Data Center Delays vs. Infinite AI Demand: The 2026 Bottleneck Trade — Jordi Visser (11 enero 2026)

[0:00] All right, welcome back to everyone
[0:01] who's back from the holidays. Uh, it's a
[0:05] good time to be uh, watching this week.
[0:08] There's a lot going on and I definitely
[0:11] have a uh a lot of slides to go through
[0:14] this week relative to what I expect to
[0:16] be both the uh investment side for this
[0:19] year, but also as you can see on on
[0:21] number five, I'm very focused on there
[0:23] being a, you know, a solid 20%
[0:26] correction at some point. Even if it's
[0:28] not for the broader market, um I
[0:30] definitely think we're going to have uh
[0:32] episodes of some risk this year that
[0:35] will be outside what everyone's looking
[0:36] for. It is not related to an AI bubble,
[0:39] but it is related to delays that I'm
[0:42] expecting in the data center side that
[0:44] are actually real. Uh it won't slow down
[0:48] the AI compute and the AI adoption. So
[0:50] you're going to get a transition in the
[0:51] market that to me is is a combination of
[0:54] a a change from momentum over the last
[0:56] three years relative to the chat GPT uh
[1:00] launch, but at the same time a 15-year
[1:04] shift related to software to hardware.
[1:07] So uh if I'm right, there's a lot of
[1:10] opportunities to make money. There's
[1:12] also a lot of opportunities this year to
[1:14] lose money and to get caught in
[1:16] positions that are just going against
[1:19] you continuously. Uh start of the year
[1:22] just to get going. For people that are
[1:23] bearish, uh I fully expect there to be
[1:26] 15% earnings growth in the S&P 500. I
[1:29] fully expect the S&P to at least be up
[1:31] 15%. On the back of that, I expect the
[1:34] economy to continue to show the
[1:36] productivity gains that showed up in the
[1:37] last two quarters. uh which means this
[1:40] is all about the AI productivity gains
[1:43] with inside the market. The replacement
[1:44] of people not firings but no more hiring
[1:49] relative to nominal GDP growing relative
[1:51] to revenues growing which means profit
[1:53] margins go but this is the year of the
[1:55] adopters. This is a broadening out. This
[1:58] is a PMI sensitive uh story where PMIs
[2:00] are rising. Uh if you're going to call
[2:03] me every week and say the PMI is not
[2:05] above 50 yet, uh you're wrong.
[2:08] go for it. Um, making money in the
[2:10] markets is not about telling me what's
[2:12] news. It's about telling me what's going
[2:14] to change over time. I'll go through a
[2:16] lot of those things, but everything has
[2:18] inflected since, for me, July, when I
[2:21] wrote my first piece on the PMIs. Uh, I
[2:23] use the markets as a tell for what's
[2:26] happening. And all you can see here is
[2:28] in the beginning part of the year, we've
[2:30] got green across the board, whether it's
[2:32] in local currency or in foreign
[2:34] currencies,
[2:36] every single part of the world is up to
[2:38] start the year, which finishes what
[2:40] happened last year. When you go through
[2:42] the sectors that have led, the S&P is up
[2:44] uh a little under 2% so far. It's all
[2:47] the things sensitive to the economy.
[2:49] Materials, consumer discretionary,
[2:50] industrials, energy leading the way. Uh
[2:53] the safety names towards the bottom and
[2:55] tech unchanged in the S&P 500. That will
[2:59] be a major theme for this year. That is
[3:00] part of the rotation. Small caps leading
[3:03] the way. Small cap, small caps are going
[3:05] to be a dominant theme for me this year.
[3:07] Midcaps, I expect the Russell 2000 to be
[3:09] up over 50% this year. Uh with the S&P
[3:12] up at least 15%. And again, basic
[3:15] materials, industrials, you've already
[3:17] got double digits there. Tech led by
[3:19] semiconductors, X Nvidia, X Broadcom.
