저자: DanIves · 외국인
요약 보기
AI 물리적(physical AI) 시장이 2035년 약 1조 달러 규모가 될 수 있으며, 그 절반가량은 자율주행차(약 5000억 달러)에서 나온다. 자율주행은 수백만 대 차량의 실주행 데이터 축적으로 휴머노이드보다 앞서 있다. 메타의 엔비디아 CPU·GPU 대규모 도입 계약은 엔비디아 훈련 우위를 강화하지만, 앤트로픽이 구글·아마존 칩으로 학습한 사례는 공급 다변화 우려를 보여준다. 사이버보안은 AI로 인한 위협 우려로 주가가 눌렸으나 Ives는 예산 확대와 M&A 증가로 오히려 성장 섹터가 될 것이라고 주장하며 팔로알토·크라우드스트라이크를 긍정적으로 본다.
다듬은 글
physical AI could be worth a trillion dollars. And in clip one today, we break down exactly where that money comes from and why autonomous vehicles are getting there first. Then in clip two, Meta just locked in millions of Nvidia processors. And there's something in that deal that tells us a lot about where the AI race is actually headed.
And then finally, we'll see Dan Ies making a case for massive growth potential of the cyber security sector. So we have the the one trillion addressable market 2035, but digging down into the notes, actually autonomous driving is a really big part of this for you. That's right. Our latest research shows that physical AI could really become a trillion dollar market by 2035.
That's 10 times higher than where the market is currently valued at. And uh that trillion dollar estimate spans four key robotics categories. autonomous vehicles, humanoid robots, advanced automation and in drones. And look, while I have really strong conviction that the late 2020s and early 2030s will be the decade of the robot, I also think that growth and adoption will likely come in stages rather than all at once.
I can see how autonomous vehicles could lead the the trend and set the stage. And in fact nearly half of that estimate uh of of the market growth comes from autonomous vehicles or in dollar terms that's about $500 billion by 2045. I think also intuitively this makes a lot of sense because autonomous vehicles clearly have a head start.
The technology has been around for nearly a decade now. The production process is there. it can leverage an existing uh automotive supply chain and I think more importantly the the AI models needed for autonomous vehicles can work with a much bigger real world driving data set that's collected from millions of vehicles out there and that's a very different story if you compared to where humanoid robots are at the moment.
So if you follow the arc of what happened with autonomous driving, what needs to happen for humanoid robots for them to make a meaningful contribution to your $1 trillion forecast for addressable market by 2035. I think the number one challenge that humanoids face right now is the lack of physical AI data.
And that's because when you think about humanoid robots, they really bridge the gap between the cognitive, the digital and the physical world. And in the physical world, the laws of mechanics and physics apply. Uh a humanoid robot needs precise instructions if it's going to function and perform properly in an unstructured world that is made for humans.
Take a simple example such as lifting a box. This is a very simple task for us humans. We have inbuilt dexterity. We have inbuilt intelligence. We know exactly what we have to do. But a humanoid robot needs precise instructions. It needs to know exactly how much force to apply, where to apply that force.
And that simple example can get really tricky if something changes. If for example, the box is actually hier than expected. If the surface is slippery, that means that the robot needs a new set of instructions. And the the challenge is that there is no dictionary out there. There's no single database that can tell us, look, for a 50 lb box, you need this amount of energy, this amount of force.
All that needs to be built from scratch in order for the technology to scale and to become more more useful in the real world. Autonomous vehicles are carrying the physical AI market on their back because they already have what humanoid robots are still desperately trying to build, a massive real world data set trained across millions of vehicles.
So, the trillion dollar number gets thrown around a lot and I get why people tune it out. But when you actually break it down, autonomous vehicles, humanoids, drones, advanced automation, the math starts to make sense. And the key insight here is that these four categories are not moving at the same speed.
Autonomous vehicles are pulling ahead, no pun intended. Nearly half of that whole estimate comes from AVs alone. The reason is data. With self-driving, you've got millions of cars on the road right now, every day, collecting realworld miles. Rain, construction zones, someone cutting cross through lanes at 70 mph.
