Transcription
to it. Uh, not going to read through this stuff. You guys can go through. Uh, I will just say that towards the towards the end or the last third, I am going to do a deep dive in the semiconductor names using an interview from Jensen Huang. A lot of you have asked for, uh, just some advice on how to use AI more and more from the investment side. So I will spend some time on that. Um, there is a theme for this one that I just want to make sure you guys, um, start the process. You won't lose your job to AI. You'll lose your job to somebody who uses AI. Um, Jensen Huang, inside or during an interview this week on the BG2 pod, which is the only, um, podcast I'm going to reference this week, um, he went into this and he did basically say that AI is the greatest equalizer. And without reading the entire quote in the entire story, I could not agree more. Um, it's the major theme that I think people have to understand is that if you don't want to lose your job, you have to start using AI. If you spend the time on AI, no matter how you've been in your career, you can reinvent yourself because of the power of it if you commit to it.
For the week, S&P down slightly, Russell down slightly, um, after good weeks. GDP numbers. Nominal GDP came in basically at 6%, reversing kind of the weakness we had seen over the last three quarters and in particular the first quarter with all the trade disruptions from the tariffs. Um, the more stable final sales to domestic purchasers just continues to hum along at very, very solid levels without the distortions. This is the thing to look at just to see how stable GDP is. Uh, again, this is the real, uh, side. So came in at 2.9% annualized. And for Q3 so far, we're tracking at 3.9%. So the economy in the first half of the year was under 2%, we're tracking at close to four right now. Let's assume that over the first three quarters with all the tariffs and everything having gone through, we're still going to be a very healthy two and change percent. And at the same time, core PCE, uh, again, for everyone screaming policy mistakes, I can't emphasize enough to people here. The word means nothing. It doesn't do anything to help you make money. It's stupid. Uh, and I still read it almost every single day. Core PCE. Now remember, these are the numbers so far. This is the six-month rate of change, which has now fallen back to the lowest level since this is all during the tariffs. So for people who were looking for the tariffs to create the inflation, not there. We still have cuts built in. We're 90% for a cut in October at the end of the month. That will be during or just before the Apex summit. I'll get into that, I'm sure, later. Uh, and then we still have 70% chance for a cut in December. Again, economy is fine, inflation is fine, and yet the sentiment is still not there in terms of people wanting to be involved in the markets.
I'm going to continue to emphasize the commodity side. The XME put up a 2% week and I just want to make sure because I haven't shown this one before. A lot of this been precious metals. It's also been copper, but now you're really getting a widening out. But I just want to emphasize these are weekly charts. So, back to April. The XME has been up every single week except one in August. No other part of the market has done this. Part of that is gold. Yes. But like I said, part of this is the copex, the copper, uh, part which is in there. Uh, got to give a shout out to Colin Fenton who's been all over this. If you haven't been following his work, uh, this is something he put out on June, uh, 30th, which just emphasized and this is the, the part I really think is an important part. The copper cycle now underway began in February of 2024. We believe copper will most likely continue to strengthen for at least 3 years in what we expect more broadly to be one of the strongest environments for commodity markets in decades. He and I are on the same page of this, um, for different reasons, I think, and that's what makes it so interesting. Colin's a long-term supply and demand person, um, in commodities, and I'm doing this purely from the AI demand story and the buildout that has to happen in power. We got another signal that this week he published this in terms of the PGM West power price surge on tight capacity. Uh, aside from the price move, the move comes as governors from several PGM states, including Pennsylvania and Virginia, convene to voice concerns over rising retail power costs and tightness in the grid. Uh, the commodity side is not isolated to the metals. XLE posted its best month since June, continues to have a great looking chart. XOP, great looking chart as well. Uh, again, best week since June. So, you're getting the oil and production side joining and Halliburton, uh, broke through the 200-day moving average for the first time since April of last year. So, you're getting changes. And one of the things that's starting to become more evident, the Dallas Fed energy survey for Q3 was released, uh, and I'm not going to read all the quotes in here, but you can see that it's going to be really hard. The US shale business is broken. What was once the world's most dynamic energy engine has been gutted by political hostility and economic ignorance. The previous administration vilified the industry. The current administration is finishing the job. This whole thing and all these quotes are just about the fact that you can't invest, make long-term investments, and why would anyone make investments right now in energy if in three and a half years the policies completely change? So the uncertainty in the energy policy is going to hurt the supply side in the US and I think that's going to contribute to some of the commodity side. The Deutsche Bank, uh, the DBC, the Deutsche Bank commodity ETF chart looks like a reverse head and shoulders. It's up. I overlaid this over the last 15 years with the six-month contract of crude. Um, commodities have a bid and I think people should pay attention. This is the basing chart of the total return index of that DBC. Uh, and then here is the chart itself. So the commodity side looks good. Um, Variant Perception put out this thing on their advanced commodity side, six-month forward also looking higher. So I would focus again on the commodity side for next year. It's part of my PMI theme.
