Microsoft Abruptly CANCELING *MAJOR* AI Spend | WARNING.

요약 보기

마이크로소프트가 미국 데이터센터 임대 계약을 취소하고 있다는 채널체크(Channel check)가 제기됨. 이는 MS의 AI 과잉공급 우려와 연결되며, 딥시크 효과와 AI 수요 증가 속도에 대한 의문을 동반. 과잉 설비투자(capex) 조정이 현실화되면 GDP와 반도체 밸류체인에 부정적일 수 있으나, 자금이 휴머노이드 로봇 등으로 재배분될 가능성도 언급. 단기적으로는 이번 주 반등 가능성을 낙관하지만 엔비디아 실적과 향후 capex 동향을 주시해야 한다는 입장.

다듬은 글

※ 자동 자막 원문(다듬기 전)

01. oh hold up is Microsoft actually cancelling data center leases and what does this potentially mean for AI and whole like freaking economy that's surrounding it or is this potentially just because of a uh you know reduced partnership between open Ai and Microsoft after all open AI partnering with Stargate which you know is a partnership between Oracle and soft bank and not exactly uh you know just Microsoft here so we got a few things to put together but uh first I got to say this hey everyone me Kevin you got to get the inro yeah a couple days ago we talked about saan Adela and this interview that he gave where he talked about hey it's it's really important to make sure that artificial intelligence Supply does not so far exceed the level of demand that you have so that you actually Escape your ability to earn a return on investment so let me put it in English if you have a lemonade stand you should not have 1 Million Cups of lemonade on your lemonade stand maybe you should have 10 and try to sell those 10 first and then make more now of course there are economies basically at this point built around supplying these lemonade stands with chips and server racks and uh artificial intelligence compute capability uh energy infrastructure you name it uh so this actually came up via a rumor first and before I made my last video reviewing the satian Adella interview I actually saw the rumor but I didn't cover it not because I you know didn't want to give you the most recent information but because at that point it was unclear if it was just Rumor Mill there are plenty of people just harping on this idea that oh the AI bubble is over uh and I don't want to just Echo what is potentially misinformation but now that I'm getting more and more corroborated data on this it it's worth noting that it could still be wrong there could be idiosyncratic problems with Microsoft or this could be the beginning of a larger Trend uh so what I'd like to do is go through this and and then analyze what could be going on here I'd also like to uh just first speculate uh that uh it is entirely possible that the Deep seek effect is still present now most people argue nah man the Deep seek effect isn't going to have an impact it's actually just going to lead to an increase in usage for AI and look I I believe that AI usage can go up but remember if you can make it so much more efficient to utilize AI you know how many more queries are you going to send to chat GPT in other words if all of us together we're in a classroom together let's say and they're 25 of us and and we send 10 inquiries a day to GPT okay fantastic we're sending uh 250 inquiries and if half of you are using grock now and a quarter of you using are using cloud and and a quarter of you are using GPT then there's even less on GPT but the aggregate is still 250 inquiries right if it now becomes 10 times cheaper to provide that answer because of supply side efficiencies in other words they can manufacture those answers for us with one/ tenth of the compute does that mean all of us are going to 10x our output are we going to go from 250 queries to 2,500 no of course not now is it possible that maybe one of you was research scientist and you're like my gosh I'd love to increase my input but the cost per cury query is too high for my research purposes and so I would like to increase my input of course or output of course you know you're going to send more quaries but not everybody will so there I don't necessarily agree with this argument that unless there's this significant Improvement in how good AI is that everybody is going to go from the 2023 you know 2022 I should say level of AI queries basically zero to the level of 2024 AI queries call that 100% I don't necessarily think we're going to go from 0 to 100 to 200 just because the supply side is more efficient maybe we'll go from Z to 100 to5 or 110 right some users will use more but not everybody will uh and and so this means the underlying deep seek problem is still present that there is a possibility that Satia Adella was correct in that we don't want to overs Supply let's actually build the infrastructure let's get a return on the infrastructure and then let's actually uh uh you know continue to reinvest at that point now a lot of folks especially on X and I think there's certainly a lot of knee bending going on here you know people like to get on the floor and and go you know for Elon um so there's a lot of talk about how great and wonderful grock is and and it might be wonderful maybe maybe it's the best AI you've used that's fine and I'm not here to say that one AI is better than another uh you know just because uh who was it uh you know I I think it was was it GPT that that beat the Minecraft uh game at the end but Claude didn't uh and Gemini I don't know had some problems with uh creating the workbench at the beginning I don't know just because one is better than the other I don't necessarily think leads to more quaries and that's the big question right how many more quaries are we really going to get how much increase in demand are we really going to get and I think Adella is actually making a reasonable approach here where he's being smart and and I've been critical of his sort of blind optimism on wanting to continue to spend but then since then and I'm not saying you know he's changed his opinion because of me and of course we're going to go