Micron Sparks an AI Rally and Why This Bubble Could Burst | AlphaTarget PULSE

We still think that after a customary summer early autumn pullback, we are likely to see a strong rally towards year end and the beginning of next year, but one day at a time. What we are now seeing is we're seeing comments and reports increasingly from companies saying that they need to curb the spend because costing too much. Apple did not build its own large language model. What they did is they did a tie up with Google and basically are using the Gemini model as a baseline to distill and build their own model. So Nvidia did a licensing agreement with them not too long and at the last Nvidia GTC Jensen said he had $1 trillion in demand until the end of 2027 for his regular GPUs and he expected that Grok the Grok product would would add an additional 25% to that. Markets Business Technology Investing Insights Alpha Target Pulse Alpha Target Pulse is a production of Alpha Target Ltd. A company registered in Jersey Channel Islands. The opinions expressed in this podcast and the information provided are for informational and educational purposes only and are not to be construed as investment advice. You should consult with your own registered investment advisor or other qualified financial professional before making any investment decisions. Welcome back to Alpha Target Pulse, your podcast for insights on markets, business, technology and investing. In today's pod we will cover what's going on in the stock market, the big rally we've seen in the AI bottleneck stocks, explosive moods in some of the other AI stocks, the on and off deal with Iran and the US and the recent earnings yesterday which were released by Micron which caused a big rally in the stock and reversed the stock market decline with bold claims made by the CEO. So we will discuss all of that a bit later on. But first can I please request you if you watch this pod on YouTube, can you please subscribe to our channel? It would really help us to improve the quality of this show and also increase the discovery of our podcast on YouTube. Thank you so much. So before we proceed, let's now bring in and introduce our analysts who will join us today in our discussion. First and foremost we have Ernest, our numbersman, CFA Charter holder, top ranked analyst in South Africa and former analyst at Society General. Next up we have anurban, Mr. Bleeding edge as he is known around at our firm, Ph.D. in computer science and former Director of Research and Portfolio Manager at the Motley Fool Australia. Next we have a big picture guy who previously worked as an analyst at Peel Hunt Berenberg and RCO Family office as an investor. So let's kick off today's pod by just going through what's happening in the stock market. And just to give you guys an overview of what we think is happening and what the outlook is for the next few weeks. Well, first and foremost we had the news from the White House that a deal has now been secured. Basically it is only a matter of time before the document is signed, sealed and delivered. We've had some ups and downs in relation to that with talks failing and then restarting and failing. So nobody really knows when the end document will be signed. But the good news is the market seems to believe it this time. We've seen a big decline in the price of oil. Brent and IMAX crude are down significantly, which is a major plus. Bond yields have come down a little bit, which is also another plus. We have seen some sort of stability in the stock market of late in certain segments, which is also a plus. On the tech side we have seen some very strong earnings announced by Micron yesterday with the CEO claiming that this is a multi decade cycle because of robotics, which will basically require 10x the amount of memory that AI further down the track. But I think you have to take these announcements with a grain of salt because you know the job of every CEO is to pump up their stock and if they are not going to sell their business then you have to ask who will. So you know, we've seen this over and over again. Yes, robotics will require a lot more memory and so will autonomous driving and so forth, but robotics are not going to happen here and now they are maybe a five or seven year story or longer. But what we are watching closely is basically the earnings cycle from the AI data sector ramp up, which according to some estimates will peak in the second half of 2027. And historically these stocks have actually peaked several months before the deceleration has begun. So it'll be interesting to see how it all plays out. We still think that after know customary summer, early autumn pullback, we are likely to see a strong rally towards year end and the beginning of next year, but one day at a time. We're not in the business of forecasting the market. We're in the business of analyzing companies or business trends and then aligning ourselves with those trends to make sure that we make money on the way up and protect our gains on the way down. In terms of macro, we had the Fed, the Fed basically said that they're not going to raise rates anytime soon. But there are basically one rate hike is on the table later this year, maybe two depending on how the economy shakes up and