저자: 미상 · 외국인
요약 보기
생성형 AI 승자와 무관하게 수혜를 받는 반도체 밸류체인(마이크론, 어플라이드머티리얼즈, TSMC, ASML)에 투자하라는 주장. AI 시장은 2032년까지 연 36% 성장 전망, 엔비디아 데이터센터 GPU 점유율 92%, 마이크론 HBM 점유율 약 25%, 고대역폭메모리 시장 연 68% 성장 전망 제시. 반도체 공급 제약이 투자자 인식보다 심하다고 주장. 영상 내 유료 광고(딜리트미) 포함.
다듬은 글
open ai's gp40 is taking the World by storm and generative AI is already changing our lives in a big way just like smartphones did and the internet before that and personal computers before that but the real opportunity for investors isn't about open AI at all it's about Google actually no not exactly it's about finding the companies that will win big regardless of which AI ends up on top that's the best way in this video Micron Technology which makes the high bandwidth memory modules for nvidia's gpus Applied Materials and their Innovations in semiconductor equipment tsmc which makes over 90% of the world's most advanced chips and asml which might just have the biggest Monopoly in the entire AI market now your probably wondering how this list of stocks has anything to do with GPT 40 or Google Gemini so let me explain that first starting with the AI Market itself the global AI Market is expected to almost 12x in size over the next 8 years which is a compound annual growth rate of over 36% between now and 2032 by the way that's about a three times higher growth rate than the S&P 500's average annual rate of return of 12% over the last 10 years software and services account for about 75% of the total AI Market opportunity while Hardware makes up the other 25% and since software usually has higher margins everyone is focused on finding the winner between software companies like open Ai and Google but while their presentations couldn't have been more different I just want to thank the incredible open aai Team Google they actually did have one big thing in common we are fortunate to have a long-standing partnership with Nvidia we are also proud to be one of the first Cloud providers to offer Nvidia Cutting Edge Blackwell gpus thanks to Jensen and the Nvidia team for bringing us the most advanced gpus to make this demo possible today now this isn't a video about Nvidia but Nvidia is dominating the market for data center gpus with a whopping 92% market share which means almost every generative AI application and service runs on nvidia's chips today and the market for data center accelerators is expected to more than 5x by 2032 which means demand for these chips is growing at a 24% compound annual growth rate combine that with Nvidia having gross margins north of 75% which is as high as most software companies and it's not hard to see why Nvidia stock tripled over the last year alone Nvidia is winning big regardless of which AI comes out on top and what all of the companies in this video have in common is being involved in making nvidia's chips that's my plan for getting rich without getting lucky let's start with Micron Technology ticker symbol mu Micron is an American company that makes memory for the entire Computing industry not just Nvidia Micron makes solid state drives Ram modules and flash memory for data centers consumer PCS industrial iot devices Flagship smartphones and even AIO automotive and each of these markets is currently transitioning to more advanced memory for AI applications from generative AI services and data centers to co-pilots on PCS smartphones and wearables all the way to self-driving cars as a result the global market for nextg data storage solutions like nvme solid state drives and mam is expected to more than 6X over the next decade which is a compound annual growth rate of over 22% for the next 10 years but that's nothing compared to the market for high bandwidth memory which could grow by almost 50x in the next 7 years that's an insane 68% compound annual growth rate for the rest of this decade now to be fair some of that crazy growth is because High bandwith memory makes up just 5% of all memory deployed today and it's got a lot of room to grow thanks to the insane demand for generative AI infrastructure which again is dominated by Nvidia today well good thing for for micron that they make the high bandwidth memory for nvidia's gpus high bandwidth memory works by stacking layers of individual memory chips and connecting them to a single Hub which helps pack more memory into a smaller space the goal is to increase data transfer rates and decrease overall power consumption by lowering the physical distance that the data needs to travel that's why nvidia's h200 data center gpus use micron's high bandwidth memory cubes which use around 30% less power than the competition and just like Nvidia micron's most advanced hbm Solutions have much more demand than Supply they're already sold out for 2024 and most of their 2025 inventory has already been bought up as well Micron currently has around a 25% share of the Next Generation memory Market putting them right behind Samsung and SK HX both of which are headquartered in South Korea Micron stock is up by around 50% year to date but we're still in the very early Innings of the AI war between companies like open aai Microsoft Google and meta platforms and as all of them keep building bigger and faster AI models they'll need tons of high bandwidth memory and better storage solutions to handle all that data and speaking of data I recently found out that dozens of online data Brokers were selling my personal data and if you've been getting more spam phone calls texts or emails