저자: CathieWood · 외국인
요약 보기
AI 학습·추론 비용이 급락(학습 연 75%, 추론 85~90%)하면서 무게중심이 학습칩에서 추론칩으로 이동한다고 주장. 팔란티어가 점유율을 확대하고 IaaS는 안정, SaaS는 하락할 것으로 봄. 자율주행·헬스케어(크리스퍼·시퀀싱) 등 응용이 폭발적으로 성장하며 딥시크 충격은 미국 기술 섹터에 경쟁을 통한 긍정적 자극이라 평가. 단기 변동성은 있으나 장기 전망은 강하다는 입장.
다듬은 글
I think what it's telling us is the cost of innovation is collapsing which it has been doing I mean the cost uh of uh AI training is dropping before this 75% per year the cost for in inference 85 to 90% um I think it uh throws the weight more towards inference chips from training chips and so kathywood highlights the rapid decline in AI costs with training costs Falling by 75% annually and inference costs even more this shift boosts the competitiveness of inference chips over training chips positioning companies like paler to gain market share she emphasizes ai's transformative potential in autonomous mobility and healthare particularly through Innovations like crisper and sequencing Technologies would views this as a key moment for the US tech sector with global competition spurring Innovation while Market volatility raises questions about AI Investments like Nvidia she believes long-term prospects remain strong with practical AI applications set to flourish investors are advised to manage risk as market dynamics evolve awaiting Clarity from upcoming earnings reports from major Tech players despite short-term uncertainty ai's impact on productivity and Innovation is expected to be profound Nvidia is highly exalted and deservedly so in the training chip space inference is a lot more uh competitive okay do we do we fully understand the implications of that because the question everybody's asking this morning is do I buy the dip so if if we're seeing the landscape changing is it too early to make that transition and understand what the implications of it are yes we've been studying the tech stack and where we think it's going we're going to put out Big Ideas in I think a week uh sh give us some hints yes so here's a hint uh we do think there are share shifts taking place within the techstack uh we think platform as a service so think paler is going to gain significant share and this news yesterday uh I think uh highlights that and reinforces that uh we think infrastructure as a service uh is is going to stabilize a share uh Sal I mean um software as a service will come down a share however I've said all of that all of them will be growing rapidly it's just how rapid and palente we should say is one of your one of your Holdings of course a company you know very well Kathy so if inference is going to become cheaper is this going to be situation that drives more demand we were talking about you know is that where the positivity comes from for the companies that don't train but do in do infer yes how positive is this going to be uh it's going to be extremely positive so the those using AI whether for productivity or other applications are going to benefit we think the two most profound uh use cases right now besides increasing productivity for everyone uh autonomous Mobility so think cars drones and so forth and uh interestingly the healthc care space uh we think the convergence of various sequencing Technologies Ai and crisper Gene editing that convergence is going to cure more and more diseases over time so these sound like huge positives for the businesses that are involved in in that stream and what does it tell us about us Tech overall we had a conversation yesterday about whether this is some kind of end of us exceptionalist and that might be in terms of stocks broadly but in terms of the US tech sector are we talking about that or is this just a you know the spotnik moment that's been mentioned many times a chance for the US now to to catch up again yes I think uh competition is great for the US and uh deep seek is competition no doubt about it I'm hearing all of the Titans uh Tech Titans uh uh speak of it glowingly I don't think we know everything about it yet uh so no I think it's uh competition is good for the United States lowering costs is great for the world uh the collapse in the cost of innovation should help all regions of the world I will say uh that new productive forces which was shishi Bing's uh new phrase uh I think 18 months ago uh seems to be showing through right now common Prosperity was the previous one and I I still think that's an effect that was more about margins this is about Innovation so um we're looking more closely at China now deep seeks breakout calling into question the tens of billions of dollars big tech companies have coughed up for AI development the move lower also raising eyebrows around the valuations of these Tech leaders joining us now Deepwater Asset Management Gan Monster uh Jean I suppose the best way to ask you whether you think this is an opportunity or sort of sign of a greater risk here is to Simply ask you are you picking up more Nvidia at this point uh deep water does own Nvidia I can't say how we're trading it in the near term but I can say we do own Nvidia and dearra I think at the fundamental question there is the near-term and then there's the long-term and I just want to highlight one Dynamic around the near-term is that the Innovation around deeps is real and uh even if you think that they're embellishing some of the cost piece around that that Innovation piece is real but there's a much bigger question that needs to be asked is not about what the cost to train the previous models were but to get to that incremental step forward closer to artificial general intelligence does deep seek have some sort of a formula to do that in a cost effective way and my sense is that they will not be able to do that so I think that the uh the spending piece is on track the reason why I mentioned there is the near term uh around how these are trading versus kind of I'll say the near term over the next few days actually is that uh all eyes are centered on tomorrow night with Microsoft with meta Tesla's comments about the hyperscalers and most investors believe consensus is that they're going to have some sort of reassuring comments that the Market's going to be just uh that the AI trade the hardware trade the capex spending is intact