Everyone Sold Last Week. That Was The Mistake

[01:00:00:04 - 01:00:05:15] socks, which is the index that   tracks the entire semiconductor group,  but recently fell into a bear market.   [01:00:05:15 - 01:00:08:11] 20% from its high just a month ago.   [01:00:08:11 - 01:00:11:07] and at one point on Friday,   it was down almost 6%. [01:00:11:07 - 01:00:22:15]   And that's exactly why I wanted to make this video  right now. Because here's what these sell-offs do.   They don't just punish the overvalued names. They  actually drag down the strong ones with them.   [01:00:22:15 - 01:00:25:12] me, that's where the real opportunity hides.   [01:00:25:12 - 01:00:32:26] Because look at what these stocks have   done. If you were lucky enough to buy some of the  biggest names in AI over the past few years,   [01:00:32:26 - 01:00:34:10] You already know the feeling.   [01:00:34:10 - 01:00:42:08] A $10,000 investment in Micron just one year ago,   and you'd be sitting at around $85,000 right  now, and that's even after last week's drop.   [01:00:42:08 - 01:00:52:13] But those names have climbed so high   that a lot of investors are now afraid that it's a  bad time to get in. And this week's pullback only   makes that fear a little bit louder. [01:00:52:13 - 01:01:02:05]   So I want to shift gears and I want to talk about  strong companies that are either earlier in their   growth or where their future revenue is already  booked and the growth is still building.   [01:01:02:05 - 01:01:05:07] And when you look on paper,   they still look undervalued. [01:01:05:07 - 01:02:46:09]   But hey, before we jump in, quick reminder  that I'm not a financial advisor and I do   this all for educational purposes. But if you're  getting value from the research on my channel,   then please press the like button so my  channel can continue to grow. And if you   want deeper research and my actual trades that  I'm making in real time along with my portfolio,   then I share all of that with my community  over on Patreon. Now with that out of the way,   let's go ahead and get into those stocks.  The first company is Sterling Infrastructure,   the contractor that most every hyperscaler is  calling before a data center can even begin   breaking ground, where they move millions of cubic  yards of earth across a 200 to 300 acre site,   grading it dead flat and burying the power and  the water underneath it. Their entire edge is that   Sterling self-performs the whole scope of work  under one roof, so the job never stalls waiting   on the next crew. And for a hyperscaler, every day  that a data center sits unfinished costs far more   than a higher bid ever would. Now this used to be  a large road builder that was losing money. Today,   the data center site work is about 70% of  revenue and grew 174% year over year just   last quarter. Operating margin climbed under 8%  to almost 17% and earnings per share are up more   than 110% in just two years, with management  guiding for them to nearly double again this   year. And the backlog for them is the big tell.  They have 5.15 of work already booked against   just 2.49 billion in revenue last year. So more  than two full years is locked in before the year   even starts. And at a peg near a 0.88, you're  simply not overpaying for that kind of growth.   [01:02:46:09 - 01:02:57:04] is Oracle. And I know I'm going   to probably get a little bit of flack  for this one, but they're the company   that rents out the raw computing power that AI  labs like OpenAI use to train their models.   [01:02:57:04 - 01:03:09:19] Through its Cloud arm OCI,   Oracle builds the data centers and leases the  GPU capacity to the companies that need enormous   compute, but they don't want to rent it from  a direct competitor like Amazon or Google.   [01:03:09:19 - 01:03:18:10] Its edge is really just being   neutral and it engineered its cloud to run  AI training faster and cheaper than the older   general purpose clouds can. [01:03:18:10 - 01:03:24:21]   And Oracle's story this past year is pretty  wild because a year ago it was one of the   hottest stocks on the market. [01:03:24:21 - 01:03:28:28]   More than doubling to around $279  a share by last September.   [01:03:28:28 - 01:03:34:16] but since then it's been cut in   half and a drop that size usually  means that something's broken.   [01:03:34:16 - 01:03:39:20] But the reality is the business is thriving,   with revenue up 17% to $67 billion. [01:03:39:20 - 01:04:00:13]   But what really spooked investors was their  spending. So to build out those data centers this   demand requires, Oracle's capital expenditures  exploded in just 4 years to almost $56 billion.   