through this document here but since that moment where where I was sort of you know engaging on X about like this is ridiculous the podcast came out where saia's actually like yeah you know we we want to evaluate some Roi before we keep spending I'm like hm okay so maybe he's hearing the criticism and some investors are saying yeah no we we need some Roi here before we keep going because if we just dilute Ai and we're not actually creating more demand everybody is just taking you know pieces of of the same pie the pie isn't actually growing then why do we want to keep spending you know does is Elon Musk really going to build out another Nashville Tennessee uh you know super infrastructure uh for for xai or does he need to now focus on marketing and getting you know I mean they've already gotten Gro to top and app down s which is fantastic you should go download it give it a shot right I I have no horse in this race I don't really care what you use try them all is what I always say but is he now going to go out and build another sort of Giga AI Factory using Blackwell chips or are we going to focus on some Roi here first and I think that's exactly what's going on here so what do we have here Channel checks indicate us data center lease cancellations by Microsoft our Channel checks indicate that m Micosoft has canell leases in the United States totaling quote a couple hundred megawatts with at least two private dentist data center operators number two Microsoft has pulled back on the conversion of uh okay basically like uh statements of qualifications these are pre-lease statements that you get saying yeah yeah yeah we're we're capable of building out that data center for you uh and and those usually convert to leases but apparently those are getting canned uh and number three has reallocated a considerable portion of its International spend uh to the United States uh which is good you know that's basically what Trump wants right is more of that reallocation to the United States so so seeing a reallocation to the United States is great but this cancellation of leases has also occurred in the United States so it's kind of like okay got it you're going to move more of the data centers to the United States probably to you know also to some extent B the need to the United States Uh current federal regime which is probably a good thing because we want our data locally anyway uh even if it's a little bit more expensive maybe we'll get an upside of lower latency though that I don't think that's a terrible terribly big issue what's the difference of a few milliseconds on your GPT response I mean I suppose maybe for for certain tasks a lot but you know this is not like we're playing video games when we're asking for J chat GPT to respond to our generic encyclopedic responses or inquiries anyway uh when coupled with our prior Channel checks this now points to a potential over Supply position for Microsoft for data centers right our Channel checks indicate Microsoft has canel select data centers and leases uh and has pulled back on conversions to leases and pulled back on International Market expansion so you've got multiple things going on here I'm going read all this word for word here but our checks indicate that in some situations Microsoft is using facility power delays as a justification for termination in other words you're looking for excuses to terminate a contract recall as we highlighted in our 2023 takeaways this is the same tactic that meta used to cancel multiple data center leases in the United States after we learned in our checks that meta had then canceled $4 billion of capex programs related to the metaverse okay so now it's it's worth noting and then they basically say two weeks later meta ended up cutting their guidance uh for capex now I just want to be clear with a few things here I'm not saying that Microsoft's not going to spend this money in capex it could just get reallocated to something like humanoid robotics I actually think all of the you know the big mag sevens maybe you know some exception uh pretty much all of them are going to get into humanoid robotics uh and I mean this is a sector that's that's just going to blow up in time this is this is obvious I don't know how soon we should actually expect humanoid robotics especially if there's a larger macro event uh that that delays all this which is entirely possible uh but uh I I would argue that humanoid robots are going to be part of our future so is it possible then that we see this sort of q1 confirmation of capex transition to different forms of capex absolutely and is that going to be good for other economies yes but then we want a heads up on that okay so we're a robotics companies we can invest into my VC invested into robotics so everybody who invested with me in that Venture Capital fund you know we're we're in on on humanoid robotics that said when it comes to the potential for slow down and capex and then no offset that's worst case scenario because now we're not looking at okay well the money is just going to go to different forms of chips it's going to go to actuators it's going to go to battery manufacturers uh and and certain software plays it actually just goes nowhere right we want money being spent money sitting in a savings account at Microsoft is not good for GDP it has a velocity of like one in other words it's spent one time goes in Microsoft pocket Microsoft's pockets and then dies that's really bad for GDP ironically the biggest velocity Creator is the government because it's so wasteful the Mone actually circulates a lot so you know if you're not getting cutting via a lack of spending at some of the big plays like Microsoft and the government is cutting it's is not an ideal scenario especially