[3:23] Again, part of this dramatic shift away
[3:25] from the last three years. And consumer
[3:27] discretionary as well. MCI World X the
[3:31] US. Again, we broke out last year
[3:34] effectively of a
[3:36] John Rog 18-year basing uh formation. uh
[3:41] a big base and we broke out last year.
[3:44] This is not ending right now. And when
[3:46] you go through the sectors of the world,
[3:49] every single sector in the large cap
[3:51] midcap is up as well. Banks leading the
[3:54] way. You don't have bare markets with
[3:55] banks making all-time highs. Transports
[3:58] breaking out finally out of a base part
[3:59] of the PMI special. I've been talking
[4:01] about transports. You're going to see
[4:02] this a lot this year in terms of the
[4:04] transport side. Everyone trying to pick
[4:06] a bottom in software. Don't waste your
[4:08] time. You might be able to catch a
[4:10] little bounce in it. But here's the
[4:12] white line. Equal weight software and
[4:15] equal weight semis on top and the S&P
[4:17] there. This is a hardware situation.
[4:20] This is the end of software. Uh I'll say
[4:23] it again and again. Software is now
[4:25] ubiquitous. The competition will be
[4:27] never ending. I can build apps. I have I
[4:30] will show you. I've showed you last
[4:31] week. I will show you more. Um here's
[4:34] IWM. Here is a 50% year coming out of
[4:38] the last time PMIs were depressed for a
[4:40] long period of time and the last time
[4:41] before we got a commodity cycle that I
[4:44] believe this is the first true PMI
[4:46] business cycle based on massive demand
[4:49] for things as opposed to re uh rein uh
[4:52] bringing inventories back up coming out
[4:54] of fears which was the story from the
[4:56] great financial crisis all the way
[4:58] through COVID. Um everything was a
[5:00] rebuild. There was never an expansion of
[5:03] any kind of uh breath. This one will be
[5:06] as we have to put intelligence in
[5:07] everything. Uh I'm going to keep showing
[5:09] this until the pay wall is launched and
[5:10] then the people who are paying for it
[5:12] get it. But I built an immune system
[5:13] which is a turbulence model has a 100
[5:15] assets in it dominated by AI looking for
[5:18] signs of stress with inside the market
[5:20] measuring it through correlations and
[5:22] volatility. So a coariance matrix
[5:25] nothing. It's completely healthy right
[5:27] now. um we did get kind of one line
[5:30] where we had during so far this um thing
[5:32] you'll get episodic like thin lines like
[5:35] this. What you're looking for is
[5:37] multiple days of seeing correlations and
[5:39] volatility
[5:41] change and then for bottoming is when
[5:43] you kind of get the pukes of these big
[5:45] red lines. So the turbulence model is
[5:47] saying everything is fine. Nothing to
[5:49] worry about right there in the immune
[5:50] system of the market. Uh Goldman Sachs
[5:53] put this out again. you have to make
[5:55] your calls. But if you're just going to
[5:56] use the numbers and you go, "What
[5:58] happens to multiples in the S&P?
[6:02] If we've got a stable GDP environment,
[6:04] which is fully what I expect, I actually
[6:06] think GDP will continue to surprise the
[6:07] upside." I think inflation, although
[6:09] it'll bottom this year in terms of
[6:10] somewhere a little lower than where it
[6:12] is right right now in the first quarter,
[6:14] it'll start to head higher, but only
[6:16] after energy prices start to move
[6:18] higher. If they don't move higher, we're
[6:20] not going to see that. Um, and then also
[6:22] in an environment where the Fed is
[6:24] cutting rates, you have both of those
[6:26] going there. They don't happen too
[6:27] often, but that's when you get the best
[6:29] side of multiple expansion. Uh, in terms
[6:31] of the payroll numbers for the week,
[6:34] just look at this red area. So, instead
[6:36] of just looking at the number which came
[6:38] in at 50,000, we had 56 last month, we
[6:41] had minus 173 in October. When you add
[6:43] up just the last four month, I mean,
[6:45] sorry, the last five months, you have
[6:46] zero. You go back and do this. We've
[6:48] created in the last 8 months less than
[6:51] 100,000 jobs. Now, the only part that's
[6:55] green is effectively education,
[6:57] healthcare, and leisure and hospitality.