whether inflation goes up or down. We personally felt that the new Fed chair, he basically handled the press conference reasonably well, and there is no surprise there. The Fed has been on wait and see mode for a while and we think that the Fed is not going to try and become too aggressive, especially if the war definitely has ended, which means the price of crude will come on. And the Fed is aware that with the oil prices coming down, inflationary pressures will subside. So our best guess is that the Fed will probably stay on hold for a while and then we'll see how things shape up. So now that we've covered the stock market, I'd just like to mention briefly that we've seen a big sell off in the beginning of the year in some of the software companies, infrastructure software companies, cybersecurity, the good ones have already rallied a lot. Some of the other ones are still, we argue, on the bargain table. They will also go up and the market realizes that enterprise software, you know, infrastructure software at the enterprise level, and companies which are actually integrating AI agents in their stack, they are going to basically benefit from this. They are not going to become roadkill of AI. So we think there is still some opportunity there. The semiconductor stocks have had a huge run historically. They are extremely, extremely overextended. They have gone parabolic. Some of the stocks are up 6, 7, 8x. Some stocks have tripled, quadrupled over the last three, four months or so. So this is not the time to be chasing them, in our opinion. We've actually locked in some gains over the last couple of weeks in some of these overextended names. And we are waiting patiently for a re entry point later this summer or early autumn, which we think will happen because history shows that when something goes up, straight up, it doesn't stay up. The market goes up and the market goes down. INHALES AND EXHALES and this is what we think is going to happen here also. So we are basically at the moment around 35% in cash, 65% invested. We are not going, you know, pedal, you know, pedal to the floor right now because we still think that there might be some chop over the next sort of four to six weeks. So that's the stock market prognosis. No major gloom and doom on the horizon as far as we can see. So now, with all this out of the way, let me bring in Ernest, who's been listening patiently. Honest. I've got a question for you. You've got some views on the use of AI within Enterprises we've seen some token maxing transitioning into token budgeting. Can you just explain to our viewers and listeners what this means and what is really going on here? Thanks Puru. I mean look, this is really a million dollar question because what we've seen going on the last couple of months, you know, when agents were, were, were becoming popular is companies really try to apply themselves and, and, and use the AI as much as possible. And you can expect that in this excitement and, and bit of euphoria. You know, you've got developers just literally using these agents in, in multiple sessions in parallel and, and applying it to whatever they can to see what they can do with it. And, and it's only to be expected when, when something's new that people try their best to see where can it be applied. It's, it's a kind of exploratory phase and I think that's normal. But what we are now seeing is we're seeing comments and reports increasingly from companies saying that they need to curb the spend because costing too much. So we're seeing, you know, companies all over reducing their spend and saying to their staff you've got to go a bit easier on this thing because it's, it's costing us too much. Interestingly, anecdotally I have a few CROs in my network globally who I've been in touch with the last month and they telling me that they finding it quite challenging to deploy AI in their enterprise at scale and do it securely. And I think the, the point I wanted to make here is with chat GPT and you know, these, these, these models that we run in our individual capacities, it's quite easy. You fire them up, you pay them your $20 or 100 or 200 and you do what you need to do and that's fine. Good. And well you give it access to your computer if it's an agent sort of thing and that's okay. But when you're working in an enterprise and you want to, you know, give it access to other systems that affect thousands of people in the organization or you want to collaborate with your team members, it's not so easy to do this securely because now your company's got governance protocols which are in place and who gives the agent right a right to do what it needs to do. So what we are seeing is really I think the beginning phase of enterprises sort of trying to figure out how they going to make the most use of, of all this technology. And it's very exciting, you know, because they need to they need to get it working and they need to make sure that it's actually going to add value and that it's all running securely. So it's not really clear whether they're going to get it right now or it's going to take a bit of time. But, but this is certainly one thing we're watching. This is why databricks as well is doing security for agents because securing the agents has typically