Jobs. So I'm going to go through this and try to go through it quickly. Uh, the month of September, there's been a lot of things on the job market. Uh, Zoom CEO came out, basically agreed with what Bill Gates, Jamie Diamond, Jensen Huang have said, which is a three-day work week is coming soon. This is really important, especially for macro people, to start to build into their expectations about what the AI situation is going to do to the jobs market. I've talked about this a lot. I've talked about it with the academic Fed. I cannot emphasize it enough that this is going to be a major theme for the next 5 years leading into humanoids. Bosch plans to cut 13,000 jobs. So, we got Germany in the play. Germany's not just there on that. You've got SAP saying AI will help us afford to have less people, but if we do it wrong, it could be a catastrophe. Tech sector layoffs accelerate. So, here we are. We have a boom going on in AI. You've seen all of the infrastructure, uh, GDP contribution from the technology sector, and yet last year we saw job losses of over 280,000, and this year we're already pointing to 235,000 by year-end, half a million in two years. Seasonal hiring for retail expected at lowest level since '09. The IBO is great for stocks, not for jobs. This is an article from Axios this, uh, this month. There's a real possibility we could kill millions of people, have could have millions of people put out of work by artificial intelligence. Upskilling is necessary. Neil Kashkari, one of the Fed presidents, uh, had this quote, "While it takes a lot of people to build a new data center, it takes relatively few to operate one." This is becoming a major story. And as a crowdsourcer, I just noticed that in the podcasts I listened to this week, inside all the articles that I sourced for in the labor market, there is real, uh, news happening. And I think this has to do with this, which is Darius Amodei, uh, the head of Anthropic, basically said the government needs to step in and support people as AI quickly displaces human work. You need some kind of policy response at the scale of disruption we expect in the next 5 years. Now, he has talked about this for the white-collar side, but it is very clear to me that with inside the tech world, the people in charge of these models are already seeing the rise of the white-collar disruption which is going to come from AI agents, and the stories that I'll keep going through here are represented. So again, if you're a macro person sitting there, policy mistake, blah, blah, blah, all this stuff that goes on, you have to start incorporating the fact that the labor market has broken away from GDP. That is a major story that has implications for rates. It has major implications for, uh, civil unrest and kind of the anger that's going on in the country. Fiverr, if you've never heard of Fiverr, think of Fiverr as a cheap outsource labor company. I've used them for for my business. Fiverr to cut about 250 jobs and refocusing effort, re-entered guidance. The whole thing they're talking about in this is related to AI by cutting 30% of their own internal workforce, restructuring. Fiverr is effectively saying even cheap outsource human work is now being replaced by AI. So, I wanted to go through this with ChatGPT to just say, what is this signal when we're getting to the point where a company that canvases the world to help you with cheap labor is now reaching a point where they have to focus on their job to make sure if they're going to survive that they're AI-focused. So, Anthropic's co-founders say the likelihood of AI replacing human jobs is so high that they needed to warn the world about it. Anthropic is the most important LLM for coding and API usage. They dominate. Okay. Anthropic employees and leadership strongly maintain that agent progress is accelerating and expect the pace to continue faster and faster. So again, the agentic side is the white-collar disruption. This is where computers will be doing and making more of the decisions. You won't need as many people. You'll have some, but this is when you really start to get the advancement on this. And this is what is happening now at a very fast pace. Anthropic predicts extremely capable AI systems and agent proliferation could be widespread as soon as 2026, 2027, potentially matching or exceeding the intellectual abilities of specialized human experts. I'll get more into that in terms of what's happening. In September, on September 9th, 40 economists sent a letter to the Secretary of Labor urging them to pay attention and monitor AI's impact. "We are a coalition of economists. While we may have differing views of the impacts AI will have, we are united in the belief that we should ensure American workers are not left behind as the rise of AI comes." On this list, Ben Bernanke and Janet Yellen. "We are concerned the government is not adequately prepared for this challenge and strongly want to emphasize the importance of high-quality data that is granular and timely," blah, blah, blah, blah, blah. The only thing that matters out of this is people are starting to realize more and more in conversations how fast this is accelerating. This is why I want to bring up this company MKER, which I'm sure very few people on this have heard. They're eyeing a $10 billion valuation. The person who runs the company is 22 years old. They already have an ARR of half a billion dollars. So what does this company do? This company helps bypass the friction of human adoption. And the way that they do that is that they bring in experts to help with the training of the models. So they bring in a lawyer, they bring in a banker, they bring in a healthcare person. And where the hallucinations and the things are because they don't have an LLM that can have all of the knowledge that a human being would have through the experience. This company pays over $200,000 to help get human expertise and experience into training fuel for its own obsolescence, in effect. And I asked the question, is this a new form of capitalism cannibalizing itself? So basically, this company pays people to come in and train the models in a reinforcement learning human feedback way, which has become part of the model training in general. But this is for specialization, and the reason that they have so much revenue is because the companies that are paying them are Meta, are OpenAI. It is the bigger companies on the