to focus on launching the next puzzle piece - each puzzle piece compounding with the others; his "free" time, fortunately for us, most recently leading to him somewhat impulsively - but clearly his impulses are existential crisis-importance driven - he bought Twitter to protect free speech and unveiled-unmasked the corruption and treason via releasing the Twitter Files.
Elon's arguably 20 years ahead of everyone else in regards to the holistic possibilities that his ecosystem of companies will allow for, and arguably every year there is the possibility he leapfrogs another 10 years ahead.
How can this happen? This sounds like some serious hype.
What I find really funny about the TeslaBot is that it looks as capable (if not less) than Honda’s ASIMO which is a 2005 era product. What makes you think the TeslaBot won’t hit the same wall that Honda’s bot did?
They've already demoed teaching a robot to repeat-mimic the movements that a human dose with its arms-hands-fingers, and so training will be as simple as that. You could 1 single robot who could clean your whole house as well, say during a 10-hour period while you're out, getting every nook and cranny - and where speed isn't really an issue in that situation; other tasks like a delivery comes in, and the bot can immediately take that delivery and put the items in their proper places - whether that's fridge, freezer, cupboard, or left out on counter for cooking a meal later.
What'd it be worth to have 1 extra human worth of productivity at your command for 24/7? How much more value it creates compared to cost will depend on your circumstances of course. Having a small apartment it may not be that useful, but maybe if you have multiple properties and at once property you've cut down a tree for firewood - the bot can move the logs into a splitter and then stack them for you, etc.
Once you realize that a single robot that can switch between very different tasks in a chain of tasks is possible, the size of production required to reach the same efficiencies of economies of scale will be far lower than currently required for mass production ,e.g. where maybe the equipment needed beforehand required needing to be able to produce say at least 100,000 of something per month but now maybe only 10,000 of those units need to be created for the same margins; which makes the environment more competitive, the barriers to entry lower, derisking and distributing power-profits by allowing more decentralization, a reduction in how much large industrial complexes can capture production and sales.
The battle will be on 2-3 fronts: 1) making sure industrial complexes and bad actors don't try to try to prevent the general population from having access to these technologies, and 2) making sure bad actors don't try to capture and corrupt-takeover these systems to then weaponize them against society, and a possible 3) preventing the companies that produce these technologies, perhaps including AI, from trying to extra value from what "their" bots are capable of doing. E.g. they start trying to take a % off of every business type, depending on the added value the bots create for them, so instead of those values and gains being distributed to all of society - they try to capture and hoard as much of that value creation for themselves, reminding me of the rent-seeking behaviour of what I call the Landlord-Rental industrial complex; "leasing" the technology rather than selling it.
https://www.youtube.com/watch?v=XiQkeWOFwmk https://www.youtube.com/watch?v=D2vj0WcvH5c etc.
Perhaps you're not thinking from first principles and extrapolating from there?
If you’re talking about the AI chip in my Model S Plaid, I don’t have high hopes for this robot.
https://finance.yahoo.com/quote/NVDA/financials
Vehicles are one thing, AI chips are another.
One is a default investment for any AI tech company to survive.
Data centers will only grow from here.
I haven't researched it in-depth, but my take is that they don't have the culture to really be aggressive on pricing.
As an example the market decided Netflix was a terrible value when they announced they were a video streaming company in like 2008ish losing around half their value. Except we all know now that was not a priced in moment.
You can go onto the market today and buy like-kind stocks at double digit discounts to valuation just because they're smaller cap/lesser known. e.g. Retail REITs, where the business and risks are almost identical.
EMH as commonly interpreted is clearly wrong in my view.
What is true is that the majority of people can't beat the market picking stocks, but that doesn't mean there aren't observable inefficiencies. Just that most people don't care or know how to observe them
"it's too big to fail", "xyz is now part of life and will never ever disappear nor change in any way shape or form",... People really lack imagination, they take the last 2 years and extrapole it to the next decades as if there were no variables whatsoever.
AI still has to bring a profit to anyone other than the company providing them. And even if their wet dreams somehow manifest (aka replacing all human labor) you're opening another Pandora's box and all bets are off when it comes to stock valuation, global finance, & c.
There doesn't seem to be a moat we'll see how high it can go but in the long run they're a ridiculously overvalued commodity (eventually) producer.
If you were willing to risk $2,100 on a "buy-to-open" long 2-days-to-expiration (2023-02-23 expirations bought 2023-02-01 right before close) 3.5% out of the money call ($700 strike), you would've had a 300% return today.
How many people (hundreds? thousands? millions?) tried to guess the right combination of earnings up/down (long/short, call/put) risking money they might not necessarily should be, high premium / theta decay / implied volatility crush be damned?
I'm not saying the stock market is "as much like a casino as it has ever been" but... like... at 50-50 coin flip... it is better odds than a casino? lol
I guess roulette red/black.