저자: Kevin · 외국인
요약 보기
미국의 대중 반도체 수출 규제(칩스법)로 엔비디아·AMD의 중국 매출(약 20~25%)이 타격받고 있으며, 바이든 행정부가 규제를 더 강화하면서 반도체주 시장 반응이 부정적이다. 스노우플레이크-엔비디아 파트너십은 기존 빅데이터 사업의 AI 리브랜딩에 가깝다고 평가. AI 혁신은 초기 도약 후 1차 미분이 평탄해지고 있으며, 서버칩 성장률이 연 50%로 외삽되는 시장 가격 반영은 과도할 수 있다고 우려. 반도체는 장기적으로 중요하지만 현재 매출 급증은 일과성일 수 있다고 주장.
다듬은 글
uh basically Nvidia sells these h100 chips and they also sell these 8800s and uh basically they sell these watered-down versions of some of the high powered technology chips some of the best products they sell watered down versions of that to China because the chips act has banned the full powered high-powered versions from actually being sold to China Nvidia and AMD have had to deal with this for over a year now and uh companies like Nvidia and AMD get somewhere around 20 to 25 percent of their revenue from China problem is now they can't sell their best chips to China so China is incentivized to manufacture their own best chips by stealing as much proprietary data as they can from companies like Nvidia or even asml where they publicly have stolen Advanced lithography euv equipment data sets from and instruction sets from for how to manufacture these actual machines that make the chip so China is realizing all right well if we're only going to get the cheap stuff then let's just use the cheap stuff until we can make our own great stuff it's literally the same thing even at asml itself asml is only able to sell their duv lithography equipment to China but not their euv equipment which is for the most advanced stuff so China obviously doesn't like that and again now they're trying to create their own companies to manufacture this equipment and then in the meantime while they use this watered-down equipment the Biden Administration has realized you know what we may not even want China to be able to buy the water down equipment let's just ban everything all nearly everything related to Advanced chips so basically what they're doing is they're going from just Banning the most advanced chips to closing some loopholes to Banning even more chips media and AMD aren't very happy about that stock wise at least in Market reaction which is understandable because this should be a hit to revenue initially we thought okay Nvidia and AMD got this solved they basically made a chip with some safeguards on it that uh now they can still sell to China well now Biden and the Biden Administration is suggesting nah still too much for China obviously uh that doesn't help with a U.S Chinese relations which actually seemed like they were thawing as of the Biden administration's meeting with with Chinese officials and just actually the last few weeks it felt like there was almost a reset of relations coming between the United States and China yet here we are with more aggression coming uh and uh you know I I'm not here to suggest everything should be perfect we got to roll out the red carpet for China obviously there are a lot of criticism staff for China but it just seems like the more of these tensions that we have the more complicated the relationship gets between the United States and China and then we start talking about war with Taiwan again which there's a lot of talk about Taiwan potentially creating election risks uh or post-election risks for China invading Taiwan the argument has been made that well hey look if China ends up losing uh or their party the party their backing ends up losing in Taiwan but buckle up but really war with Taiwan is a topic for a different video but it just gives you a little bit of a sense of how much of the tension there really is between the United States and China and just as we thought it was cooling it seems like the Biden Administration has figured out how to reactivate some of that pain now uh so we've got that in the world of AI then we've got this snow Nvidia partnership people keep asking me about so let's touch on that quickly and then after we touch on that I really want to give you some of my personal thoughts about is this whole AI craze just hype and how should we really be positioned around it so we'll talk about exactly that okay so regarding the snow in video partnership uh this has been done before this isn't like something super crazy and unique these Partnerships have literally been done before snowflake had virtually the same partnership and I think a rough easy way to think about it is when you look at your calendar and you see an expiration on June 30th you go to the link down below like meet kevin.com and you sign up for the programs and building your wealth because you know you're going to lock in the best price guaranteed and you're going to continue to get amazing content for free as the courses grow in value link down below right anyway so basically what was announced is there's this official partnership that now uh well let me backstep for a moment before AI was AI Big Data was like this sexy phrase that everybody talked about back in the day and anybody wanted to understand how are we going to analyze big data and it was companies like Snowflake and palantir let's suggested hey put all of your big data into our bucket and our bucket manager will draw insights for you insights that you might not have even realized exist it's kind of like what palantir does with their criminal database and their terrorism database and their War databases where it's like give us everything that you know and we'll Stitch pieces of the puzzle together that you didn't even realize were happening this is obviously leveraging neural Nets which is a form of artificial intelligence it's just all basically been rebranded as AI because that's a sexy thing to say these days point is Nvidia has a Nemo platform that allows the development of language models and a lot of these database companies are like wait a minute we don't want people taking their data off of our platforms and running away to these you know newfangled AI things so how can we figure out how to just Rebrand what we have as having Ai and then maybe adding in some you know updated features the problem is these features don't really seem that fantastic so I'll give you their marketing pitch and then I'll give you some of their example use cases and then I'll let you decide so their marketing pitch is that quote together Nvidia and