hyperscaler side who want their models improved. This is one of the places these companies are going to get lots of revenue from selling this to law firms, from selling this to specialized. So, they're building specialized models, specialized LLMs specifically for certain industries, and they're using humans to help speed up the process of the reinforcement learning. Reinforcement learning was a big part of the deepseek side. All of this has become there because RLHF, using human feedback, is part of what is speeding this process. Also, Google researchers come out with a paper also in mid-September warning of a looming AI run economies. Basically, what this one is going through is that it's going to be an AI agent economy. The transactions will be happening AI to an agent, AI agent, which means less people involved in the process. It goes through this. I'm not going to read all of this. You guys can go read the paper on your own. The paper is careful not to claim all jobs disappear. It acknowledges limits, challenges, and transitional friction. But it does talk about the fact again that the increase or the correlation between the job market and GDP is what's breaking down. Now, the reason this is so important for everyone to focus on the global labor market is about $60 trillion annually. Global corporate profits are about 6 to 7 trillion. This is how important AI is and why this will not stop in terms of going through the wage bill, and this is where people believe the productivity is going to come from. It's going to come from cutting that $60 trillion. Theoretically, if you cut it to zero, the question is how high would global GDP be? If it stays at 105 trillion, then obviously you have a massive boom in corporate profits. Does that mean is that the productivity boom? So this is where I go through it. This is my angle on this. I've talked about the dangers involved in terms of the job market and why I don't, I, it's very hard for me to imagine comparing the industrial revolution productivity boom to what we're going to see now. And the reason is the replacement of jobs. So in the industrial revolution, even though it boosted human labor productivity, labor was still central. Factories needed workers to operate, manage, and use it. Scarcity preserved. Output per worker rose. Scarcity preserved is very important in this because this is my belief is that we start getting to scarcity. So in an abundant economy where machines are actually creating machines to replace themselves, people are not needed to the degree that they were. And I'm not in any way believing there will be no people working. I just believe that the ability to have the correlation between GDP growth and labor has completely broken down and will only worsen in the next 5 years, particularly as we get to robo taxis and humanoids. There is not a fast enough way for people to be reskilled because of how fast this is moving. As scarcity collapses, if energy, data, and raw materials are abundant, the marginal cost of goods and services trends towards zero, prices collapse, transitions vanish, GDP contracts. Even if material prosperity goes higher, isn't GDP to a great degree based on scarcity and inflation? GDP is built on a framework of scarcity and monetary exchange, not on abundance. This is the reason why I say GDP cannot measure it. If something is abundant, air, sunlight, it has little or no price, and GDP ignores it, even if it's crucial for life. GDP is isn't a measure of value and use, but a value of exchange. Labor costs are a huge component, over half of global GDP. As wages rise, so does GDP. If wages disappear, that feedback loop breaks down. That's the thing we're getting to is that you're disrupting this whole process and all the measurements, um, in 5 years. And I wrote this in here. I don't know if GDP will be around in 10 years. I'm sure the measurement will not be here. We're in the final stages of GDP being a useful measure. It will need to be completely rethought, uh, based on what's happening. So I put this Substack in where I go through the disruption that's happening and there's a rebirth that's coming out on the other side. It's going to be a new system, but people need to adjust.
So here we are on the, on the labor side. This is the year-over-year payroll numbers, which are now at 85 basis points. We've got 85 basis points of increased labor. We've never been at these levels without there being a recession. The unemployment rate. Marco Papage put this out, or reposted something, and this is basically showing again how we've got the recent graduates unemployment rate up to a very high level. The last time we were at this level, or every time we've been at this level, we've had a very similar thing on all workers. So the unemployment rate is hanging in there, but the pressure is growing. So this is a chart of the unemployment rate inverted, the purple line in here. You've got a whole bunch of different things. You've got temp jobs. This is the New York Fed expectation of or probability of getting a job within a year collapse, or this. And again, you've got all of these things happening. I've got the JOLTS quits in there. Uh, and I've got the jobs hard to get in the, uh, that's the red line here from the Conference Board, and this is tenure rates. Anyone still focusing on shorting bonds because of the deficit and debt, yelling policy mistake, thinking there's inflation, the gravity pull right now is on the labor market. You're going to need to see a dramatic shift in the labor market as AI accelerates, as agents are coming, as robo taxis are coming, as humanoids. This doesn't seem like a hard thing to figure out. Uh, these deflationary pressures should continue, and I think that's what's going to, uh, continue to put more pressure on governments to move rates lower because of the anger that's going to be happening from people that are having trouble in their jobs, that at a minimum are worried about their jobs. Uh, Variant Perception just had another one. I, I just bring this up here because of the rate situation. It's not just US rates. This is the, the business cycle financing index overlaid with with small cap. Again, I think commodities are going higher, PMIs are going higher, all on the back of the AI buildout, and that means that to me, small caps are going to end up being the surprise, uh, winner next year, just because a rising tide will lift all boats.