And that pushed free cash flow from a positive  $12 billion to nearly $24 billion in the red,   [01:04:00:13 - 01:04:01:29] a $36 billion swing.   [01:04:01:29 - 01:04:12:07] here's the thing that that build   out is already spoken for. Oracle is spending  against a backlog of signed contracts that   grew 363% in a single year [01:04:12:07 - 01:04:13:29]   to $638 billion. [01:04:13:29 - 01:04:17:20]   That's nearly 10 times everything  that it sold last year alone.   [01:04:17:20 - 01:04:27:04] And like I said, that revenue   is booked and it's set to carry Oracle from  $67 billion in sales today to around $140   billion within just 3 years. [01:04:27:04 - 01:04:35:00]   at that pace, as free cash flow flips back to  positive over the next 2 years, the stock has   room to climb quite a bit. [01:04:35:00 - 01:04:41:21]   Like I said, I know I'm going to get a little bit  of flack about this company, but the reality is it   is undervalued and I see a lot of potential [01:04:41:21 - 01:06:02:04]   Now there's one small cap that you might want  to put on your research list, which happens   to be today's sponsor, Biostem Technologies,  which is a regenerative medicine company that   uses proprietary perinatal tissue oligraphs to  heal chronic, non-healing wounds, the kind that   come from diabetes and surgeries. And that matters  because the global wound care market is projected   to reach $27 billion by 2027. And the reason that  Biostem is now on our radar is because the company   just filed a Form 10 registration statement with  the SEC, which is a required step toward uplisting   from the OTC market to the NASDAQ, which would  expand access to capital and put the stock in   front of a much larger pool of investors. Their  bioretane processing method creates placental   oligraphs, built to improve wound closure  for diabetic foot ulcers. In a randomized   controlled trial published in the peer-reviewed  International Journal of Tissue Repair, wounds   treated with bioretane had a 53% probability of  healing, compared to 31% for the standard of care.   The company backs it with accreditation from  the American Association of Tissue Banks and a   portfolio of issued and pending patents protecting  its amniotic tissue technology. Biostem is putting   real clinical data behind its products, while it  works towards a NASDAQ uplisting. And that's the   type of momentum worth digging into. As always, do  your own diligence and check out the link down in   the description to learn more about Biostem. [01:06:02:04 - 01:17:09:03]   Now we'll move on to Modine, which builds the  entire liquid cooling system that keeps a rack   of AI chips from really just cooking itself. Now  a normal server rack used to draw between 5 and 15   kilowatts. It was cool enough for air, but an AI  rack now pulls 50 to over 130, forcing the entire   industry onto liquid cooling and it's going to be  piped straight to the chip. Now under its Airedale   brand, Modine sells the whole stack, the chillers,  the coolant distribution units, and the controls,   not just one piece of it. That full stack position  is why hyperscalers are reserving capacity years   out. And one customer committed to more than $4  billion of cooling gear from 2027 through 2029,   with about $165 million paid up front. Now a  few years ago, data centers were just a rounding   error here. But last year, the business grew  73% to about a third of the whole company.   Management is targeting close to $2 billion  by 2028, and adjusted earnings grew 24%,   while the company spins off its legacy auto parts  arm to become a pure play. The stock does trade a   touch rich against its own fair value, but at  a peg near $0.6, you are paying a fair price   for real growth. Now we're going to move on to  Broadcom, and they're the company that designs   the custom AI chips that the giant hyperscalers  are using to try to escape from Nvidia's pull,   and they're also building the network silicon  that wires them all together. In this case, when   Google, Meta, OpenAI, or Anthropic wants its own  accelerator instead of renting them from Nvidia,   it designs that chip with Broadcom. So in this  case, Broadcom gets paid no matter which model is   going to win. Its AI revenue is on track to nearly  triple in a single year, and it's already sitting   on about $73 billion in booked AI orders. That's  more than the entire company sold last year across   every business combined. All of it is throwing off  almost $27 billion in free cash flow. So that's   about 42 cents of every revenue dollar. So when  you look at it on paper, it looks expensive at 60   times trailing earnings, but that's an illusion  from VMware amortization burying those profits.   