The UNTHINKABLE Just Happened. Prep for the Dip!

[01:00:00:02 - 01:00:11:14] Iran just doubled down on closing down the Strait   of Hormuz again. And we all know how critical that  choke point is for oil, helium, sulfur, and really   every major input for AI infrastructure. [01:00:11:14 - 01:00:16:21]   And moments like this have caused some stocks to  just skyrocket because they're the most limited   resources that are feeding AI. [01:00:16:21 - 01:00:23:16]   a critical resource gets squeezed, the people  who are controlling it are the ones that   are making all the money. [01:00:23:16 - 01:00:37:01]   During the California gold rush, thousands  had packed up and rushed out west,   and most of them just went broke. But the  man who actually built his fortune wasn't   even panning for gold, but selling denim  pants and shovels to the people who were.   [01:00:37:01 - 01:03:59:09] And when the internet exploded in the 90s,   everyone piled their money into Cisco. And of  course they were right, but if you had bought   at the absolute peak, then you waited more than  20 years just to break even. And the thing is,   today is no different. Everyone is chasing Nvidia,  and I understand why. But in every one of those   moments, the real money was never in the obvious  play. It was in the companies that were sitting   on top of the physical constraint. You know,  the thing that the entire build out can't move   without. As of right now, artificial intelligence  has about nine of those types of constraints here   in 2026. I've spent roughly the last week  mapping out every single one of these, and I   tried to find the dominant stock that was sitting  on top of each one. So there happens to be about   nine choke points and nine stocks. So let's go  ahead and jump into the video. The first company   is going to be Vistra Corp. And I want to start  with a statement. 3,800 megawatts of contracted   nuclear power for AI data centers. That happens to  be double the nearest competitor. Vistra generates   power from nuclear plants that are across Texas  and the Midwest. And unlike regulated utilities,   they operate in a deregulated market, which means  that they can sign direct, long-term contracts   with hyperscalers at real market prices. Meta  locked in two gigawatts in a 20-year deal. And   Amazon Web Services signed a 1.2 gigawatt deal  out of Texas, also for 20 years. And the thing is,   those are not fixed price contracts. With Hormuz  closed and natural gas prices starting to climb,   nuclear's fixed cost structure means that every  dollar of rising energy costs, they're going to   expand Vistra's margin while everyone else's  begins to shrink. That means that they have   a win-win on revenue and margin both at the same  time. In fact, their revenue grew 13% last year,   and the company is guiding for a 22% Ibita  growth in 2026, as those contracts are just   ramping up. And they happen to have a peg ratio  of a 1.3 that's already priced for some growth,   but not for the upside I just spoke about. Now  we're going to go ahead and move on to the next   pick of Eaton Corporation, where transformer lead  times right now are 128 to 144 weeks. Like if you   do the math, that's nearly three years. And half  of all planned US data center builds in 2026.   They're already delayed or canceled simply because  of this. Eaton happens to own the full chain that   solves this problem. They've got the transformers,  the switchgear, power distribution units, and the   busway that moves electricity from the grid to  every server rack that's inside. They happen to   capture 3.4 million in revenue per megawatt of  data center capacity that's installed. And they   are the only company that owns every piece of that  chain from end to end. Data center orders alone   were up 200% in a single quarter because when a  project finally does get grid access, Eaton is the   first one they call. And their current backlog  represents 11 years of installed capacity at   2025 build rates, which means the demand they  are sitting on goes well beyond anything that   this AI cycle alone can absorb. Their earnings  per share has nearly doubled in just four years,   and their free cash flow has nearly tripled over  that same time period. An operating margin has   expanded from around 14% in 2021 to 19% today,  with the company already targeting 30% in its   core data center segment in 2026. So it's a  transformation that is far from finished.   [01:03:59:09 - 01:04:49:03] Now I have to point out that one   of the hottest sectors this year has been copper  and gold. And that brings us to today's sponsor of   U.S. Gold Mining Inc., which is on the NASDAQ.  They own the Whistler Project, a gold, copper,   and a silver exploration and development  opportunity just 100 miles from Anchorage,   Alaska. And copper and silver are now officially  on the U.S. critical minerals list. Now the   Whistler Project just completed a Preliminary  Economic Assessment, or PEA. And the results?   Estimated production of 2.7 million ounces  of gold, almost 600 million pounds of copper,   and nearly 6.6 million ounces of silver over a  nearly 15-year mining life, averaging 246,000   gold equivalent ounces per year. At $3,200 per  ounce gold, estimated payback of capital is at 2.1   years. But at current spot prices, that actually  drops to just 1.2 years. That PEA covers just one   of three gold copper deposits on the property,  with 25 exploration targets still ahead. And   nearly 80% of outstanding shares are tightly held  by the parent company and its insiders. Please do   your own due diligence on U.S. Gold Mining and  check out the link down in the description.   [01:04:49:03 - 01:05:57:15] Now I'm going to move on and talk about   what gets my attention about this next company of  Vertiv. And it's this. The latest NVIDIA AI racks   pull 132 kilowatts of power. But a traditional  server rack pulls in 10 to 15. That's nearly   10 times the load. And it broke every cooling  system data centers had relied on for decades.   Vertiv makes the power management and the liquid  cooling infrastructure that goes inside a data   center once electricity arrives. So UPS systems,  power distribution units, and liquid cooling as   one integrated system. In fact, in 2026, they are  the only major player that owns all three under   one single roof. Air cooling can't handle 132  kilowatts per rack. Liquid cooling is the only   answer. And Vertiv is the company NVIDIA named  as their preferred infrastructure provider for   AI factories globally early this year. Earnings  per share grew 199% last year, and the company   is guiding for 30% revenue growth in 2026. They  entered this year with $15 billion in committed   backlog, which is more than a full year's  worth of revenue already signed in orders.   [01:05:57:15 - 01:08:45:00] Now I'm going to move along to the   next one on the list, which is Micron Technology,  where every supplier of high bandwidth memory on   the planet is completely sold out through 2026.  All three of them, with waitlists running already   into 2027. HBM memory is physically stacked on  top of every NVIDIA AI chip. And the chip simply   doesn't function without it. And there's only  three companies that make it. SK-Hynix at 62%   of the global supply, Samsung at 17%, and Micron  is at 21%. There is no fourth option, and there's   no substitute. So when demand exceeds what three  companies can possibly produce, the entire AI   build-out is simply going to wait. And of course,  there's a backstory that changes everything for   Micron specifically, because the Hormuz closure  is disrupting the helium exports out of Qatar, and   it's tightening the chemical supply chains that  semiconductor fabs depend on at scale. SK-Hynix   runs Korean fabs that import those particular  inputs. Micron runs on US fabs with access   to domestic helium reserves and domestic supply  chains. So when Korean input costs begin to rise,   Micron either gains market share because SK-Hynix  produces less, or the whole market's price expands   because supply tightens even more. Either one of  those outcomes improves Micron's position. And   my expectation is that this plays out over the  next 12 to 18 months as Korean input costs just   begin to get priced in. Now Micron's revenue grew  56% last year, and earnings per share expanded   175% in a single fiscal year. And here's the  kicker, their peg ratio is currently at a 0.25,   which is absolutely crazy low because of the  expected growth. And that is why they are the   fastest growing company on this list. Now we're  going to look at a company that you may not have   heard of, and that's called Amcor Technologies,  where chips on a wafer substrate, or co-wars,   demand is growing at 80% a year. But the supply is  only growing at 50%, so as you can tell there's a   gap in the math, and it's not going to fix itself  in 2026. For reference, co-wars is the advanced   packaging process that assembles every NVIDIA AI  chip, bonding the compute die, the HBM memory,   and the inner connects into a single high density  module. Without it, the chip can't be built. TSMC   performs most of this work, but NVIDIA already  controls over half of TSMC's co-wars capacity,   which means every other AI chip maker, think  Broadcom, AMD, every custom ASIC builder,   they're going to need a second source. And  lo and behold, Amcor is that second source,   and their Arizona facility, funded in part by the  CHIPS Act, sits directly adjacent to the TSMC's US   FABs. Their advanced packaging business is  expected to nearly triple this year alone,   as that overflow demand begins to fall in [01:08:45:00 - 01:09:53:12]   Now we're going to move on to one of my  favorites to pay attention to which is Broadcom,   where the company entered 2026 with $73 billion  in signed AI orders, locking in 18 to 24 months   of committed revenue with the five largest  AI spenders in the world before the year even   started. Broadcom's a little unique because  they own two pieces of the networking silicon   layer simultaneously. The first is custom AI  accelerator design, when Google, Meta, Amazon,   and Microsoft decided they needed chips optimized  for their specific workloads rather than buying   Nvidia straight off the shelf. They went to  Broadcom. So think Google's TPU, Meta's MTIA,   and Amazon's Tranium are all co-designed with  Broadcom's teams. That happens to give them   60 to 70% of the custom AI chip market. Now that  second layer happens to be Ethernet switching, the   fabric that connects every server within the data  center. Now there's a battle between competing   networking standards and it's now settled. Open  Ethernet 1 and Broadcom's Tomahawk 6 controls   80% of that market. So when a hyperscaler builds a  million GPU cluster, every GPU in it communicates   through Broadcom silicon. AI happened to go from a  footnote to $8 billion just last quarter, which is   43% of total revenue for Broadcom and guidance  is already up 27% for the next quarter.   [01:09:53:12 - 01:10:17:14] Now on to one of the most   underappreciated plays on this list, which happens  to be Marvell technology. And the reason starts   with a single number, 63,800,000 G and higher  optical transceivers shipping this year. It's   a 2.6 times jump from last year alone, with  Marvell chip inside every single one of them.   