lately they might be selling yours too that's why I decided to join delete me the sponsor of this video delete me is a hands-free subscription service that will remove your personal information from hundreds of online data Brokers you just sign up enter your information and let their experts get to work they reviewed over 32,000 listings for me since I signed up after 7 days you'll get your first privacy report that shows you everything they removed so far I just got my fourth one and delete me removed over 300 pieces of my personal information since I joined not just my name and address but my fiance's and my mom's too and I love that after they remove your information they'll rescan these websites regularly you even get your own privacy adviser if you want to talk to a real person and make a custom removal request if you care about your data and your family's privacy you can get 20% off any consumer plan with my code symbol2 by going to join delet me.com symbol2 which I'll also link to below and a big thank you to delete me for supporting the channel and for keeping my family's data safe all right so Micron makes the high band memory for nvidia's Blackwell gpus but that's just one piece of the puzzle Applied Materials ticker symbol Amat is a Silicon Valley company that makes specialized tools and equipment for chip fabrication and according to Jensen Wong nvidia's gpus would not exist without Applied Materials Amat has revolutionized the semiconductor industry by inventing fundamental chip processing tools let me highlight three each Incredible One one it was among the first companies to manufacture epital reactors for growing semiconductor layers two amats chemical mechanical polishing tools planarized semiconductor surfaces three its batch edging tools revolutionized circuit patterning all technological Marvels thanks to amt's contributions today we can build three nanometer chips with features 100,000 times finer and processing capabilities millions of times more powerful than when my career began this progress of semiconductors allowed Nvidia to invent the GPU and accelerated Computing which led the world tackle problems that traditional computers can't like computer Graphics scientific Computing drug Discovery artificial intelligence and self-driving cars Amat is pushing the state-of-the-art of Chip Manufacturing in a few key ways first they make machines called epitaxial reactors don't worry I'll keep the industry jargon to a minimum Applied Materials cural line of epir reactors is used to apply different layers of materials onto a semiconductor to improve its electrical properties like making the right areas more conductive to get it ready to become a microprocessor or a memory device Jensen also mentioned that Amat helped Pioneer something called planarized Wafers Amat makes chemical mechanical polishers or cmps which use a special solution to smooth out each layer from the epir reactor polishing these layers helps them fit together without any gaps or electrical faults and a planarized wafer is one where all the layers have been added to the wafer and Polished in preparation for the next processing step the third thing that Jensen mentioned was called batch etching and deposition at a high level etching is where the tiny channels for circuit patterns are carved into each layer of the wafer after etching the empty panels are filled with conductive metal during the deposition process which turns the planarized wafer into an electric circuit Applied Materials makes many more machines for a wide variety of Industries but I wanted to at least show you why Jensen Wong says nvidia's chips wouldn't exist without them the global semiconductor manufacturing Market is expected to more than double in size by 2032 which would be a compound annual growth rate of just under 10% for the next 8 years and I honestly think that estimate is conservative since generative AI is driving so much demand for advanced chips that companies like Nvidia and Micron are currently limited by Supply and these Supply constraints are actually twice as bad as most investors realize remember micron's high bandwidth memory is made by stacking layers of individual memory chips and as a result it takes roughly twice as many dies to make compared to standard Ram the same is also true for nvidia's Blackwell gpus which are made by connecting two dieses and since both Nvidia and Micron are using twice as many dieses semiconductor manufacturers can only make half as many products with the same amount of resources on top of that they need additional materials for steps like stacking dieses for micron's memory or connecting the dies for nvidia's Blackwell gpus all of which means more demand for Applied Materials Solutions as a result Amat expects their revenues from high bandwidth memory packaging to quadruple year-over-year thanks to generative Ai and just like with Micron it doesn't matter whether GPT 40 or Google Gemini comes out on top a matte stock is already up by close to 40% year-to date and the revenue should keep expanding for years to come no luck needed all right so Micron builds the memory for nvidia's gpus and Applied Materials makes the machines but tsmc is where it all comes together the Taiwan semiconductor manufacturing company is a Pure Play Chip Foundry that's supports chip designers around the world and tsmc doesn't make any chips for themselves so they don't compete with their clients as a