is still an unsolved problem. So I think you know, they'll eventually get solutions but I don't think it's solved at the moment. Thanks for that Ernest. I'll move on to you Anirban next. Apple made an interesting announcement recently regarding AI on the phone with audio information as context. What are your thoughts on this and will the phones be able to run a comprehensive enough model locally? Will they need to call the big models too? How does this link in with their tie up to use Gemini? Yeah, thanks Farooq. So I think Apple is doing something interesting here. So Apple did not build its own model. So backtrack a bit. So Apple did not build its own large language model. What they did is they did a tie up with Google and basically are using the Gemini model as a baseline to distill and build their own models. So they're building models on top of of Gemini so that when Apple has their own foundation large language model, it's basically built on top of Gemini. It's still their own model but it's distilled from Gemini. They've also what they've done is they've distilled small models. So basically models that can run on the phone, several billion parameter models for example that sits on your phone. But I think the interesting thing that Apple is doing is it is providing at least currently the native app. So this is all in the developer beta. The native Apple apps can provide context to the AI so it can provide context to Siri. So effectively it can read or make an index or semantic index of your messages, of emails, of, of your photos, right? And once it builds a semantic index and that semantic index actually sits on your phone, it can become a very personal assistant, right? So for example I can query saying hey can you find me images of say my daughter's birthday celebration from XYZ day to doing something and it will be probably able to do that or it might be something like it might see that while you have a calendar engagement today at some place and it could for example proactively recommend based on the weather, hey bring an umbrella or, you know, bring sunnies or whatever. Right. So it can do things based on the personal context that I think your other agents don't have. So that's, I think a very useful way in which you can enhance how the AI actually interacts and does things for you. Right. Then the other interesting aspect is basically orchestration. So it's trying to, when you give a query, a lot of the queries can be handled on device. Like if you want somebody to, for example, just proofread something that can be probably done on device. But if you have a complex query that requires it to go basically be given to this foundation model running on the cloud, then it'll send that request to the cloud and deliver the answer back. So I think this orchestration and personalization I think adds a layer that I think the. So the large language models that don't have context can't really do, which I think is an interesting thing. And I think this goes back to sort of this even in the enterprise context. It's the same thing where if enterprises have certain context or certain data that's only available to them, which is not really available to the large language model, then the large language model is essentially a genetic repository of information. You can then use this specific semantic information to do a lot more. And I think this is where I think a lot of the evolution should likely happen is that you'd have orchestration around whether you're going to use on device or locally or to the cloud, what type of model you want to use. Do you want to use the latest frontier model or do you want to use an open source model? Because for some model, some work, an open source model might just be very good. You don't need a frontier model for something that's not very complex. So I think this orchestration would, I think we'd see more of this in enterprise, just like what Apple is building. Apple is basically turning Siri into an orchestration layer. And I would think that in the future you could allow the Siri layer to basically use any agent. It just is providing the foundational context information and then you can turn it into, well, I want to use some other agent, whether be it Grok or Gemini or whatever else you remember probably 10, 20 years ago, the big thing was bring your own device. Where workers would bring their own smartphones into the workplace and then you know, they didn't want to get a company issued phone or device and then the company would like put an application on the phone which sort of guardrails certain things. They're allowed to do on the, on the corporate network when they go home, then they don't access it in that way or be it as it may. Now what I would like to know is with Apple, with their new personalized agent that you've just described, how do you think that will interact in a corporate sense? Because at the end of the day, you know, we have to ask ourselves what's the perfect device for a person? And I mean it looks like it's going to be this phone, you know, your iPhone. I would imagine that if you are running, doing work, then there's going to be some other application that's going to be sitting there that's going to be probably acting as a filtering agent or