and uh when that becomes consensus if that doesn't yield kind of a rebound in some of these stocks I think you could see a little bit of a near-term selloff if you're curious typically when you have these shocks in the system looking back to the run from 95 to 2000 it's about 30 days it takes from that initial shock that we saw yesterday to when things actually bottom which puts us right in sync with when Nvidia is going to be reporting their quarter I mean you own Nvidia Steph's got a move that I'm going to get to after this CU it plays right into what we're talking about here but how are you assessing the damage here Morgan Stanley today cuts the price targets of a number of semi stocks including Nvidia and broadcom and Marvel and Micron Nvidia goes to 152 from 166 MH whereas be today says they'd be buyers of Nvidia on this dip so 152 just to use that number on Nvidia would still be a nice 20% rise from here um which would indicate that you should not sell all your Nvidia what you should do though is evaluate risk and I watched the show yesterday obviously I'm here today we are all talking about risk management and for the last two days I've been talking to a lot of portfolio managers within City Partners outside of City Partners there are very few in fact I don't think think there's anybody who said that now is the time to back up the truck on AI by the same token nobody is selling is saying sell all these names outright that's not the conversation the idea that these are busted in some sort of post 1999 way is not what people are talking about now maybe that's wrong but we'll find out what people are talking about and by people I'm referring to professional portfolio managers is this did we go into the end of last week with simply too much risk were we too Blas consider Oracle for a second one of the stocks that's in the headlines uh the last two days it shot up like a moon rocket last week on Project Stargate and now it's given that all back you have to be comfortable with the level of risk that you have now what I'm saying to you just to be explicit is most people that I'm talking to are talking about trimming risk what I'm doing in light of that is I'm waiting because we are going to get more information as the week goes on I don't want to get ahead of where we are in today's uh rundown but I will say we've got earnings coming up from two of the major hyperscalers Microsoft meta let's see what they have to say there are going to be questions about deep seek on those conference calls I will wait for that information but I think this is a time for everybody to say not should we sell these all outright but did we have too much risk Nvidia um is at the highs of the day you know as we're sitting here talking and Josh makes the case that you know some of this might be overdone to at least some degree it's trying to get 5% back of yesterday's um huge selloff has I I got to ask you about Nvidia let's wrap it up with that yeah we got to you know I wish I came on the show at 9:00 the setup for Traders especially the Traders watching this you know you watch that 200 day moving average it's the first time it broke below it since December of 2023 maybe October 20123 it's been a while uh we finally eclipsed it every buy stop in the world was set to 122 as soon as it hit 122 was Off to the Races so your risk reward it was set up for a nice rebound trade and long term I still think it's great this story out of China with deep seek uh we shot first we're asking questions now and the answers we're getting uh Kyle bass was on your show he he doesn't trust it and I have more questions and answers so I think this was a buying opportunity just like August 5th with the Japan Yen carry trade remember that we sold off ridiculous amount uh we got an opportunity to buy and if it does fail the 200 day you get it we're going to wrap it there Jay Woods you have as much confidence today in Nvidia as you did last week in their ability to to beat and then continue to if not raise their guidance to the degree they had in the past cuz most think that that's not possible anyway but enough to justify the excitement and the move in in those shares um well yes I do I me we've been doing a lot of research on this and one of the things to understand is that like um training consumes a lot of uh computing power but inference uh especially at the edge I when you use some uh you do something on your iPhone that requires any kind of agent AI any kind of a AI function uh that inferencing uh function could be uh multiples of compute power needed um and uh you know the interesting result deep seek is run on Nvidia chips it just run on last generation Nvidia chips versus say upcoming Generations or highest end um you know Blackwell is coming from Nvidia we think there'll be uh plenty of people pursuing both ends of the AI Spectrum right there will be some still pursuing the absolute high-end Cutting Edge think about um internet search there was Yahoo uh there was then then there was Google but in between there were ask geves uh altav Vista uh bing has been around um that battle to be the ultimate dominant uh AI model is is going to continue for a while still and at the same time with the lower cost a lot of uh more pragmatic practical uh applications can be possible uh particularly through um you know software and uh uh Hardware the falling costs of AI training and inference are reshaping the tech landscape with companies like paler set to gain from increas competition in inference Technologies this shift boosts Industries such as autonomous mobility and Healthcare where Innovations like crisper Gene editing are advancing rapidly while leaders like Nvidia remain key investors must navigate evolving risks and opportunities in the market Global competition and declining Innovation costs highlight ai's transformative potential particularly in the US tech sector as companies prepare updates balancing enthusiasm with caution is crucial signaling a pivotal moment for reshaping Industries and practical AI applications how do you see declining AI costs impacting industry Innovation which sectors do you believe will benefit most from AI advancements
Comments 0