I've seen a lot of that before. The system sees all of it and gets better. That data set compounds over time in a way you just can't shortcut. Companies trying to enter this space aren't starting from zero. They're starting from years behind. And the gap widens every single day. Humoid robots face a completely different problem.
The physical world is brutal for a robot. Lifting a box sounds trivial to a human because we've been doing it our whole lives. Our bodies know how much force to apply, how to adjust when something's heavier than expected, how to catch ourselves if our foot slips. A humanoid robot needs all of that spelled out, and there's no master database to pull from.
You have to build it from scratch in the real world at scale, which takes years. That's why AVs are a decade ahead in deployment terms. What I find interesting about this framing is that it gives you a road map. AVs set the blueprint. They prove that physical AI can work. They generate the revenue. They build investor confidence in the broader category.
And then humanoids follow that same arc just later. So when you hear trillion dollar forecasts, the right question is who's furthest ahead along the curve. Right now, the answer is whoever owns autonomous vehicle deployment at scale. And the data advantage story doesn't stop with autonomous vehicles.
In clip 2, we'll see exactly the same dynamic playing out in AI model training, where the AI company with the best chips trains the best models, and everyone else is playing catch-up. Meta just committed to millions of NVIDIA processors, not just GPUs, but CPUs, too, which is a bit of a new wrinkle.
And buried in this deal is a detail about Anthropic that tells you something important about how Nvidia keeps its grip on the AI market. Meta planning to deploy millions of Nvidia processors in the next few years. A deal that could add billions of dollars to Nvidia's balance sheet. Bloomberg Intelligence global head of tech research Mandep Singh joins us now.
Mandep first of all I mean does Nvidia need this? It's up pretty sharply off the back of this. I thought there was already like pretty robust demand for what Nvidia is selling. There is and look this is probably Meta's way of locking in that Nvidia supply. So not only are they using the GPUs but they're also talking about using Nvidia's CPUs.
So that's new. I mean traditionally all these companies used x86 uh architecture and Intel CPUs along with the GPU cluster that Nvidia had. But now uh you know with this deal they're talking about using uh Nvidia CPUs. So that's one big change. The other thing for Nvidia here is Enthropic has had a very steep rise in the last few uh months.
Entropic models aren't trained on Nvidia GPUs. They're trained on Google GPUs, Amazon. So from that perspective, you have a company that's doing really well on the model side that is not trained on Nvidia GPUs. and and so from that perspective I think for both of them it's important that you know Meta comes up with a model that uh is really good in terms of being a frontier model like Enthropics that is trained on uh Nvidia's latest cluster and obviously Meta needs to lock in the supply so it's it's I think symbiotic
that's that's a good point too because also Gemini I mean of course they're yeah using Google's I mean in all of this can you will other companies look at the advances of anthrop ropic of Gemini and start to rethink their supply chain because of that that they don't want to be too reliant on Nvidia.
There is that aspect but what Nvidia is saying is our latest black wall chips will give you 30x more token output than our prior architecture from Hopper. So for the same unit of power you can have uh 30 times more output and so that throughput aspect of Nvidia chips is always there and that ties into how good the model will be that is trained on the latest chips.
So really what Nvidia needs is companies like OpenAI like Meta training their models on Nvidia's chips and having uh somewhat of an advantage compared to the ones that are not trained on Nvidia. Nvidia's training advantage compounds over time because the best models get built on Nvidia hardware which means Nvidia customers keep winning which means everyone needs to train on Nvidia to stay competitive.
So the analyst man deep makes a point here that I think is worth sitting with. Anthropic, one of the world's hottest AI labs, trains on Google and Amazon chips as well as Nvidia. And Meta looks at that and essentially says, "We can't let that be our story." Because if a Frontier model can be built without only relying on Nvidia, suddenly your entire competitive moat looks a bit shakier.
So the metad is partly about securing supply, but it's also about making sure the best model trained on Nvidia's latest cluster is there. The throughput argument is the crux of it. Nvidia is claiming their latest architecture gives you 30 times more token output for the same power as the previous generation.