That chip is an optical DSP. It's the processor  that converts electrical signals from a server   into light pulses traveling through the fiber.  Right now the market is a duopoly. You've got   Marvell and you've got Broadcom, [01:10:17:14 - 01:15:49:15]   which means when 63 million transceivers  are shipping, two companies are splitting   that revenue and Marvell holds the number one  position. The reason every hyperscaler is moving   to optical right now happens to be physics.  At 224 gigabits per second signaling speeds,   copper cable reaches less than one meter  and it consumes 30% of the cluster's total   power. Optical or photonics cuts that by 70%. The  data center industry does not get a choice here.   They're going to have to make the shift. So  look, four years ago, Marvell was an Ethernet   chip company. Then they happened to spend $10  billion acquiring InFi in 2021 and they built the   optical silicon position completely from scratch.  Earnings per share tripled in a single year on a   business segment that barely even existed when  that deal even closed. Now we're going to move   on to the next one, which may surprise a few  of you. And that's Southern Copper Corporation,   which sits at the base of every AI data center.  And the constraint is pretty straightforward.   27 to 33 tons of copper per megawatt of  installed capacity, which means a single   100 megawatt hyperscale facility consumes more  copper than most mid-sized city infrastructure   projects all combined. Copper is inside of  everything within a data center that is not   silicon. The power distribution infrastructure,  the cooling loops, the transformers, the bus bars,   everything that all the other companies on this  list are going to depend on in order to do their   job. All of it depends on copper and none of them  have a substitute. Here's something of importance.   The S&P global is projecting a 330,000 ton deficit  in 2026 alone, and that's widening to a 10 million   ton shortfall by 2040. This is all happening as  AI demand compounds on top of electrification and   grid expansion. Southern Copper operates the  largest reserve base of any publicly listed   miner on earth at 51.1 million tons, and they  produce at 42 cents per pound net cash cost.   Every other major producer happens to cost five  times more than that. Their revenue nearly doubled   over the past five years, while the company ran a  52% operating margin, which is essentially unheard   of in the mining area. And now we're going  to go ahead and round this out with Corning,   where the constraint is one multiplier. AI  data centers require 36 times more fiber   than a traditional server rack, and lead times on  ribbon fiber are already stretching past 60 weeks.   Corning makes the optical fiber cable  that connects every server, every switch,   and every GPU cluster inside a data center. And  every one of those 63 million optical transceivers   that we just talked about with Marvell requires  fiber to carry that signal. The photonics adoption   that drives Marvell's growth is the same  wave that drives fiber demand. It's just one   layer deep within the stack. Corning operates  the world's largest fiber cable plant in Hickory,   North Carolina, and in January of this year,  Meta signed a multi-year agreement worth up   to $6 billion for fiber, cable, and connectivity  supply, with new manufacturing expansion already   under construction. Demand for fiber today  is growing at 22 to 25% annually, but supply   can only expand at 11 to 19%. New fiber preform  capacity takes 18 to 24 months to come online,   which means the gap is not going to close  in 2026 and maybe not even in 2027. Their   operating income doubled in a single year as the  supply squeeze accelerated, and the hyperscaler   segment of their optical business grew 61%, and  the Meta deal has not even fully landed yet. Now,   before I cover how I'm going to be investing in  these AI choke points, if you're getting any value   from my videos, please consider pressing the like  button so my channel can continue to grow. So,   if I'm going to think about how to size these nine  positions, I'm not treating them all equally. I'm   thinking in three different layers. To me, the  first layer is the foundation. Think Vistra,   Eaton, and Vertiv. These are the power and the  cooling companies. Nothing in AI works without   them. I'd plan to put 40% here because there's  lower volatility, it's steady compounding,   and the demand story doesn't depend on any  one company winning the chip wars. Now,   the second layer is the silicon stack. Micron,  Amcorp, Broadcom, and Marvell. This is where the   highest upside lives over the next five years. The  photonics transition, it's just in the beginning   stages. Custom silicon is also accelerating,  and HBM demand has no ceiling in sight. I'd   be placing another 40% here, with Marvell and  Broadcom as the conviction positions inside of   that group. Now on to the third layer, which is  the physical materials. That's southern copper   and corning. The supply deficits are real and they  don't resolve quickly, but these are longer time   horizons. It's more value-oriented plays, so I'm  only going to put 20% of my money here. That's   just the framework as I would lay it out, and  of course a person doesn't need to own all nine,   but if you're building a portfolio around AI  infrastructure, these are the nine choke points   that I would want exposure to. As always, this is  for educational purposes, and it's not financial   advice. If you want the full breakdown with all  the numbers, the slides, they're all shared on my   Patreon. And as always, thank you so much  for watching, and I'll see you next time.