It was another week for the AI bubble theme. And this is where I want to make sure this is the best advice I can give you guys. If I do work, I definitely do work more than all the people writing these, all the people speaking about it. I will take you through the AI situation and the spending, but you have to connect all the dots. These articles are just random BS that comes out and people put up charts like this. Oh, look at all this spending. That means it's going to collapse. This is a once-in-a-lifetime situation of converting the entire world from non-AI to AI. That's what it is. Every single thing that you own in your home, every single thing that you're in in a car, every single thing in a phone will have artificial intelligence in there. To get there, we are not even close to the amount of supply that's necessary. Will there be bad investments? Yes, absolutely. If you ask me, I'm sure at least five of the Mag 7 will have a hard time justifying what they're spending money on, but that doesn't matter for making money in the next few years. Uh, playing short those companies is just going to kill you. So, you had the Nvidia investment for $100 billion that got everyone going. The incestuous behavior between these companies now, Sam Altman on worries about OpenAI, $850 billion. I totally get that. I'm not going to go through that guy's stuff. Um, again, the capex spend, $1 trillion of capex, $20 billion of AI is a bubble. David Einhorn sounds warning on AI spending splurge. And again, I agree with the fact that not all these companies are going to be able to justify it. Um, tech bubble on the GDP side because we got another quarter of just massive tech spend that's showing up in GDP. And Oracle. You can't see the Oracle thing in here, but they went out and borrowed money and everyone freaked out about that. And of course, to start the week, our usual clickbait. Let's put this out there. This one from Bain. An $800 billion revenue shortfall threatens AI future. AI companies will need $2 trillion in combined annual revenue to fund computing power by 2030, but their revenue is likely to fall $800 billion short. This might be the most ridiculous thing I've ever read. Uh, how anyone can predict exponential what's going to happen when everyone was wrong when the iPhone came out by a factor of, of infinity. Uh, and I'll go through this. I asked the question and went through this on based on the report that they put is safe to, uh, that Bain did not think about a broad basis for their analysis, meaning going through all of the potential implications, robo taxis, healthcare, everything of where revenues could come as opposed to just isolating it to the seven companies in terms of their spend. So if you're doing it versus the hyperscalers versus the revenue, again, I still think you're going to be wrong in there, but we'll go through this. The internet forecast '90s, they modeled revenues solely as advertising, smartphones focused on device sales, not apps, not this. You have no idea what part and what spending will be redirected towards AI. Jensen Huang on the interview said they're already, already at trillions of dollars purely based on the cloud revenues in terms of what's coming through. Now again, his argument is not about incremental, it's just about converting all the stuff they have. So we'll go through that as well when we get to the interview. But the main point is here, what, what you missed when you predicted on on the iPhone, e-commerce, social networks, cloud is, uh, all of these things came out of it. You couldn't have it without it. AI is moving much faster. In 5 years, you're talking about what's equivalent to 50 years of innovation, even faster than what happened with the smartphone. And again, when I go through and talk about robo taxis and humanoids alone, the revenues are extreme. In one of the Mag 7 and one of the hyperscalers now is Tesla in terms of spending money on building out their own AI. So smartphones in 2007, no one got anything related to it. How could they? You're going to have the same thing going on. You will have new categories to generate it. I just put in these about how much money and the savings that will come through per year if AI is just used in these forms. Uh, drug discovery, all of them. So the AI bubble thing is just ridiculous. That being said, this was good, a good article by Michael Semplist, and really it was just a thing. He called it the data center blob, just to talk about how big and massive this has been. And I think the most interesting part that he went into, first of all, AI direct impact to the S&P 500, basically 181%. 65 for these, the cap equipment, X AI, basically nothing. It's responsible for all the earnings growth. It's responsible for all the capex and, uh, R&D. The blob is overtaking markets. So his whole theory here is that just 40 stocks are driving the bulk of the market. Uh, and that's been what's been going on. And I think obviously that's going to change. That's why I said a tide that lifts all boats. It's been isolated to these groups. I think this is going to change. Borrowing is now expanding it. This gets into the Oracle thing. So he just highlights that this is a huge concentration, and he said he's never seen anything like this before. The positive side, the concentration means that when AI-driven names falter, the broader market could suffer disproportionately. This is what I want to get into because everyone has felt this. When I started doing stuff with 22V, I started having more and more conversations with portfolio managers, and the frustration they run into in every single vertical, whether it's utilities, whether it's the industrial side, or whether it's the AI side. Whenever something happens in Nvidia or in some name, the entire space falls, and people just have to get used to it. When you're that concentrated and that everything is driving the market, that's what happens. Since ChatGPT came out, Nvidia is up 12 times. 12 times. During that 12 times, there's been one, two, three, four retests, five, six, seven, seven, 20% corrections during the period. Five of those went before they got to new all-time highs. Two of them where they came down, retested, came back, and then went through this. You're going to have a lot. That's in three, less than three years. We had five 20-plus% corrections. We had three 25% corrections. People just have to get used to the fact that this is part of what goes on with AI because the majority of the index and the majority of the country is still in a recession, and that's where the majority of people work. So you're in this thing of, that's why it feels like a recession all the time because technically, using the old methodologies and not including AI, this is a recession now. I asked ChatGPT based on this, 40 names within, would it be difficult for port managers to risk manage during these times when 40 names are driving 70, 80% of earnings growth, they are the index. This creates two interlocking problems: correlation spikes and stress periods. I think everyone sees that that's what I was talking about. These things can be uncorrelated for periods of time, but when there's a correction and the market sells off, or AI sells off, they all go down together, and they have multiple standard deviation moves because if a utility stock is going up 30% a year, of course, it can have a 15% correction. That might be higher than historic, than history, but it can sell off. It can also go sideways for a year just because it's had a big move. We've seen that in some of the REITs this year that are still, nothing has changed on that front. Um, option markets and what I basically got into is puts become fairly useless in this point at the index level. Uh, but where I go through and ask the question, and this is the hedging strategy where I think people should start to pay attention to, which is vol, vol, they should be looking at VIX options, they should be timing these things. Uh, there's more details you guys can read on your own. And this is what has happened is when Nvidia has gone, has peaked relative to the S&P and gone down, and that's it. Just when it's peaked on a relative basis and gone down, we get these episodic spikes eventually after it goes on. We've had them in all of these places, um, where it's gone on. Well, right now we have another Nvidia peak. This happened literally at the ChatGPT 5 launch. There's no reason for it. Their earnings have been great. There's nothing going on. I did come out and say in this period in here that I thought we would have a period of Nvidia underperforming. I still think that's likely. Uh, Jensen Huang tried to change my mind on this, but I think the timing of this is just, uh, not good for them relative to other semis, which I've been talking about, and I'll go through. But we're in one of those situations. So the question is, are we going to get a spike? VIX calls are cheap right now relative to where they will be if we get a spike.