like another firewall sort of thing, you know, that prevents I guess leakage of enterprise information or sensitive information to elsewhere. I mean most of Apple's things are actually happening on device. It's not really leaking anywhere. And when they're sending a query to the cloud, that query gets deleted and they provided or the information gets deleted, whichever information is being acted on. So they're providing guarantees that researchers can actually validate that they're actually doing that. So they're providing the guarantees for privacy here. There's no data logging that's happening to facilitate it. But I would imagine if this is in the enterprise context, there's going to be another layer trying to block things or guardrail things and so on. Right, I'll come back to you again changing the topic this time to memory. We've recently come across something quite interesting. Bernstein thinks the memory players EPS will come down significantly in 2028. They are forecasting that EPS for this cycle. Earnings for the memory stocks will peak in the second half of 2027, which if true has some serious ramifications for the stocks. What did you think of this and what are your findings on this one? Yeah, I mean, thanks Puru. I mean it's a, it's such an important part of the market right now this memory trade. And we got these big three players which pretty much do most of the high bandwidth memory. And it's really a story about cyclicality. So big shortage, memory price spikes up, then you got margins lift, you've got huge lift in profits and, and, and all that leverage and then it takes a long time to build new capacity because these are big step functions in terms of fabs. New fabs takes a few years and then eventually you get supply that matches demand and the whole thing reverses. So prices come down, operating leverage in reverse etc. I don't know why they chart there. I don't know why it says, you know, gradually declines in 28. I mean if you look at these charts it's kind of falling quite a lot. It's like more than halving but nevertheless. So I think it's going to be important and the key question Is it 2028 or is this going to last a bit longer? Could it be another year or two after? And we actually don't know because we, you know I was chatting to Oyvind the other day, we don't know how much demand there's actually going to be. In fact the companies don't even know the big models. They looking a few months ahead. They don't know what's going to happen in two years time. So it's very exciting and we all watching and tracking it. Yeah, I agree. I think you know there's memory trade is pretty mature now. Multiples still appear cheap. But the one thing I do point out is you know memory is a cyclical industry and we the reason why the multiples are cheap is because the market knows, or a large chunk of the market knows that the earnings are going to peak and then they will decline significantly year over year. So when the comps come down, the earnings come down and the know the PE automatically inflates and that's why cyclicals usually trade at what appear to be cheap multiples near peak cycle earnings. So that's our two cents. I'll move over to you next. Ain you've been looking at the premium inference market and writing a report on it for us which we will publish on our website in the next few days. What should investors be paying attention to in this space? I know you've done a lot of work in this area, the CPUs, the infancy and so forth. If you can just give us your high level thoughts on this please. Yeah, no, it's a very interesting market. I think it become quite large. So one of the key players in this space is Grok, the one spelled with the Q not the Elon Musk K1. But yeah. So Nvidia did a licensing agreement with them not too long and at the last Nvidia GTC Jensen said he had $1 trillion in demand until the end of 2027 for his regular GPUs and he expected that the Groq product would add an additional 25% to that. So that's 250 billion. So a huge number, so a huge potential market and we've Also seen an IPO in this space with Cerebras which raised over $6 billion, which I think they said was the largest semiconductor IPO in history. And there's also a bunch of other private companies in the space too, so we might see more of these IPOs in the not too distant future. But yeah, the reason this space matters is because these new processors are significantly improving inference speeds, with Cerebras claiming the improvement is up to 15 times. So this will not only boost productivity for humans and agents using AI, but it'll also improve the experience of things like voice and reduce dead time when you're waiting for a prompt to end and you get distracted and you kind of lose that flow state. And Cerebras founder has a quote that I like which is what's the market size of slow search? Which is quite powerful I think, because it kind of implies everyone will eventually want super fast inference. So yeah, it's a quite exciting market I think. Interesting. And what is enabling these companies to deliver much faster inference? What is the secret source here? Yeah, so they all have a bit different architectures and approaches, but one thing most of them have in common is that they're using this