Because on forward earnings, it's closer to about  24 times. Now I get it, I did cover Broadcom   literally about a week ago, and I'm bringing it  right back because it sits at a peg near 0.48. So   it's still cheap relative to its growth. And now  we're going to move on to Lamentum, which owns   the EML laser chip, the tiny light source buried  inside every high end AI optical transceiver that   shuttles data between all the racks. So every 800  gig and 1.6 terabit module, well, it's going to   need several of these. And Lamentum makes 50 to  60% of them. And it is the only supplier shipping   the 200 gig per lane version at volume that  the newest links depend on. That is a genuine   choke point. And it's why Nvidia designated  Lamentum's optics for the next generation Rubin   platform. Look, for years, this was a struggling  telecom name bleeding money, running an operating   margin as low as negative 25%. Then the AI orders  began to hit, revenue climbed 90% year over year   last quarter, and the operating margin swung to  positive 22%. The point where a company crossing   into real earnings builds its fastest momentum.  The loudest signal really came from Nvidia, which   paid $2 billion as an investment into the company.  And I try to bring that up every time because   honestly, if Nvidia is willing to invest in them,  there's going to be some growth potential. But I   do want to point out that on top of the lasers,  Lamentum landed a multi-billion dollar optical   switching deal that's already carrying a backlog  north of $400 million. And honestly, that demand   is only just starting to hit their numbers. Now  we're going to move on to Credo technology, which   builds active electrical cables, a copper cable  with a signal processing chip built right into   the connector. Our reality is that plain copper is  going to die past a meter or two at these speeds,   and optics is expensive and power is not going  to be as expensive as it is. So Credo's cables   re-transmit the signal to run several meters  farther, cheaper and with far less power. So   what matters inside an AI cluster is reliability,  because one flaky connection can stall a training   run worth over $100 million. And Credo created  this category and they own roughly 88% of it.   And that has made it the fastest revenue more than  tripled last year, up 206%. And at a 68% growth   margin, which is software territory for a cable  maker, an operating margin swung from negative 19%   in the red to a 33% profit. And that's what I'm  really looking for. When we see that flip from   startup mode to real profitability, and that's  where the momentum really begins to build. And   for all that growth, the stock is still relatively  cheap on a peg basis, which is well under one. So   you're paying a lot less than a dollar for every  dollar of growth that's still ahead of it. Now   the next one up is SanDisk. And that's probably  going to be a little controversial in this group,   so please just hear me out. And they're known for  NAND flash. And that's the chips that store data   permanently. The same memory in your phone and in  the drives packed into the AI data center. Now to   make the point, the memory that's bolted next to  an AI chip today is HBM, or high bandwidth memory.   It's blazing fast, but it's expensive and it's  cramped. Holding only 50GB in a stack. SanDisk   pioneered an alternative called High Bandwidth  Flash, or HBF, that essentially matches that speed   but holds 8 to 16 times more on the order of 512GB  at a similar cost, which could roughly double the   size of SanDisk's market. And it is co-writing  the industry standard with SK Hynix, while a   giant like Samsung is still at the early concept  stage. So this is a company with only about 13% of   the NAND market holding the pen on technology that  could reset the entire business. As a reminder,   SanDisk had spun out of Western Digital last year  near $38 a share, and revenue just exploded 251%   in a single quarter. It used that windfall  to wipe out its entire debt and authorize a   $6 billion buyback. As of today, it's trading  under 10 times next year's earnings. And while   memory is cyclical, the multi-year supply deals  that are already signed and HBF still ahead,   they give real reasons that this run is going  to outlast all the ones before it. And now we're   going to move on to Tower Semiconductor, which  if you're familiar, they do not design chips.   