So here's the interview. Highly recommend you go spend it, not just because Jensen Huang is on, but because Brad Gerstner is so smart in this, that the questions are really, really good in the way he approaches it. So, just to go through it, clear the inference ship, which I've highlighted time and time again, with with reasoning, inference demand is expected to grow by orders of magnitude, 100 times, a thousand times, a billion times. He goes through the three scaling laws. We did pre-training, okay. Post-training, reinforcement learning. This is the thing that I mentioned with MKER. This has become one of the big additions. Then you've got the inference, reasoning, thinking before answering. This has been the one that has used up more capacity. Goes through the deal, talks about why he talks about all the multiple exponential demand, and then he goes through and highlights to you, general purpose computing is over. So everything that you're using today is a dead machine. The future which is coming, every single day. I ordered my first iPhone in the last four years because I decided that whatever limited AI was on the phone, I wanted to start to get used to it and figure it out. I did the same thing in my day with AI, and there's enough going on in it to pay attention. The hyperscalers shifting workloads from CPUs to GPUs. This alone equals hundreds of billions in infrastructure refresh, labor augmentation. Again, this is going through the argument of the revenues that are going to be associated. He's talking about how big the capex is going to be. So for people that think it's a capex bubble and it's going to come down, it isn't. Token generation is doubling every few months. Energy and compute are now direct revenue drivers. Energy and compute, it is the focal, focal point of everything that I talk about. It is the only thing I think from an investment standpoint that makes sense is to focus on energy investments and compute. All the other things, including shorting things, doesn't make sense to me other than maybe software, but in terms of like focusing your attention, you want to spend your time on energy and compute. The bubble debate, and he goes through it, and what he says is, until all classical computing shifts to accelerated AI, the chance of a glut is extremely low. This is really critical. Prisoner's dilemma dynamic. Every hyperscaler must build aggressively. No one can risk being left behind. I've talked about this for the hyperscalers. This is also true for China versus the US. They are spending, and I'm going to go through China's spending. So it's not just the US companies like Meta. They may overspend by $10 billion, but the risk is existential, not optional. That's why they will continue to spend. Their boards cannot convince them otherwise. Accelerated computing plus AI, not just GPUs. When he talks about accelerated, he means the broad shift away from general purpose, again, to move away from computers. Now, what I wanted to go through, and this is where I start the question for everyone here is, what does the world look like the next phase? So as the world moves to AI embodiment, they're critical for edge, and NPUs fit into the equation. I've talked about NPUs within the interview with Jensen Huang. Did he acknowledge that they continue to make efficiency gains? Yes. He said massively, they have efficiency gains with every chip generation. So the next time you hear someone say, "Oh, are there going to be efficiency gains?" Every single chip that they release is focused on a metric in tokens per watt. They are absolutely, every single one of the new architectures delivers massive jumps in performance per watt. So the next time you hear, "Are they going to go through this?" It already is happening. It's been happening every single day. So it is ridiculous at this point to sit there and worry about even that the demand is just higher than supply. And here's the thing. If AI demand was not moving so fast, would this allow other chip companies to catch up? And the answer is yes. You can go through. They'd have more breathing room. If it wasn't moving so fast, and Nvidia would go to a one-year release cycle. So everything would change, except for the fact that AI is going so fast. With demand exploding, hyperscalers need the full stack solution now, not a partial answer late. If someone tapes out an ASIC today, they're solving last year's problem. By the time they deliver, the industry has moved on. This is the reason why Nvidia has a moat. It's moving so fast. They built the scale for this. Everyone's now trying to compete with them. But they have a new chip constantly coming out with better efficiency. So why would you buy the older stuff? You can't do it. And you certainly can't take the time to try something if it's not going to work. Your time is over. This is why Nvidia's had the moat. Their AI server orders are so big, they're unimaginable, says a separate computer place.
