The Top 5 AI Stocks for Each Layer of the Machine

In a gold rush, the people who will get  rich aren't the ones digging for gold,   but the ones selling the shovels. AI is  the biggest gold rush of our lifetime,   and everyone is crowding into the same few  names. But in reality, there are 5 different   layers to this whole thing. It starts with the  equipment that makes the chips, then the memory,   to the chips themselves, and  the platforms that run them,   and finally the software you actually use. And each layer has its own shovel seller that's   genuinely hard to replace, with one that's even  quietly running a 72% profit margin, something   which Nvidia themselves can't even do. In this  video, I'll highlight one example in each of the 5   layers, and exactly what makes each one unique, so  let's not waste any time and let's jump right in. First up, in the equipment layer, we have  ASML. Without ASML, we wouldn't have any   AI chips at all. No memory chips, no GPU  chips, no Google TPU. Because before you   can even make any advanced chip, you would need  a machine that can print all the circuit stuff   onto a silicon. The process for making this  cutting-edge stuff is called EUV lithography. And just to give a sense of how far ahead ASML is,   right now, ASML is the only company that  makes 100% of the EUV machines on Earth,   and there's not even a single company  that is even close. The nearest attempt   is a state-backed program in China, which after  years of trying, still hasn't produced a single   working chip. By ASML's CEO's own estimate,  China is probably 10 to 15 years behind. So because every AI chip needs more advanced  manufacturing than the last, and because every   new fab needs ASML machines to fill it, the demand  for ASML machines just keeps growing. In fact,   the demand for these machines is so strong that  customers are now paying years in advance just to   reserve them. In 2025, ASML's revenue grew  16% from the year before. And they've just   raised their 2026 guidance again, guiding for  as much as another 22% of growth this year. However, just because ASML is a monopoly doesn't  mean they are immune to risks. While ASML is   safe from competition risk, their real risks  come from politics and timing. In the past,   China used to make up about 33% of ASML's  sales. However, export controls have since   dragged that down to 19%, and on top of  that, US now wants to cut off servicing   the older machines in China too, which would  shrink the China revenue down even further. Then there's the timing. ASML's  revenue is lumpy. Because each   machine is such a big-ticket order that when  chipmakers pause or delay their spending,   it can hit ASML's financials  hard. It happened in 2024,   when a pullback in chip spending left ASML's  revenue basically flat for the whole year. And it's happening again right now. ASML's  biggest customer, TSMC, just delayed moving to   ASML's next-generation machines until 2029, which  pushes the next big wave of orders further out. Valuation wise, ASML trades at around 45  times forward earnings. That puts it near   the 99th percentile of its 5-year history.  Though that's expected, given that ASML has   one of the most unbreakable moats in tech. On GuruFocus, if we assume that ASML can   continue to grow its EPS by 20% over the next  10 years, just like it had done in the past,   the stock is currently trading at  about 70% above its fair value.  Though that hasn't stopped analysts from being  bullish. Analysts are giving ASML a Strong Buy,   seeing around 19% upside from here. In short, ASML is the closest thing to   an unbreakable monopoly in this entire machine.  However, it isn't a stock trading at a discount.   Instead, you're paying a high price for one  of the safest businesses in the whole chain. One of the most common questions investors  ask is, "How do I find a good stock to   invest in?" With moomoo's Industry Chain  feature, you can uncover companies with   strong competitive advantages that often have  greater growth and price movement potential.  So if you go to the Markets tab,  then US, and scroll all the way down,   you'll see the Industry Chain. Then tap on the  AI theme, you'll see the whole thing in one view,   from the infrastructure layer that  made this whole thing possible,   to the algorithm layer that turns all that raw  computing power into the AI models themselves,   and the application layer that puts those  models into the tools people actually use. With this, you can find every stock sitting  inside each layer of the chain, including   the smaller names you'd never have thought to  look up. And if one of them catches your eye,   you can dig into the stock by heading over  to its company tab to see its earnings,   analyst ratings, valuation and company financials. Otherwise, if you don't feel like digging through   all of that yourself, you can also ask  moomoo AI to help you out. For example,   you can ask moomoo AI to summarize  the key companies in the AI sector,   and in seconds, it hands you back a  full analysis that pulls all the numbers   together and explains what they mean for you. From 6 to 20 July, moomoo is giving new users   free SK Hynix shares. Just deposit SGD 3,000,  and you'll get SGD 20 worth of SK Hynix shares. This is on top of the S$1,200 welcome rewards  when you fulfil the terms and conditions,   which includes a S$100 worth of NVDA shares  exclusively for my channel. Plus you'll also   get to enjoy zero* commission for US, HK, and  SG stock trading. Simply follow the steps here   and key in my promo code: KELVIN88 to get  your exclusive reward. For more details,   refer to the linked promotion  page in the description below. Next after the equipment layer, we have the memory  layer. And the king of this layer is SK Hynix,   a Korean chipmaker that's quietly become  one of the most important companies in AI.  