when the cape Shiller index put you at you know pre Great Depression and pre Doom bubble valuations that said we're probably going to bounce in markets tomorrow I I would anticipate that uh unless there's some kind of crazy news uh everything over the weekend seems to have occurred pretty much as expected uh German elections uh and otherwise uh so I I would anticipate some form of Bounce in markets tomorrow and I think it's an easy sort of bet to make unless there's some bad news we get pce this week I don't think that's going to be particularly bad uh mostly because the pce components from PPI and CPI were actually pretty decent that'll carry over to PC so I don't see that as a problem and I don't see any kind of like unemployment Catalyst I we really don't get another jobs report until uh you know two weeks from now so we've got some time on that that's like one and a half weeks from now uh but um I I suppose what we want to pay attention to is the potential for any kind of downside announcements from these capex companies so um let's see here let's take a look at what else is happening here so the article then goes on to say that uh well they sort of give a little precursor here and say hey you got two weeks which by the way I should be Jet rided by then which would be really cool I'm actually in Florida right now in my short shorts ready to go work out uh but I'm in Florida because uh I'm doing my multi-engine wrapping up my multi-engine trading over here and then uh getting my jet rating uh and uh yeah I mean that's kind of cool uh but anyway at the same time we're still going to be doing our course member live streams over the next few days uh and we've got that price increase on the courses coming at the end of this week so uh hop into those if you haven't yet for uh my uh my signals on fundamental analysis trade alerts real estate analysis you name it we've got some wonderful things there and some new lectures coming out on trumponomics but take a look at this why is this happening says TD Cohen TD Cohen says while we have yet to get to the level of color via our Channel check so that we would like to in why this is occurring our initial reaction is that it is tied to Microsoft potentially being in an over Supply condition okay we we to know that as we highlighted in our recent takeaways we learned via Channel checks that Microsoft walked away from multiple deals as in early mid-stages of negotiations on other deals that have now been cancelled and let uh OnePlus gws of ladds of intent on larger footprint sites expire has also walked away from at least five land projects that it had under contract in multiple tier one markets at the same time we highlight the magnitude of potential Data Center capacity and walked away from and the decision to pull back on land acquisition in our view indicates the loss of a major demand signal for micros that was originally uh you know responding to what we believed was a shift in their appetite for capacity tied to open AI which recent reports seem to reiterate keep in mind open AI has has just partnered with you know the whole Stargate project uh but uh we believe this indicates a decision to pause construction on data centers in Wisconsin that was a big plant by the way that was a big big big big deal uh prior channel in checks indicated uh those were designed or the Wisconsin facility was designed to support open AI so maybe somebody else was picking up that contract instead of Microsoft but anyway soft Bank who knows but there's also a capacity that is likely procured particularly in areas where uh capacity uh blah blah blah blah okay basically they think that there's a potential Microsoft has some new internal forecast that's like all right look we got plenty if we need more we could just go rent it from somewhere else because remember this is not just Microsoft building out their own Data Center capacity like you could pick up excess capacity from other uh data set of providers as well there are plenty of them if you want to go rent an h100 you can do that there are plenty of providers just Google rent h100 and you can kind of track prices over time I actually recommend you make a little spreadsheet and you just track those prices over time and you can sort of see what overall market demand versus Supply is those current excuse me per hour prices help you keep your finger on sort of the pulse of the market and they've been declining because well now you've got potentially better chips coming out uh and everybody's been getting into Data Center buildout so is all of it really necessary not necessarily and now the question is what happens to the economy if Microsoft pulls back Amazon you know and then that opens the door to Amazon pulling back which opens the door to meta pulling back which reiterates why Apple didn't blow all the money on the data centers on commoditization in the first place which then you know makes people wonder okay is Elon Musk raising for xai at Peak market prices probably uh you know so I don't know you know what does that then do to uh capex plays like uh you know uh rack installation companies rack providers qualcom broadcom uh I mean the whole soft the whole chip sack basically now you know jansen's like oh you know investors got the Jeep deep seek uh movement wrong okay now we know we have Nvidia earnings coming up and they're probably still going to be okay I still maintain that it'll probably take 6 to 12 months wants to actually see a Slowdown but I guess we'll see when the numbers come out we'll certainly be covering them so anyway uh that's what I've got for now thank you so much for watching I find it all very interesting and I look forward to your commentary check out the courses over at me kevin.com