So, this is where I wanted to go, and I, I wanted to get me to a point where ChatGPT answered the question using both what Jensen said and what's out there. So, if there's still a restriction on V3, which is the video usage from, from, uh, from Google Gemini, there can't be a glut yet. V3. And yes, you've connected the dots. Well, let's frame it clearly. You go through it. Video is too compute-intensive. You still can't run it at scale. You can only run it in limited access. It is rationed out. Demand is greater than supply because they won't allow you to use it. No glut until all content generation is AI-driven. This is what Jensen said. Glut air pockets. There's going to be. Then it asked me, and this is now what it's happening, uh, to build a checklist of signals. So here we go. These are the things to look for for when supply is starting to get ahead of demand on just any one of these six. None of these are happening, and I'm not going to go read them all. Um, H high bandwidth memory supply tightness. I'll get into that because this is there. We got power constraints, the secondary market. These are all the checklists of what would be there. So in glut signals, you know, when V3 is offered more broadly, uh, capex growth is flattening, you start seeing them cancel some of this stuff. Right now, all we see are shortage signals. A true glut cycle would require all three layers to saturate. So we're not there. So every time that you get this bubble talk that happens, the good news is we're going to get a correction. When you get the correction, go to technicals or have some VIX call options on, uh, beforehand, and then put yourself in a position where you're still focused. I do believe there's going to be a rotation, and I, you know, I've talked about that. I'll get into that in some of the semis. Now, if you don't believe, if you think the US is in a bubble, then why is China doing all this stuff? So, Alibaba stock surges as Eddie Wu announces $53 billion AI spending plan. Uh, so they increased their spending and saying the speed of AI industry growth and suggested global investment could reach $4 trillion in five years. He stated Alibaba will increase its data center capacity by 10 times between now and the end of the decade. This is actually more than the four to five times that Jensen said in there, and they talked about this on there. If Alibaba alone is going 10 times in power, then Nvidia's four to five estimate might actually be conservative. Nvidia's revenue is directly correlated to power. It's correlated to watts. It's another demand signal that there's no glut. And it's not just them. So, Bank of America went through it, put the number that the Chinese hyperscalers are now at for this year at 85 to 98. Now, this 85 to 98 is different than the US one. It doesn't include the grid capacity or the energy side, which a lot of the numbers for the capex for some of the hyperscalers, that also they're spending tons of money right now on the energy side. Private sector in the US or in China, they're spending tons of money on this. And to give you an idea, in 2024, China spent over $625 billion on clean energy alone. This one put the number at clean energy investment at $940 billion, 10 record, 10% of GDP in China. China observers are shocked at how US grid limits and lack of coordinated investment are slowing AI demand center buildout. Contrasting with China's rapid demand, there's a bottleneck. They talk about the fact that demand is a key reason it expects to leapfrog new data center demand in the US to leapfrog the US in AI infrastructure. This could hurt in being competitive at all. It looks like a competitive red flag. This is one of the reasons why I've talked about you're going to have for the first time for the Mag 7 global competition. It's certainly going to be in Asia. It's certainly going to be partly in the Middle East, but they are doing a lot. And what ByteDance came out and basically said, the rapid growth in token consumption from 50 trillion in 2025 to 10 times the number in 2026. So again, you just had 10 times for Google. You're talking about everything around the globe accelerating. Huawei unveils their own data center side to compete with Nvidia. So, another sign of there being no glut, the semiconductors.
So, I've shown this chart before. This is the one of the DRAMs overlaid with the PMIs. It has had another sharp move higher. Uh, so you've got DRAM prices going. But just so you guys can see this one, this is September alone. So what I did was take that chart, paste it along with the description into ChatGPT. Please describe to me what this chart shows with DRAM prices and what company should benefit from it. This is how you make money with inside the market. You don't need to pay an analyst. You don't need sell-side research and go through all of it. You get a chart like this, you put it in. There's a sudden severe shortage or demand shortage in DRAM memory chips. DDR5 is the backbone of the AI servers, tight supply, blah, blah, blah. The companies likely to benefit. Well, Samsung, SK Hynix, Micron makes perfect sense. They have an oligopoly. We've talked about it. Micron reported, I'll go through that in a second. Uh, but here are the companies on the second-order beneficiaries. ASML is a company that I mentioned last week. You've got TDK, Electron. You start building out and looking at what this shows and what this means, and you keep going. So Teradyne is a name that came to my attention this week. So as I was doing this, I said, what about Teradyne? Do they benefit from this move? Yes, Teradyne benefits indirectly from DRAM price surge, though with a lag. So near-term, zero to two quarters, limited impact. Medium-term, two to six, strong memory test demand, higher orders for Teradyne. So you start to build as to when they would start to benefit. So it asked me if I wanted a list. Said, "Yeah, give me a list of the hierarchy and go through the companies at each stage." Okay, so this is the one that's already worked. You can go through these. Micron's up from $60 to $170 as of, uh, last week before earnings. Uh, so what else? What's the next line? Well, here are the next ones in terms of the upstream materials. There's Teradyne sitting at number three in here. You can look at the other names in here. You can do your own homework. Given Micron has already said there are bottlenecks in supply, does this move the stages up? So what I basically asked the next question was, since there's already bottlenecks and since Micron is saying that they've got bottlenecks and delays, what does this mean for Teradyne? What does this mean for all of that supply chain? It pulls everything up, pulls their their benefit forward by one to two quarters. So you should start seeing the demand. So in terms of this, it's moved it up. So that means that these stocks should be seeing more demand. At the same time, it's bad for companies like Nvidia that need it. If there's a bottleneck, they're not able to get their chips done and they'll have delays. So if the supply chain is not there, you want to go with the names that are going to see demand coming from nothing right now. And that's what's happening. A lot of these names like Teradyne, uh, like, uh, ASML, they're kind of unchanged over the last four years. And the reason is because the phones, because the computers, because the autos, because of home, the, the sensors in, in washing machines, there's been no demand for this stuff. So now the AI demand is starting to impact them because we're getting into the next stage of this. And