And when you're training a model at the scale that Meta or OpenAI is operating, that efficiency gap translates directly into model quality. More compute, faster training runs, more iterations, it compounds. Companies that can do five training runs in the time it takes competitors to do one will pull ahead and stay there.
The CPU piece is underrated. Traditionally, everyone used Intel or AMD's CPUs alongside Nvidia's GPU clusters. Meta moving to Nvidia's CPUs, too, is a signal that Nvidia is trying to own more of the full computing stack. Not just the flashy accelerator part, but the whole system. If that works, it deepens the lock in considerably.
You're not just relying on Nvidia for one component. You're relying on them for the whole infrastructure layer. And zooming out, this is why the metadal matters beyond the headline numbers. It signals that the AI infrastructure buildout is still very much in full swing. Companies at the frontier need more compute, not less.
And Nvidia is positioned as the default choice for anyone who wants to train a model that can compete. Anthropic success on alternative hardware is a real data point, but it's also a warning shot that Nvidia takes seriously, which is exactly why they need Meta building the next great frontier model on their chips.
Now, let's watch Dan Ies making the case for the massive growth potential of the cyber security sector. Analyst Dan Ives, he joins us now from Wed Bush with what to expect this evening and what's been taking place in in that group. It's good to see you. Great to see you. So, why has cyber been swept up and everything that's been going on?
Yeah, I look I think part of it is just this software, right? I mean the view is that can cyber be disintermediated by some of these tools whether it's anthropic or some of the you know really the startup tools that are coming from an AI perspective to my point is look come into this year all of our checks are showing cyber security should actually see anywhere between 3 and 500 bit acceleration I think some of that is because you're see as you see the AI use cases deployed you look at pow they don't acquire cyborg if it's not for AI crowd strike all the stuff Georgia and team are doing to I think this is one I get it's a guilty till proven innocent but cyber security this is going to be I think it it's a moment that inlects and ultimately it's going to be a huge growth area for the sector that I don't think is being factored in here right now it's being used as a headwind but tonight it's very important to hear from the cash he is a great person I
I've I mean part of me wants to say well h how can you have checks on things that we don't really know what we're checking for yet. Like we don't know what these other tools are going to do and how they're going to disintermediate certain industries until we actually see it. Yeah. I think to even get more granular typically going into a IT budget cycle, let's say about 5 to 8% of maybe that budget is what I'll call it cyber security related.
Now what's happening because so many more of these comprehensive solutions AI platforms end to end I'd say about 70% of IT budgets have an AI you know sort of angle cyber security is now 5 to 8% you could argue from everything we're seeing 20 25% of now the potential budgets that they could go after so when I look at Palo Alto crowd strike zcaler like Scott I think this is one I know we've talked about a lot in software but cyber security I think is being mispriced that AI is a headwind where in fact I think it actually is going to change the view of the sector for years to come as well as also I think it's going to be a huge cause for M& I mean you've looked at PaloAlto specifically I think on this very program and declared that it's in your words table pounder right I mean you still believe that or is in some respects the jury still out even for you want to see what they deliver tonight because the stocks had some upset in the past around earnings it's not like this thing just goes straight up
yeah no doubt look there's three reasons why this is I I view it as double table pounder because the cyber art deal will be viewed as the deal that changes POW out there. I think what Nikesh did there, you're talking about 25 billion I could argue is going to generate hundreds of billions in the future of market cap you know ascension given what that gives them a full end platform two if you look where they play just like crowd strike
we are now getting to an inflection point where you are going to start to see AI I think 25% of the deal flow ultimately will be positive influence that's not factored into the street numbers sentiment super negative and that's Why? To me, Palo Alto, Crowdstrike, those are the ones that I view as just names that not even just in CyberScript, but overall attack or two of the top.
And YouTube isn't just entertainment. It's one of the best client acquisition tools because it builds trust at scale. We've helped businesses grow from scratch to a $100,000 month just by launch and let's see how YouTube could help your business scale.
Comments 0