They manufacture them for the companies that do.  Tower is a specialty analog foundry, and its most   important platform for AI is Silicon Photonics,  the chips that turn electrical signals into light   inside those optical connections. It's only one  of about three companies making photonics at real   scale, and the open neutral one that the whole  industry can share as well. Which is why names   like Marvell and Nvidia run their photonic chips  through Tower. So it's Silicon Photonics revenue   more than doubled in a year. And operating margin  climbed from about 9% to nearly 16%. Photonics is   only about 15% of Tower's revenue, so this is a  forward bet on top of a very broad foundry base.   And that premium multiple only works if the AI  leg keeps compounding. But with earnings growing   better than 50% a year, its forward pay ratio is  near 0.91, which keeps it very reasonable. And   their customers are pretty much just voting with  cash, against roughly $1.57 billion in annual and   Tower has already booked $1.3 billion of Silicon  Photonics orders for 2027, and they've taken   $290 million up front just to reserve their  capacity. And now we're going to follow that   up with COHU, which makes the machines that test  semiconductor chips after they come off the line,   holding each chip at a precise temperature,  pressing it onto the tester and sorting good   from bad before it ever ships. The parts that  matter for AI are its thermal handler, which   tests GPUs while controlling the intense heat they  throw off. And its NEON system, which inspects the   stacked memory and AI chips, with that inspection  revenue guided to grow about 80% this year alone.   Now to be very clear, AI is only about 2% of  its sales today, so this is really a deeply   cyclical test business coming off of the bottom.  At its last peak, COHU had earned $3.45 a share,   and it is now only beginning to climb back out of  that loss, with orders already up 57%. And behind   all of that is a pipeline of AI test work worth  around $750 million. It's not even booked yet,   but larger than its entire revenue last year  alone, and with 60% of sales recurring consumables   all underneath all of that. So you're not going to  be paying up for peak earnings here. You're buying   for a business that's still priced for downturn,  it is only now beginning to climb out of. Right as   the recovery and the AI demand start to arrive all  together. And then on to our next company, which   is InoData, which produces the expert-labeled  data that AI models are trained and tested on.   This is essentially the textbook and the exam  that a frontier model is going to learn from.   It already does this for five out of the seven  largest tech companies. And honestly, I think   its biggest wave is still ahead of it, and this is  where I'm speculating a little bit and trying to   see the trends. Because the way I see it, as every  large company starts building its own AI agents,   they're going to need the same data, plus a  way to continuously test whether those agents   are correct or not. Which turns lumpy project  work into recurring revenue and its market from   a few labs into potentially the whole economy,  with the agentic slice alone projected to grow   more than five times by 2030. Now a couple years  ago, there were some doubters about what they do,   so some short sellers actually filed some fraud  accusation against them. And those accusations   completely collapsed, where the DOJ and SEC  both closed it with no action. At the same time,   their revenue had grown 48%. And to give a little  more context, their revenue has nearly tripled in   two years to $252 million. And the company  made a complete swing from a loss to up to   $32 million in profit. And yet the market  is still treating it like an afterthought,   even though its growth is accelerating. So there  you have it. Those are the undervalued names that   I'm watching right now. It's the companies  that are actually building the AI boom,   and they're still growing extremely fast, and  they're still priced like the market hasn't   quite caught up to them quite yet. Now if you're  interested in what I'm looking at for specific   entry points on these stocks, I am going to be  looking to share that on my Patreon. So if you're   interested in the community, I'm going to do a  research to try to share that. In either case,   I hope you got some great value in today's  video. And as always, thanks for watching.