Because the GPU chip that powers AI processes  data so fast, it needs memory that can feed it   data just as fast to keep up. That memory  is called HBM, or high-bandwidth memory,   and SK Hynix is the one that makes it. SK Hynix controls about 60% of the market.   So when you own most of a part no chip can  ship without, you get serious pricing power. In Q1 2026, SK Hynix's revenue nearly  tripled from a year ago, while operating   profit jumped by more than 400%. But the  number that really stands out is the margin.   SK Hynix posted a 72% operating margin,  which means for every dollar of revenue,   72 cents was pure operating profit. That's  even higher than Nvidia's best quarter ever. Right now, the demand is through  the roof. HBM is the single biggest   bottleneck in AI right now. It's  sold out all the way till 2027,   and the shortage isn't expected  to clear until 2028 or later. But here's the thing. Memory  isn't a clean compounder, instead,   it's a boom and bust business. The good news is  the downturn isn't expected until 2028 or later,   so it's not around the corner just yet.  However, the bad news is that almost all   of this demand is riding on a handful of  buyers, with Nvidia right at the front. So   the day Nvidia slows down its spending,  SK Hynix will be the first to feel it.  On top of that, there's also a fresh lawsuit  accusing the memory makers of teaming up to   keep prices high. If this is true, it  would mean some of these record margins   were propped up on purpose, and that's not  a profit that you can count on to last. SK Hynix currently trades at  around 8 times forward earnings,   which makes it look attractive. If we assume  that the company can continue to only grow at   15% EPS for the next 10 years, this would put its  stock price right around its fair value. Meaning,   there's currently no safety net in the  price for when the cycle eventually turns.  However, that hasn't stopped analysts  from being bullish. Right now,   analyst consensus is a Strong Buy, with  analysts seeing roughly 35% upside from here.  In short, this is one of the higher  upside opportunities on the whole list,   but also the one most tied to the cycle.  You're buying the most profitable company   in the whole machine at a bargain price,  but only for as long as the boom holds. Next, at the compute layer, we have the good ol'  Nvidia. Nvidia has been the poster child of AI,   and rightfully so. It creates the engine  that powers the entire thing. Roughly   80% of the world's AI accelerators  are Nvidia. But the chips are only   half of what makes Nvidia so hard to beat. The other half is CUDA, Nvidia's software   layer that lets developers actually program  those chips. CUDA has had a 15-year head start,   with more than 6 million developers building on  top of it. So once your team is trained on CUDA,   moving off it would mean rewriting  years of work. And that is the moat   that makes Nvidia almost impossible to replace. In its most recent quarter, Nvidia grew  its revenue by 85% from a year ago,   with the data center part up 92%.  And it did all of that at roughly   a 75% gross margin. And the demand isn't  slowing. Nvidia is printing cash at scale,   with its order book already full into  2027 thanks to the next Vera Rubin cycle. So what's the catch? Think of AI as having  two stages. Training is when you build and   teach the model, and inference is when  you actually run it to answer people's   questions. Nvidia completely dominates  the training stage. But for inference,   cheaper custom-built chips are  starting to eat into its share.  Some analysts think that Nvidia's share of  inference could fall from over 90% today   to as low as 20 to 30% by 2028, as Google,  Amazon and Broadcom roll out their own chips. Then there's also China. The US finally cleared  Nvidia to sell its H200 chips there again,   but this time it was China that  pushed back. Beijing blocked the   chips at customs and told its  own companies not to buy them,   so even with US approval, Nvidia's China  sales have basically fallen to zero. As for the valuation, Nvidia now trades at only  around 19 times forward earnings. This puts   it right near the bottom of its 5-year range,  meaning it's about as cheap as it's ever been.  And Wall Street sees it too.  Analysts rate Nvidia a Strong Buy,   seeing roughly 59% upside from here. Though, if we want to be conservative   and assume that Nvidia could continue to  grow its EPS at 20% over the next 10 years,   it would mean that the stock is currently  trading at slightly above its fair value.  In short, Nvidia is still the engine  of the whole machine. However,   while Nvidia is dominating today, its lead  in inference is already slipping. And that's   exactly why you don't want your whole AI bet  riding on the one chip everyone already owns. And that brings us to the platform layer. For  this layer, my pick would be Google. Unlike   ASML or Nvidia or SK Hynix, which each  focus on just one layer, Google is the   only one that owns all of the layers at once. It makes its own chip, it builds its own model,   it rents out its own cloud, and it even  owns the distribution layer through Search,   Android and YouTube. So whichever layer ends up  winning, Google is always there to take a cut.  