저자: CathieWood · 외국인
요약 보기
AI는 향후 5~10년 최대 부의 창출 촉매이며 파괴적 혁신 시가총액이 2030년까지 19조→220조달러로 연 42% 성장할 것으로 전망. AI 소프트웨어 지출이 1조달러 이상이고 AI 칩 수요도 그에 맞춰 커짐. 과거 암호화폐 사이클처럼 엔비디아 등 하드웨어 주문 과열과 기대 선반영 우려가 있어, 포트폴리오를 하드웨어 중심에서 생산성 이득을 누리는 소프트웨어 기업으로 이동 중. AI 학습 비용은 연 4배씩 하락 중.
다듬은 글
is booming so a lot of people are paying uh a lot to try and figure out the risks Associated but the opportunities Associated uh with AI are transformational this this is going to be the biggest um Catalyst behind wealth generation uh during the next 5 to 10 years and the and the super exponential uh growth opportunities that it will enable most people who waited until the S&P gave them permission to move into Tesla are underwater right now because there as as is always the case with disruptive innovation there's controversy impact of technology on real GDP growth when there are big breakthroughs and so as Brett mentioned the the breakthroughs in the early 1900s telephone electricity internal combustion engine um created a step function increase in growth and you can see it here there's a lot of compelling evidence that this decade is going to be remarkable from a macroeconomic perspective right and it should translate into um very nice returns for these Technologies as you can see uh we if you look at the market today um roughly 19 trillion of the 117 trillion in global Equity market cap is associated with disruptive innovation so let's let's say a little more than 15% um if we're right and and with Brett's guidance our analysts have uh you know built the building blocks uh for us to be able to say this uh if we're right that 19 trillion is going to scale to 2 20 trillion uh during the next really seven eight years seven years so that's a 42% compound annual rate um of return um and whereas the rest of the world we see appreciating little or nothing at all why would that be it's back to that creative destruction that the the traditional World Order is going to change radically um whereas AI 37 % uh that is off of a substantially larger base and just to give you a sense of the drama here even for us and all we do is focus on disruptive innovation um in I believe this was 2020 our expectation for the market cap of artificial intelligence out there um in five years from now was I think $40 billion now it's up to $400 billion do perhaps but and part of that is like we we opened up the model by saying well how much value is going to be delivered with AI software and um you know if I can just like if I can deliver a robot to an an Amazon or to a Tesla and that can increase the productivity of their manufacturing Base by X they'll spend some fraction of x to install that robot and then they'll get great Roi the same with software and how it improves me and all of our analysts as knowledge workers people pay for software that improves worker productivity uh and if you look across all knowledge workers there's almost $30 trillion in knowledge work wages that is going to be paid in 2030 so if you can improve their productivity by multiples and just pay a fraction of that you still end up with trillions of dollars and in fact more than1 trillion dollar in our estimation of AI software spend to help knowledge workers and there you need a trillion dollars plus of AI chips to create that AI software uh and and so um kind of rather than approaching it from like hey this is uh this is it spend and what people are going to do we approach it from this is the output of the technology that's going to be installed and therefore the amount of kind of um justifiable really Roi positive Hardware we need to support that uh and you end up with a much larger market begs a question you know we have been shifting our emphasis from um in our various strategies from uh Hardware Centric that would be Nvidia uh AMD depending on the portfolio um it towards software and towards companies that are going to capitalize on the productivity gains that um that uh we're going to that AI is going to deliver um why would we be pulling away well it's a little bit back to uh the question we're talking about and the confusion that we expect to uh see in in in the next year around the cycle um because believe it or not even though AI is a force it really is we are seeing what we saw during uh covid there's double and triple ordering uh in a shortage uh situation or or perceived shortage POS situation uh in covid we're we're still suffering from uh the aftermath of that uh and you know even though this is a huge SEC secular growth story there are there's so much in the way of expectations now being built into the check the box Ai and believe me it's an important box Nvidia has done a magnificent job we bought it from uh you know from from from the time we started the company we're involved with it when it was $5 and just when we understood we built it enormously uh but the but it is subject to cycles and and and the last one was the crypto cycle 2017 couldn't get enough gpus uh and uh and there was double triple ordering and the stock in the uh fourth quarter of 2018 was cut by 2/3 now that's not going to happen this time because this AI wave is just so powerful but the expectations do get ahead of themselves we saw that with you know uh uh alphabet I guess uh earlier this week and AMD just to expectations had gotten ahead and uh you're going to have the the the two and fro we talk about cost declines a lot at ARC and and on AI in particular it's the cost decline is happening more quickly than anything we've ever studied so Kathy alluded to it but um cost to train a model we think is going to fall by four times now last year we thought by three times per year prospectively going forward through 2030 and on a real-time basis costs are falling even faster than that for AI training which is separate from the cost to actually deliver an AI model which is called in and that's following at you know 85 plus% per year but then that's also separate from what it means for the cost to do something within the business or human context at an in application basis I can't remember what's on the chart 12 cents 16 cents 16 16 cents per thousand words so you've gone from $4500 to 16 cents well then anthropics Claud 2 model uh scored 90th percenti on the Gres and it cost 4 cents so we went from something it was the same amount to produce written content for more than a century suddenly it falls 10,000 fold and is now going to be on a cost decline trajectory it's not like it's just going to Plane off essentially written authored material is going to both be ubiquitous and abundant with all that is good and bad about that as in you will have your own personal chatbot that you can chat to that will be you know really compelling and interesting and have all kinds of information to share with you uh and you will be at risk of somebody you know kind of like infiltrating you pretending to be human when they're really not and so I think that the the implications for like businesses um humans like all of these systems that we use where like kind of written and oral speaking ability is a mechanism by which we verify that this person is a person that I should pay attention to uh is just going to get totally um turned upside down by this and it's not not just going to happen in the written word context it's going to be image generation it's going to be kind of ability to look at data sets and forecast the direction that they're going so kind of like the the the ability to take like an idea and convert it into something that amplifies and spreads is going to be massively magnified here with all kinds of interesting implications um by Ai and so like the one of the promises of technology is to reduce the friction between interact between agents corporations everything like you can imagine there's like the transaction friction is going to collapse in all kinds of different ways both with AI and blockchain I think that should like have a profound macroeconomic effect
Comments 0