저자: TedMortensson · 외국인
요약 보기
인텔은 파운드리(18A) 기술 문제로 침몰 중인 상태이며, TSMC의 지분/합작 참여가 있더라도 TSMC가 IP와 2·3나노 기술을 쉽게 내주지 않아 구조적 한계가 있다고 봄. 암(Arm)의 자체 AI칩은 엔비디아 대비 통합 솔루션에서 수년 뒤처져 성공 확률이 낮고 고객사와 경쟁하는 구조라 불확실. 클라우드 AI 투자는 과열 후 소화(digestion) 국면이 필요하며 추론 가격 하락이 진행 중. 엔비디아는 Blackwell 전환 과정의 마진 압박과 GB200 구리 백플레인 대신 광학 솔루션 대기 이슈가 있어, 2월 실적과 4월 가이던스가 관건이라고 언급.
다듬은 글
there's been a few shifting pieces recently in the chip race as companies fight for a spot on the chip sector leaderboard tywan semiconductor reportedly considering running Intel's us factories according to Bloomberg Intel shares pulling back from a jump earlier in the week where US vice president JD Vance highlighted the importance of us-based chipmakers here with the latest on what's ahead in the semiconductor landscape bar technology desk sector strategist Ted mortensson um Ted it's really great to have you here because we keep getting these headlines Fast and Furious today the latest a report that the Administration has approached Taiwan semiconductor about this stake partnership what have you with Intel and Taiwan semi to manufacture chips in the US what do you make of that well Intel's somewhat of a sinking ship right now um they're having a lot of trouble on their 18 18a um Foundry uh uh node they just cannot along with Samsung that they're really having at the 3 nanm uh range now you take TSM which is I would say the goldplated area of manufactured at these Advanced nodes they're already Manufacturing in volume at 3 nanometer they've cracked the code at that node and they're really moving to 2 nanometer which is the magical node and the reason why they have such an advantage is that at 2 nanometers you get about 15% better performance and about 30% uh better power profiles and when you talk about geni you need to be there and Intel's not so in some respects uh the administration's move here is not unexpected uh uh Intel is a National Asset not only from a uh security standpoint but Manufacturing in the US and they need some help and tsm's the only one that can render the life Boat Ted is it enough for Intel though I mean you call it a sinking ship would this be enough to guess plug the holes in that ship well they got to get rid of Foundry I mean it it's kind of a boat anchor on their profitability if they get the 18a node to work at least they'll be on the same node uh as AMD is AMD is um has significantly uh beat Intel up both on the client and the server side the server side is where they really need to be at that next Generation node because of power and that's where all the gross margin is so if they can get help on Foundry if you look at uh Intel's Foundry their largest customer is Intel uh at the server at the server level which again carries that high gross margin they got to get it right and they're just not there yet Ted is a potential for either a stake sale partnership whatever it is with t TSM is that a reason for investors to own Intel right now it seems to have been this week well I think the weightings in when I talked to portfolio managers what um the weightings are very di Minimus so you're really looking at evaluation play and if you look at the call options that were done early in the week they're roughly at the 25 level um this would dramatically increase I would say Intel's valuation of getting rid of that Foundry business but TSM is not going to do it for free uh the Trump Administration has has to negotiate a deal with TSM that they get pre uh preferential treatment you know not only on the on if they do buy The Foundry at the price they can buy it at number one and then an earnout provision uh they're not going to give up the the sacer count of of their IP at the three and two nanometer range they're just not going to do it it's it's uh really puts their model at risk if it if it's not structured the right way that they uh truly profit from that combination Ted I want to shift gears to another company here report out earlier this week that arm will be launching a new AI chip with meta as a customer C curious what you just make of that headline that report is that something that you think overall makes sense for arm are they going to be able to compete in that chip chip space in that way it's a great question too I mean arms in a arms late here they're a royalty company that that companies a contract with on the royalty basis for their designs their designs are fantastic I mean if you look what's happening on the PC and also on the data center side on gen a lot of these uh designs are arm-based now the problem though is that they are late Nidia is years ahead of them as it relates to an integrated silicon solution whether it be networking whether it be Memory whether it be GPU and CPU that's going to be a very tough nut to crack um if arm is going to be successful they're going to have to be 20 to 30% better than any other chip uh out there and that's a tall order the other thing is they are now directly competing with their customers and that would be Amazon on trainum and their inference chip Microsoft meta it's interesting that meta is going uh along with arm and also Google I mean um a lot of these designs are Arm based and if arm gets in Into The Fray here going to be a direct competitor not only with Qualcomm also so uh I don't think it's that clear and I don't think it's um quite frankly there's there's a lot that we don't know and from a p&l standpoint they really don't have a choice they've got to get in this the semic space Ted I'm curious your take more generally on kind of where we are in this gen gen aai investment cycle obviously we just got a lot of big announcements about big big numbers that the hyperscalers are spending um but but it you know you don't sound very enthusiastic certainly about some of these chip makers the market you know doesn't feel as enthusiastic right now what do you think well take a look at the like the Microsoft chart or take a look at some of the the cloud Titans they're they've they've kind of um gone sideways for a while and I think what's happening in the Gen area it's it's one of the biggest Technologies In Our Lifetime we're just spending a ton of money and there's been some advances not only on the Silicon side but the software stack where if you look at where training pricing and especially inference pricing if you look at listen to the paler call it's a second quarter that they basically said that inference prices pricing is imploding so it's getting cheaper to run these models and until we Baseline and see really what the model is the unknown is how much these Cloud Titans are spending I mean we're up 50% in Cloud spend uh over 200 billion for the big four in 2025 that's an astronomical number and I just think uh for AI really to get going we're going to have to have a digestion phase and then the market will really really pivot up to the upside TB we've talked AI chips for five minutes here and we haven't really talked about Nvidia I can't let you go without asking you about it overall where do you feel like sentiment is at with that name and to sort of what you were talking about there are we sort of at fair price at this point and it feels like maybe Nvidia needs to start doing something a little bit we need to see a little bit more over the next couple quarters or how are you sort of feeling about that name that I'm sure you get asked about quite a bit so there's a lot of open questions on Nvidia from a retail perspective this is a very very lazy long and in some respects a retail investor is taking a knife to a gun gunfight on what uh the institutional um investors are are are looking at um this quarter is going to be very interesting and the reason why is um Nidia is going through an architectural shift from What's called the old h100 h200 GPU architectures to Blackwell and there's a there's a server called the GB uh 200 without getting too specific this is a copper based uh back plane server that Nvidia um introduced and was supposed to take off in reality what we're finding through our research is the hyperscalers want more of this Optical Solution that's associated with the B300 so they're waiting and if the hypers scales are waiting on this infrastructure transition uh Nvidia is going to be forced to sell more uh Blackwell Standalone gpus that will put some Pro some pressure on Gross margins to what degree we don't know and if you look at the a January quarter coming up to be reported in a couple weeks here a week week and a half or so um that comp is up 72% you overy year where the street is right now um we just have to see them exceed that that very difficult comp and then where the institutional investors are really really focused is kind of seeing the the the actual print for April if that exceeds then I think you're going to see Nvidia move up and uh past the 140 level uh but there is there is some real big focus on on on this transition infrastructure um discussion on uh and also gross margins on how that April guide comes in that's what everybody's waiting for is that April guide well in the meantime of course we'll get those earnings on February 26 and see if we get some guidance and how that looks Ted thank you so much for your perspective I appreciate it well have a good have a good weekend thank thanks
Comments 0