And right now, the layer that's carrying it  the most is the cloud. Just last quarter,   Google Cloud grew 63% and crossed USD 20  billion in revenue for the first time. Its   cloud operating margin nearly doubled,  from about 18% to 33%. And its backlog,   which is the work it has already signed but  hasn't delivered yet, is now over USD 460 billion. On the model side, Gemini has now crossed 900  million monthly users. And because Google runs   Gemini on its own TPU chips, it can serve  AI cheaper than almost anyone. So while   rivals like Microsoft have to pay Nvidia's  markup on every chip, Google mostly skips it. Google's risk here isn't so much  about the competition, but rather,   the courts. So back in September 2025, a judge  ruled in Google's favor and allowed Google to   keep Chrome. But now the government and a  group of states are appealing that ruling,   which could put the Chrome  breakup back on the table. Then there's also the question of whether AI  will eat into Search. While Google Search is   still dominating, last quarter,  we saw that Google's third-party   Network ad revenue fell by 4%, even  as its core Search revenue grew 19%. As for the valuation, Google currently trades at  around 28 times forward earnings, which puts it at   the 74th percentile of its 5-year history. But  even if we assume a 22% EPS growth rate, which   is on the low end of what Google has done in the  past, the stock still comes out trading very close   to its fair value. And again, analysts are very  bullish on the stock, rating it as a Strong Buy,   while seeing around 24% upside from here. In short, Google is the closest thing to a   one-stock bet on the whole machine. The stock is  definitely not cheap right now, and the risks are   real. But it's also the only name here that  wins no matter which layer comes out on top. Last but not least, we have the application  layer. This is where the AI dollar lands at the   very end of the chain. So after every layer below  is built and running, the money will finally flow   to whoever isn't just using AI, but is also  selling it to everyone else. And Salesforce   is one of the biggest names doing exactly that. In case you've never used it, it's the software   companies use to run their entire sales and  customer operations on. And it's the giant in   the space, ranked the number 1 CRM software  for 13 years straight, holding around 20% of   the market with more CRM revenue than Microsoft,  Oracle, Adobe and SAP combined. But what really   stands out isn't its size. It's that a company  this big is still growing incredibly fast. So one of Salesforce's biggest bets is a  product called Agentforce. Instead of a   human doing the job, AI agents can now do  the actual work, from answering customers,   to handling support tickets, or  even following up with sales leads.  And the smart part is how it's priced.  Instead of charging a flat fee per user,   Agentforce charges by usage. So the more work  the AI does, the more Salesforce gets to earn. So far this strategy is working. Agentforce's  annual recurring revenue is already up 205% from a   year ago. And because Salesforce already owns the  customer relationships, the data, and the workflow   those agents run on, that usage would only  keep compounding as more companies lean on AI. However, there's still a real fear that's  hanging over that old per-seat business.   Because if AI agents can now do  the work instead of employees,   companies would need fewer seats, and  that older revenue would start to shrink.  In fact, some big enterprises have already  started trimming their Salesforce seat   counts as they roll out AI agents. And the  core, older part of Salesforce's business is   already slowing to roughly 9 to 10% growth.  So as to whether the new usage-based money   can grow faster than the old seat money  shrinks, we'll just have to wait and see. On the valuation front, after  all the AI-kills-software fears,   Salesforce now trades at only around 20  times forward earnings, which puts it among   the cheapest it has been in the past 5 years. And analysts seem to agree on this too. The   broader consensus is still a Strong Buy, while  analysts seeing more than 50% upside from here.  And on GuruFocus, even if we assign  a 20% EPS growth rate, which is quite   conservative relative to its historical  growth, Salesforce is currently trading   at less than 60% of its fair value, making  it one of the cheapest names on this list.  In short, Salesforce is the clearest way to gain  exposure to the view that AI ends up helping   software instead of killing it. The catch is that  the seat-versus-usage race is still unsettled.   But if the usage side wins, this could  be the most mispriced name of the bunch. So those were my 5 examples, one for each layer of  the AI machine. If there's one thing to take away,   it's this. Don't just buy the one chip  everyone already owns and call it investing   in AI. Own the whole machine instead.  And don’t buy these because I said so.   Do your own research and make sure each one  actually fits what you're trying to build.  Anyway, that's all for this video. Like, share,  and subscribe, and I'll see you in the next one.