저자: DanIves · 외국인
요약 보기
AI 인프라 수요가 폭발적이며 Moore's Law 종료에도 GPU 가속 컴퓨팅이 10~20배 비용 절감 효과를 낸다고 주장. 생성형 AI 소프트웨어·디지털 엔지니어 확산 강조. 트럼프 행정부 하에서 규제 완화로 AI 인프라·소프트웨어 M&A가 대폭 늘 것이며, 테슬라·팔란티어·애플 등 기술주에 긍정적 전망. 엔비디아 중심 AI 인프라 수혜 지속 전망.
다듬은 글
the same resource which saved even more money the use went up once more also all of this stuff that's been going on for the last 10 to 15 years has hidden the main Trend which is that Moore's Law is ending we found another two times in cost stings which hid the end of transistor scaling and CPU scaling then all of a sudden we had already found ways to make these things more useful and cut costs so we're done that's the reason why our data enters and computers are getting bigger right now the first thing that's happening is faster computers a 20 to1 speed up is normal when you take your data processing work and speed it up with Nvidia in the cloud this is because spark is probably the most popular data processing engine in the world right now you get 10 times the savings but you have to pay for the Nvidia GPU that helps the CPU the cost of computing might go up a little maybe even double but the time it takes to compute goes down by about 20 times so you save 10 times as much this kind of return on investment borrow I is common for Accelerated computers I would tell everyone to speed up everything they can and then run it with gpus after they're done improving it that's an immediate return on investment when you speed things up later we'll talk about generative AI in the first wave infrastructure players like us and all cloud service providers put the infrastructure in the cloud so that developers could use these machines to train the models fine-tune the models protect the models and so on this is a great investment because there is so much demand that for every dollar they spend with us they rent something worth $5 this is happening all over the world everything is gone and this is in such high demand it's crazy of course we already know about some of these apps open ey chat GPT GitHub co-pilot and code writers that we use at work it's amazing how much production has grown today there isn't a single software worker in our company who doesn't use code generators these can be the ones we make ourselves for tbaa or USD another language we use verilog C or C++ that being said I don't think every bit of code will be written by software Engineers anymore in the future I think all of our software Engineers will have digital Engineers working with them all the time when I think about Nvidia I see that we have 32,000 workers but around them are hopefully 100 times more digital Engineers viia is that the system you built today for training will work great tomorrow for inference I think that most of chat GPT is learned from the same kinds of systems that it was just trained on you can draw conclusions from it if you can train on it you're also leaving behind a lot of Machinery that will be very good at drawing conclusions there's no doubt in your mind that you can use the money you earn from the investment to build a new infrastructure that can grow with your business you know you'll leave something useful behind in just one year Nvidia and the rest of the community will work to make the algorithms better so that the rest of your infrastructure gets five times better it's going to stay that way the way people think about infrastructure is I built it for training today so it must be good for training we know it will be good for inference too there will be more than one measure of inference it's helpful to have a bigger model to distill from if you want to distill a smaller one because of this you will still make these amazing Frontier models they will be used for groundbreaking work making fake data teaching smaller models and Distilling them down there will be huge models small models and even very small models at the end the tiny models will work pretty well though they won't be able to be used in as many situations in one tiny domain they will do a very specific job incredibly well in that one domain we will see superhuman performance from a very small model maybe not a small language model but tiny language models tlms or something like that in your new note you say that more business deals are made when Trump is in office why is that the case and where yes I believe this time for M and I will be just right Regulators have been a big problem as has overconfidence in stock prices I believe that there will soon be a huge amount of mergers and acquisition itions mainly involving AI infrastructure software and protection I believe we've seen a lot of them in public and private it's kind of like a tech prom but no one is really asking anyone to dance on either side and now I think this thing starts I think we could see a lot of mergers and Acquisitions under a trump government okay so that's in the public markets I'm going to come over to you Al as we already said you're an early stage investor in both Pinterest and eron you're in touch with the private markets a lot so now that President Trump or president-elect Trump is in office what does that mean for tech companies in the private markets there was talk that prices were falling before Trump took office does that mean they are no longer falling once more I believe it is very good for private companies in general I think there are some companies especially ones with ties to Elon Musk like SpaceX that are just very positive we've seen what Elon Musk can do when the White House is basically hostile yeah I think it will be amazing what we see if the White House is neutral or friendly I think this is true in all private markets oh Alex you know we talked about this off a you call it the musk economy we have a picture of it that calls it the Elon Musk corporate Empire there are a lot of companies there it's xai the boring company X Tesla and SpaceX so when you look at these different companies what kind of value increase do you you think we have now that Elon Musk may be part of a government that seems to be more friendly in a way I think SpaceX and Tesla are so important that they helped Trump win the election now that there are fewer rules and more people are interested in building in the US I think these potentially amazing infrastructures will be put in place much more quickly across the country but especially in the redder states that voted for him Dan I'd like to talk to you again in your new note you say that you expect to see more mergers and Acquisitions M and A under a trump President we were together though on Election night you were in the room with us while we talked about this many stocks went up including Tesla which we've already talked about in detail but you say that you see other stocks like paler Google and apple doing even better why does that happen this is going to be a great time for a technology and I think Panter will be at the Forefront of it all everything carpent his team do will cost the US Department of Defense a lot of money especially when it comes to the Belt look at Microsoft Amazon and Google they are all hyperscalers this is really going to be a new age of AI infrastructure also when it comes to Apple I think this consumer ey Revolution goes through Cupertino even though there are some fears about tariffs with China that's why I think Trump getting involved is good for Tech I agree with everything Alex said the new era of the this Tech bull market has begun so now that there's a chance of less control some areas have seen an improvement do you think Tech will keep getting that much better in the future it seems like everyone agrees that Lena Khan will no longer work at the FTC it will just be a different regulatory setting Khan has been a pain for Tech and I think that when she leaves the whole area will be different I agree with some of what Alex said the rules will be so much friendlier for deals and mergers and acquisitions as this ey change continues I think big Tech will become a little more aggressive when it comes to mergers and Acquisitions especially when it comes to I
Comments 0