memory called sram. So SRAM has much more bandwidth than the high bandwidth memory or the HBM memory used in GPUs currently. And memory is a key bottleneck when in the decode phase of inference, which is essentially the token generation stage to produce the output. Also, SRAM and compute, they sit very close together as SRAM sits on the chip, whereas HBM sits off the chip. So this means the data from SRAM doesn't have to travel so far, which reduces latency and power consumption. So historically SRAM wasn't really considered to be viable as the primary memory because it didn't have enough capacity like the size in terms of gigabytes. But these companies have built new architectures around that constraint. But yeah, it's important to note though that GROK and Cerebras aren't outright replacements for GPUs as GPUs are more multi purpose. Rather they are providing specialized hardware that is very good at a specific portion of the inference and they can actually work together like they work together with the traditional GPUs. I mean, clearly it's a very interesting market at the beginning of the S curve adoption. We've discussed this before. The S curve adoption is the hockey stick adoption curve where the demand suddenly grows exponentially and takes off for a number of years. We've seen that across technological shifts over the last 20, 25 years, we've seen that with the Internet, we've seen usage with the mobile cloud. And now with AI, you know, most people are still, I think, and as a team, we think are underestimating how much AI will proliferate over the next, you know, 5, 10, 15, 20 years. You know, we would argue that AI is probably where the Internet was in maybe 2004, 2005 sort of thing. So we still have a long way to go. And if you just think about it, you know, the AI agents are being deployed at scale. Less than 1% of the world's population currently uses AI agents, as per some research I read recently. So, you know, agentic AI is the future. AI is the future. You know, there will be a lot more tokens which will be generated down the track, and companies which basically enable faster inference workloads and cheaper ones will, you know, thrive. We don't think that the trajectory is going to be smooth. You know, it never is. There will be macroeconomic problems down the track. There will probably be some geopolitical problems, maybe a Fed which has gone too far with monetary tightening and maybe a recession or two thrown in. But over time, we feel that AI is going to become huge in the future. So we don't necessarily subscribe to the school of thought which says that all this investment which is being done, the capex which is being spent by the hyperscalers and the new clouds, is all going to go to waste. We do think that this will generate an roi. In fact, there is some research which was produced recently which showed that companies which actually use AI are already seeing a big jump in the return on equity and also in the roi. So, you know, clearly AI is beneficial for companies, AI is beneficial for consumers. It will increase productivity, no doubt about it. You know, people are now doing a lot more with their time because of LLMs than they did before LLMs, you know, and it is helping different industries, different professions and so forth. So, you know, we don't think that this is a problem of the technology. What we are mindful of is basically timing the stock market, because the stock market is not the economy. You know, sometimes what we see when the market gets super excited, like it did during the dot com bubble, is that the stocks just go up and up and up. You know, the valuations become completely unhinged from reality. Investors just discount everything, bring everything forward, all the future returns are brought forward. And then, you know, you have a recession or you have an economic softness period, and then the earnings disappoint and the stocks go down 60, 70, 80%. Now we are not saying that there is going to be a recession, you know, this quarter or next quarter. All we are saying is that, you know, if something goes up 5, 6, 8x over a year or a year and a half historically there is not a single precedence of these gains being sustained or durable. You know, whether you look at the technology bubble in 2000, whether you look at the housing bubble, you look at the emerging markets bubble, precious metals, commodities, biotechnology bubble after that, and then you know, the Bitcoin bubble and then the SAS bubble during COVID and the E Commerce bubble. One of the key defining features of all these bubbles is that they always collapse. There is no historical precedence or example of any industry or any group which has gone up 8, 10x over a short period of time and found a permanently high plateau. It doesn't exist. So we think that there is still some money to be made in this memory stuff maybe for the next few months. Even if the market gets a whiff of any detailer agents here, these stocks are going to drop like flies. So you know, if you're going to enjoy the trade, by all means, you know, enjoy the trade, make the money, but have a trailing stop in place so you can actually get, get out in time before the music stops. Some of the other stuff looks interesting to us. You