it's not just about what Micron said. So what did Micron say? "Demand has driven a significant surge in high-value data center products, especially bandwidth memory. It's accelerating an industry demand. Supply is tight." Attributing the improved forecast directly to AI agent workloads. Again, the job situation and traditional tasks initiated by those agents. The agents are going to accelerate the demand for everything. And what did they say about AI demand for phones and computers? Again, an upgrade cycle is massive for the market. Apple just made new all-time highs. You're starting to see as they release their phone that people are starting to get a little bit more sense that we're getting closer and closer to AI-enabled smartphones and AI PCs. He specifically discussed adoption of it as a central growth driver for Micron memory. He cited greater adoption of AI PCs and the end of life for Windows 10, improving the demand outlook. All smartphones, as described by Micron, are super companions delivering new multimodal experiences with advanced memory. An improved PC demand outlook is directly tied to increasing sales. In short, Micron management sees AI adoption in phones and PCs fueling a major upgrade cycle, bringing multi-year tailwinds to both DRAM and NAND flash. Again, you have a situation that these areas have been in a bear market. So, go find the companies that will benefit as AI gets integrated into them, which is what the podcast told you. Then you go through your work in AI. This is how you start from a very high-level systems thinking approach. Listen to everything that Jensen Huang said, have AI go through the transcript and then pick out the parts where there's opportunities, and then you go through. And if you do this every single day and you go through it, you will start to be using AI in a way that is very easy to do. And by the way, this has been Gemini. This has been ChatGPT. This is Perplexity. Give me a summary of what Susquehanna said in their upgrade this week. So Susquehanna upgraded Teradyne. That's how I found Teradyne is because they were up over 10% on Monday when the market was not that strong. So I went through and said, "How can Nvidia be down but this be up? What's going on? And why would the stock be up 10%." So they raised their price target from 133 to 200. First of all, it's a big raise. The the main thing here is that I see the reaction of the stock as something very, very important. So as I go through, analysts pointed out strong exposure to auto, PC, and smartphone markets where AI and assisted driving and NPU proliferation are accelerating test demand for new chip families. This is a major thing because this is the, the PC. This is the smartphone. The auto is related to robo taxis. Behind that is going to be the brain for humanoids. The note called attention to Teradyne's leverage to rising AI hardware demand across autos, PCs, and smartphones sectors that are driving a fresh surge in semiconductor test requirements, especially for advanced. Again, this is how you go find things. All right. Uh, in their most recent earnings commentary, what did they say on NPUs and AI chips? "We're seeing a step function increase. Mega trends in automotive, especially assisted driving and infotainment, are driving strong demand. As more products integrate AI features, NPUs and neural inferencing, the brain inside the machine, test complexity and market demand increase." These things are not used yet, guys. This is a new demand side. They're saying it is happening now. I can't, I can't say that loud enough. This is a new thing. So, here's
The stock. Pterodine was up big on Monday. It continued higher the rest of the week. Again, since March of 2021, it is unchanged. This chart looks like the PMI looks like energy. Again, hardware, hardware, hardware. Uh, Trump moves toward a deal for an equity stake in a lithium mine.
Again, I'm looking for things that suggest that there is zero chance that the AI bubble is in place. One is all the comments we made that the hyperscalers are in a prisoner's dilemma. They don't have a choice. They'll get their revenues one way or the other. Now, the government has made three investments. One in MP Materials, a rare earth company, AI. Now, it's lithium batteries AI. Intel compute AI.
Trump signs executive order supporting Pros deal to put TikTok under US ownership. Uh, again, this is another signal of something I said last week, which I expect. All signals point to the fact that we cannot divorce China. We cannot decouple from China. We depend on the rare earths. We have no choice. So anything where we can find deals with China means the probability of a grand bargain at the Apex Summit increases. If that happens, I went through the beneficiaries last weekend. You can go see them again.
Tesla. All right, here we go. Another positive week for Tesla despite the market being down, up three and a half percent. Uh, I'm going to go through why this is so important. I'm going to take it in more details because we're getting closer and closer to an event. This is Jason from the All-In podcast. My last three Tesla Full Self-Driving rides were absolutely perfect. Alan Musk responds. They're buddies, so not surprising. Version 14 goes into early wide release next week. Then 14.1 about two weeks later, and finally 14.2. The car will almost feel like it is a sentient being. 14.2. So again, I've said this before, check your biases at the door, and that includes valuation biases. That includes anything you have with Elon Musk.
Tesla is a major story because it is the gateway to humanoids. Humanoids, as I showed you, $50 trillion to $60 trillion of spending is on labor. Anything that speeds humanoids up, and you start to get a valuation shift in Tesla. If you're shorted or underweighted, do it at your own risk. This FSD version 14 is important. Please go through the importance of FSD version 14 if it enables the safety driver to be taken out of the car. So people can sit there and say, he's still a safety driver. Okay, great. You're in a probability distribution. If the safety driver gets taken out, that's the holy grail question. If FSD version 14 or its successors enable safe removal of the safety driver, it would radically change Tesla's valuation, operating model, and the broader mobility market.
So let's just go through this. The economic impact, robo-taxi revenue model. Instead of selling a Model 3 for $40, $50,000, blah, blah, blah, you get, everyone has probably read this at this point. You get a margins jump because utilization goes up dramatically. Software monetization, hardware sales, you get all of a sudden a business that completely changes and starts to have somewhat of a similar thing to Apple. I'll go through that as well.
So, you're you're looking to see now for humanoid valuation. Given the FSD 14 importance of computer vision, if the safety driver is taken out, what does that mean for the possibility of humanoids? I've said this before, they are going with computer vision. Waymo went with geo-fencing and lidar. Not only a difference in cost, but again, you drop a Waymo on Mars, it has no idea what to do because it doesn't have GPS. If you drop a computer vision, anything up there, humanoid car to work on Mars, that's the only way it's going to work. That's why this is so important. It will immediately bring the valuation of humanoids into the equation. Even if the scaling of it is not there, because computer vision is one of the three challenges along with scaling and the hand that Musk has mentioned as recently as two weeks ago.