snowflake will create an AI Factory that helps Enterprises turn their own valuable data into custom generative AI models okay so the whole point of Snow's existence has always been that you could take a lot of data and do more with it now it's just being rebranded as an AI play okay so what is this AI Factory what are some of the use cases well here's the type of use case Health Care insurance models could answer complex questions about procedures that are covered under various plans okay first of all we have language Bots Bard gpt4 gpt4 is pretty exceptionally great and quite frankly what it seems like everybody is doing is just taking gpt4 and plugging it into every stupid thing that exists in the world and it just does the same thing that gpt4 does everywhere call me a little bit cynical but the reality is use the chatbot doesn't matter if I have to go to the tab for open AI or here now the idea is that oh well you could train the language model on your own proprietary data fine I'm just not convinced how profitable that kind of service will be more so than what the company has already been doing do I really think this partnership is a game changer where all of a sudden I'm going to look at the income statement for Snowflake and go you've plugged the in essence I'm oversimplifying right gpt4 into snowflake I will Mark your revenues up at 10 a year beyond what I previously thought no and that's one of the frustrations that I have with AI but I'll get to more of that in a moment here's another one a financial services model could share details about specific lending opportunities available to retail and business customers based on specific circumstances okay we don't need AI to do that I can hand you a pamphlet that's like yo if your credit score is 640 here's what we got for you a crap credit card and if your credit score is 740 we got a HELOC for you and a personal loan for you and you know uh you know a happy ending I don't know whatever point is it's not it's not the the examples they're giving aren't that impressive uh deploying custom llms used for generative AI applications such as chat Bots and intelligence search we get it like that's that's basically what everybody is plugging in right now everybody's figuring out how to harness the power of this gpt4 Innovation plug it in so where does my cynicism come in and what do I think is a danger for this AI craze now keep in mind I I feel like honestly I think sometimes it's the most frustrating part of of what I do is last last fall uh I started really getting excited about chips because chips were crap you know nobody wanted to touch chips it was like October uh everybody's like dude PC demand is in the toilet chips are dead nobody's gonna want chips anymore and I'm like I like that chips oh and the many fundamental reasons for this before even AI came around like these suckers are oversold the orange recession is over for those particular plays here's an opportunity anyway so I went really heavy in chips but now the story played out to with the benefit of AI much faster than anybody thought it would but now as it's playing out I'm catching myself being like no no no I have my concerns about this as excited as I am about artificial intelligence and some of the Practical use cases you know I I think right now you're probably looking at maybe a 15 20 efficiency boost if you harness AI correctly uh whether that is learning how to speak better in presentations and videos and being more concise and and tone checking your your letters or dispute resolution I mean there are so many different things that we could do with it it's phenomenal I mean I have a course but this is really you know a income it features AI because we've got about 40 AI lectures and they're really really good and these are practical lectures on how to actually use this stuff rather than just the you know the basic kind of stuff you see on Twitter or whatever but anyway the frustration that I have is beyond this initial boost the Innovation that we're seeing has gone to like zero in my opinion I know that sounds harsh and rough and I'm not here to be like an AI Luddite and say oh ai's you know crap AI is great and it's going to do great things over time but let me do this I'm going to graph and this is why I'm I feel frustrated uh and I'm trying to get to that point so let me graph you this if I'm going to graph the first derivative of the AI Innovation craze here's what you got no change no change no change GPT and Bard and all that craze and no chain oh no change no change no change okay this is what it feels like so it feels like we had this big explosion of innovation and really like if we if we don't go to the first derivative the graph of AI usage probably looks like this skyrocketing and they kind of like staying here right so like we've definitely leveled up but when you look at the first derivative you you you're kind of like yay we had an innovation moment but now you're kind of flat again or it feels like the re-innovation isn't that great again don't get me wrong there's certainly improvements happening so I think maybe the way to put this is maybe maybe there's a you know a a slow and steady Improvement right but even that is is it doesn't seem to be super rapid right now what kind of risk does this create for chips well here's the risk it creates the risk it creates is growth that's being priced into these chip companies and this is a realistic concern it's not just trying to bandwagon off oh it's a red day for chips something that I've been really trying to I don't want to say cope with but like deal with it's like I have this problem because I worry about people doing this so here's what's happening first people are like okay nvidia's quarterly revenue for chips is now estimated to be you know server chips seven billion dollars so oh I can't wait to get over this so what people are doing is they're annualizing that and they're saying all right times four so in video make 28 billion dollars on server trips and then what they're doing is they're gonna say all right well it's going to grow at well if it used to if it grew at 100 in a quarter let's just say it'll grow at 50 a year okay so now what people are doing is they're saying all right well let's go 24 25.