result over 60% of all chips on Earth are currently made by tsmc even crazier they make 90% of all advanced chips in the world from Tesla's chips for full self-driving to the a series and M series chips in Apple's iPhones and MacBooks and of course nvidia's gpus in fact tsmc has a custom 4 n meter process specifically for NVIDIA called 4np and last time I talked about buying TSM stock I said this I think we'll see many more specialized chip fabrication processes like tsmc's 4N some processes will focus on Graphics chips others will focus on optimizing CPUs and maybe others will be better for flexible server chips or data Focus chips and so on the thing is these specialized processes only really make economic sense in large volumes so as new AI application start requiring new chip designs and architectures that should ACR to tsmc's benefit because they're really one of the only companies that can make such specialized chips at scale that was back in June of 2023 when TSM stock was just under $100 per share and since then tsmc did exactly what I predicted they made big advancements in their process nodes with a strong focus on building chips for AI I'm not trying to toot my own horn here my goal is to show you why it's worth understanding the science behind the stocks well tsmc's 3 nanometer process node N3 is currently in volume production with their two nanometer node on Pace to begin production for apple and nvidia's chips in the second half of 2025 and I wouldn't be surprised if they pressed their advantage and came out with new specialized process nodes to do things like stack more high bandwidth memory together for companies like Micron or just fit more transistors on a chip with higher power efficiencies like they already have for NVIDIA since power consumption is the bottleneck for AI data centers today and here's a fun fact tsmc uses nvidia's AI to make its chip manufacturing processes more efficient Nvidia created a tool called autod DMP which optimizes the placement of different chip components tsmc uses autod DMP to optimize different chip designs up to 30 times faster than previous methods tsmc also uses nvidia's cou with the library to accelerate computational liography by 40 to 60x which was a huge breakthrough because computational lithography is a big bottleneck in modern chip design I'll link my video walking through that entire breakthrough in the description below but long story short tsmc uses nvidia's AI to build better chips which improves nvidia's AI which helps tsmc make even better chips and so on so this is AI building Ai and of course we can't talk about tsmc without also talking talking about asml asml is the only company on Earth that makes extreme ultraviolet lithography machines which use ultraviolet light to instantly etch tiny chip designs onto silicon Wafers each euv machine has over 100,000 parts and needs dozens of trucks and multiple cargo planes to be transported around 90% of the parts for these machines are actually made by hundreds of different companies around the world and asml acts as the system integrator that that way asml doesn't need to be the best in the world at making each individual part and subsystem many of which need insane amounts of engineering to be built to spec for example these euv lithography machines have special mirrors that are made by a company called Zeiss these are the flattest mirrors on Earth they're said to be so flat that if Zeiss made a mirror the size of the United States the biggest bump would be the size of a speck of dust so it would be much riskier if asml did try to make all of the these complex components themselves because if one part has an issue the whole $200 million machine has that issue instead they've built up a massive network of exclusive deals with hundreds of specialized suppliers experts say it could take decades and billions of dollars for any other company to even begin to challenge asml Mo which is why I said they might have the biggest Monopoly in the entire AI Market their next Generation machine is called the high na extreme ultraviolet lithography machine which can etch lines up to 1.7 times smaller than the previous generation which ends up enabling up to 2.9 times more transistors per chip and more transistors per chip means higher processing speeds or more memory or more compute per watt everything that's in high demand thanks to AI Intel recently confirmed that they received and assembled the first High na euv lithography system which will allow them to make 2 nanometer chips and help them make Headway on their Intel 18a and 14a process nodes funnily enough tsmc says that they don't need these machines at all and they expect their next Generation process nodes to enter mass production in 2025 without them either way the increasing demand for AI chips is bound to benefit asml stock which is already up 29% year to date all right I know this video is a little long but I wanted to be thorough since investors might be missing the big picture by focusing only on gp40 and Google Gemini there's no doubt that generative AI is creating huge investing opportunities in a wide variety of Industries but they all rely on AI processors high bandwidth memory and the machines that build them and in my opinion the best way to find the winners in these quickly growing Hardware markets is to understand the science behind the stocks that's the best way research like this Google sure why not
Comments 0