know, we've raised cash, we've changed our methodology of investing a little bit now to you know, keep up with the times. The market has just become so volatile and things have just become so crazy in both directions. Their stocks go up, you know, 2, 3x in months and then go down 50, 60% within weeks. So in this scenario, you know, we think buy and hold on individual companies at least was not going to do the trick. You are going to have, you know, a roller coaster experience like going to an amusement park. You know, you enjoy the ups and then you enjoy the downs. But when you basically finish the ride, you are no any richer than you when you started. So you know, if you are in that game, you know, by all means buy and hold. You know, we like to, you know, grow our own money and grow the money for our subscribers. So we've actually amended our strategy now to you know, time the intermediate swings in the market. So when we have a huge run in equities, you know, the market gets frothy, take 40, 45% off the table, lock in some gains, be patient, wait for the market to sell off as it always does and then you buy back again. So, you know, this way you are locking in, hopefully generating a higher CAGR over time and also reducing your drawdowns, which they look okay with the benefit of hindsight, but when you actually in the midst of one, when your account is down 30, 35%, most people don't feel very sanguine about it. So we will try and keep evolving our methodology. We've recently changed the way we hedge the portfolio too. We're not hedging the portfolio by shorting the NASDAQ now because we've seen that the correlations have been tossed out of the window. So we are basically buying puts on some of the overextended names. We don't short indices anymore because we've realized it no longer works as effectively like it used to before. So, you know, portfolio management, managing money, running your own money or running, you know, somebody else's money is, you know, and it is like a process. You just don't find something and you stick to it. You know, the market changes, what the market appreciates changes. So you have to keep evolving your own method and keep learning if you want to basically succeed in this game. And you have to leave your ego at the door, you know, something is not working. I think one thing we are pretty good at as a team is if we realize when something isn't working, you know, we're pretty quick to put our hand up and say, hey, you know, this is not working. We need to find something to improve the system. And as long as we can do that, we should continue to generate a decent return for ourselves primarily because we only invest our own capital and also for our subscribers. So this brings us to the end of today's pod. Unfortunately Damien wasn't here today, but hopefully you found the discussion useful and information beneficial to your own knowledge and investments. Before we head off, I would just like to mention that Alpha Target publishes Deep Research, institutional grade research on disruptive, rapidly growing companies in the public markets. Our subscription service basically provides you a monthly report which is a 250 page heavy duty month end publication which is published actually on the 2nd of each month, which has company reports on all the 15 to 18 companies in our portfolio. We also publish a weekly update every Monday highlighting the past week's news flow, the company earnings analyst notes and so forth, and stage analysis. So we basically combine fundamental and technical analysis to run our portfolio because we feel that if you combine both these disciplines, that's when you get the maximum bang for your buck. We also send out trading alerts to our subscribers. So whenever we make any changes to my own portfolio, whether I'm buying something, selling something, putting on a hedge, trimming or adding to a position, we send out alerts to all our subscribers via email as well as sms. So subscribers are always on top and aware of what we are doing with our own money, which hopefully will help them or helps them with their own portfolio management. Our performance over the past two years since we started in June 24th has been good. We're up about 65% versus I think about a 37 38% gain on the S&P 500. I don't know the numbers, the latest numbers at the top of my head, so I may be wrong by a couple of percentage points, but not by much. But we have actually outperformed the S and P by a wide margin. We can't promise that we will do that every year, but hopefully so far the results have been good and hopefully this will continue. Well, thank you so much for tuning in. If you would like to learn more about Alpha Target, please go to alphataget.com thank you so much and we'll see you again in a couple of weeks. Thank you for tuning in to Alpha Target Pulse. We publish deep research on disruptive, rapidly growing companies we invest in. Our subscribers gain exclusive access to our entire portfolio including exact portfolio allocations, in depth research coverage of companies news flow and our earnings breakdowns, trading alerts of our portfolio moves and our proprietary hedging indicators. You can find subscription information@alphataget.com we'd love to have you join us there.