26 27 and they're extrapolating out all right we're gonna go uh and grow this over and over again uh by 1.5 uh let's do this a little bit better uh whatever we just quickly put this together here so they're growing this Revenue so rapidly that uh that basically you've got this market pricing and Perfection for AI growth of of 50 of extra you know spending on server chips but my concern is ignoring China if your AI Innovation is maybe a little better than this like you had a big boost and then and then maybe you've got you know some kind of like uh I guess if this was a first derivative chart we'd probably be a little bit more like that you know where you've leveled up a little bit but if we go over here that's not consistent with 50 growth over and over and over again it's really consistent with a lot of growth really quickly in the next quarter or two and then a Slowdown in that growth rate pretty rapidly once people have these servers in fact I've gotten to the site at this this point of thinking to myself if everybody gets their age 100 how much processing power do you really need for these chat Bots and then you upgrade to the h100s and then what then you're flat again and then you get back to the slow Innovation curve of AI until the next big explosion in Ai and all of a sudden now instead of projecting Nvidia having 141 billion dollars of server chip Revenue in in 2027.
maybe the more realistic thing is okay they had 28 Bill even annualizing that could be generous but maybe the growth was only 25 percent and look how remarkably the numbers change after a few years oops let's actually do it correctly here if your growth rate is half and and the difference is is massive so look at that you're less than half by 2027 if your growth rate is is that much slower in this case it's 25 uh you're less than half because of the comp nature of compound and growth uh and that just gets even more extended so when people do these DCF models it's a discounted cash flow and they're running these numbers into 2033 you know trying to do these 10-year DCFS on AI they're assuming that the AI Innovation is going to keep keep growing as rapidly as GPT did here but it would what if it was just a flash in the pan so I personally am frustrated and this kind of goes back to why I said it was frustrating because I love the chip so much at the end of last year and I think chips are going to be so important for the next decade but part of me has the struggle of like what if this these sales are just a flash in the pan like we had a big boost in sales and then we level back off slightly above where we were before like slightly above trend in fact that's another way to analyze that is like think about it like stimulus money for covid right so so uh let's say that the trend of personal uh you know disposable income the trend of personal disposable income looks like that let's call that the trend okay and then you get kova that comes around and all of a sudden you get this this whole and this massive Spike uh of of money that people get and then you're back to this sort of negative hole and then you're kind of you know bouncing back to turn this is a return to trend and maybe maybe that trend is a little bit higher right so maybe we say all right now we're kind of like this and that purple trend line actually needs to kind of start bending up right you do something like this and so this extra bit right here is just the net benefit from coven that's fine but is the trend line here now no of course not and so that is a concern I have for AI and I'm concerned that analysts are pricing in this sort of goldilock-ish this here rather than what might just be a nominal boost a more nominal boost so that's my take uh let me see what other notes I had here on the uh partnership between Snowflake and Nvidia AI Factory was really what they were talking about you know I I think they uh they're trying to lean into this idea of like don't know what to invest in in AI well snowflake is now the shovel seller no you're not you're not the shovel seller Nvidia and and tsmc and uh and asml they're the shovel sellers but uh oh yeah and then this like supports text to image text image to text like all this stuff with their great Innovations already for it and now what companies are doing is they're just plugging it in like yesterday who was it it was um I think it was Ross he's like have you ever seen a lion eat it was responding to somebody's hypothetical joke uh Omar here on Twitter was making a joke he's like all animals are vegans why can't we be too obviously we know all animals aren't vegan so he's joking but Ross replies ever seen a lion eat like Ross he's joking but anyway so I replied I go Omar's story checks out and here's a little AI generated image of a lion and his you know fruit that he's eating obviously this isn't real but it looks pretty dang awesome and it took me all of like 10 seconds to make okay great is that like did that I mean it increases my productivity somewhat but you know to some extent uh in this in this example it's also just a joke uh look for some purposes it's it's fantastic again I think AI can give you that 15 productivity boost for sure maybe even more if you uh implement it correctly but uh but as far as this uh this all in on AI uh software makes me the most nervous but I've started to get nervous also about chips so uh do you mean AI is transitory no I don't want to say it's transitory like it's going to be here to stay right this is not to say it's like okay chat GPT is here and now it's not no no that would be like transitory I would say the rapid boost of chips sales might be somewhat transitory potentially are some Industries more ripe for Innovation from AI well you mentioned Transportation Hospitality entertainment I actually think it would be more like health care and to some extent Finance those probably more in the future I think yes with Transportation but that's kind of like your FSD style artificial intelligence I actually think that's probably your most intelligent artificial intelligence to some extent that exists today is full self-driving the Tesla fsda I think it's really really incredible uh Hospitality entertainment I think these might be a little bit harder uh these are this is so very labor involved but uh look I remember working at Jamba Juice back uh how long ago was that I don't know 15 years ago and uh one of the managers is like oh we have this software that's like schedule three people for today and the manager was looking at it going like I think it's going to be a slow day why does it say schedule three people like I think it's gonna be slow and the software had all the big data of the prior 10 years of Jamba Juice sales and and when people come in you know when the humidity is this level and the weather is this level and this is that and a lot of the holiday schedule it puts all that big data together and the system's like no no listen to the AI schedule three people so the manager like all right whatever I'll schedule three people and you know the first hour of the shift the boss is like I knew it stupid computer wasting our money again and then all of a sudden you get like a three hour long Rush and then afterwards the manager's like damn how'd the AI do that they didn't call it AI then they kind of the computer program so it's all just like fittery branded you know uh uh anyhow those are my thoughts on on this this AI stuff uh okay goo]
Comments 0