FSD's advances in vision-based autonomy are directly transferable to humanoids navigating factories, warehouses, and homes. If Tesla can prove cars can drive without humans, cars are just humanoids on wheels, it signals their vision stack is robust enough for real-world uncertainty, a core requirement for humanoids. So, the perception alone could create a valuation flip way before humanoids generate revenues. That is the argument that I'm making is that if you're going to be short this, pay attention to what he has been signaling. I didn't write a paper on this purely based on some speculation. It was crowdsourcing of information. So much of it, including the stuff on version 14, including the deal done with Samsung to have his new chips brought in, including the changes in Dojo, including the changes on getting rid of the virtual reality side for training Optimus.
So the valuation impact if Tesla shifts from an $800 billion automaker, it's already obviously over a trillion dollars, to a multi-trillion dollar AI mobility platform. Why the capitalization of labor will lead to trillions in value? Again, here's the annual spend that can be expected or that is there on mobility. This is the humanoids in terms of the TAM. You can go through the TAM for autonomy and humanoids on your own and figure out what you think it'll be. Investors are already speculating on this. They've been speculating on it. The question is how valuable is that? These are just different things that were put in the practical timeline. 25, 26 real driver out pilots. If they believe it's going to happen in '26, they're going to front-load the valuation.
So again, I'm doing this more for people who are negative on it and for people who don't have anything on it. For traders that are looking to trade the thing, trade it technically. The charts look great at this point. They don't look like I mean, they're still looking good until we get up to the all-time highs. Uh, we're approaching it soon. Um, since none of this vision is possible without AI, is Tesla an AI company or a car company? I do that because people keep giving me valuations to it. That's not the way to value it. How do you value OpenAI uh the way it is? And I would say that Tesla has a far bigger advantage than OpenAI in terms of revenues and the ability of making the money. And this is where you get through. If you're talking about revenues for the hyperscalers, is Tesla at this point a hyperscaler when XAI is building out Colossus and data centers? And I think the answer is yes. Uh, even though it's not the same company, it doesn't really matter. Are they going to get the revenues? And obviously there's a linkage there. The car is a smartphone on wheels.
So I just wanted to go through this has been a discussion that has happened and again this is another way the smartphone replaced the PC as the primary computer node. An autonomous AI-native car could become the next primary compute environment after the smartphone, especially given one to two hours a day driving. Again, you have to think of this differently. And remember, he's got the hardware, the car, he's got the brain, Grok, the network, Starlink, he just paid for Echoar, the app layer. Just don't fade it. Um, I'm back to this one again. You have to think outside the box.
All right. Bloodbath in crypto this week. Um, so the panic set in not just Bitcoin breaking down below 110, but you have all of these others breaking. Ethereum broke through 4,000. It had been a long road to go from 2500 to 5,000. Now we're back this way. So, a bunch of longs were wiped out. Massive liquidations.
Um, here's what I just want to do again because to me the linkage between AI and Bitcoin is is inexplicably uh it cannot be separated. So I wrote that Substack which goes through why I think AI and creative destruction is so important and such a driving force with Bitcoin. We're at the stage now where AI is accelerating to the point where people are going to start losing their jobs. That is exactly the time that the anger starts and where governments are forced to do things like cutting rates. Even if inflation is higher, as gas at the pump goes higher, even if GDP is at 4%. They have to focus on the labor side. They're going to put their own Fed chair in. All of these things are positive for Bitcoin. But the pressure continues to accelerate as AI accelerates.
So there's a direct linkage between Bitcoin and Nvidia. And you can see it here. Nvidia. This is the absolute chart, not the regular chart. Whenever Bitcoin or whenever Nvidia is weak, then Bitcoin is weak. Here's the relative trade that's been going on this entire year. They peaked right around the exact same day and then we went down. So again, Nvidia peaks right here on August 8th. Bitcoin peaks right here. They're moving together.
Um, every time we've had a correction over since ChatGPT was launched, the white line here is Nvidia versus the S&P. So consolidation here, Bitcoin consolidates. We get this little point here. We get weakness. We get a rally. Then we get another consolidation here. Consolidation here. Then we get this point after a rally, another consolidation. Now we're going through another one. It goes in stair steps. So Nvidia over the S&P is the way to look at this stuff. And again, I reiterate, Nvidia has had plenty of corrections. Bitcoin will always have plenty of corrections as well. But in the end, here's the differential.
Um, this is from a uh, sorry, November 2nd. So this is about three weeks before the release of ChatGPT. 13 times for Nvidia, five times for Bitcoin. The Mag 7, four times. Gold only 200% and the S&P 500 170.
That's it for this week. Uh, I appreciate you guys checking in all the subscriber stuff. Um, I like seeing a bunch of you uh out on the street trying to help you guys make money. I cannot emphasize it enough. The way I started this. Uh, AI will replace people, but AI is really going to replace people who don't know how to use AI. Keep using it. Keep watching what I'm doing. Try to replicate it. I'll probably be promoting some videos or something to help people more and more. Uh, it's a kind of feeling of being powerless right now with AI coming when the reality is people should be feeling empowered because it allows you to do that. I'll see you next week.