[7:00] If you strip these out,
[7:02] you're dealing with negative numbers for
[7:04] a long time. Now, this is ex those two
[7:06] sectors, we are at minus 71,000 on a
[7:10] five-month average. So, we are in
[7:12] recession territory for the bulk of the
[7:14] economy from a jobs perspective. But
[7:18] again, in my opinion, this is all
[7:20] related to AI. This is related to AI
[7:23] driving the stock market higher,
[7:25] allowing more people to leave the labor
[7:27] force part labor force completely. Uh
[7:30] whether it's a two household, someone
[7:32] just gives in or they turn 50, they lose
[7:34] their job or they're getting pressured
[7:35] and they take a package. It doesn't
[7:37] really matter. They've created enough
[7:38] wealth uh via their house and via this
[7:41] the benefit of QE. But this is really
[7:43] the benefit of having AI an exponential
[7:45] innovation that started after the great
[7:47] financial crisis. So ADP confirmed the
[7:50] fact that we have no job creation. So
[7:52] it's not just that. And if you look at
[7:54] the jolts numbers, we have finally now
[7:57] have 685,000 more unemployed people than
[8:00] job openings in the US. This is going to
[8:03] be a continuation because this is the
[8:05] year that adoption picks up, meaning
[8:08] digital employees pick up dramatically.
[8:10] I'll talk more about this, but this is
[8:12] the main story is that you were at the
[8:14] bottom of adoption. So in the same way
[8:16] that when Chad GBT came out, we were at
[8:19] just starting the data center buildout,
[8:21] we are at a new investment theme that
[8:24] will forever only go one direction.
[8:27] Productivity will only go one direction
[8:29] from this point forward because we're
[8:31] not just dealing with AI agents. It
[8:34] eventually gets to humanoids. And as you
[8:36] go through this, you're replacing human
[8:38] beings that are very expensive with
[8:42] computers that work 24 hours a day. So
[8:44] per hour on everything. And if you go
[8:47] through productivity numbers, you're
[8:48] going to see per hour. When you get the
[8:51] GD, the employment numbers, what we
[8:53] normally care about is the wage
[8:54] component and the amount of jobs. We
[8:57] don't spend time on the hours worked.
[8:59] But if all of a sudden we go from 34
[9:02] hours work to just multiply 7 days a
[9:04] week, 24 hours, you've got a massive
[9:06] productivity boom, which is coming,
[9:10] it's not a guess. It is a certainty that
[9:12] we will have productivity gains at the
[9:14] bottom that will continue. So for people
[9:16] trying to make the argument that this
[9:17] isn't going to get better, this is the
[9:18] productivity over the last two uh
[9:21] quarters. And again, you can see per
[9:23] hour,
[9:25] 24-hour workers per hour. How much work
[9:28] is going to get done when people are not
[9:30] working a 40-hour work week, but they're
[9:32] working a 100 change hour work week. So,
[9:37] this is all going to happen and the cost
[9:38] of those people goes down. So,
[9:40] productivity is going to only go up from
[9:42] here regardless of what your views may
[9:44] be. If you haven't thought about it in
[9:46] those simplistic terms, you will. Here's
[9:47] what's happened in productivity or at
[9:49] least nominal GDP. nominal GDP which
[9:51] absolutely does not purely capture
[9:55] uh the intangibles of AI. This is going
[9:57] to be a bigger issue as the velocity of
[9:59] money increases with stable coins and
[10:00] with tokenization. I'm not going to go
[10:02] through that today, but if you haven't
[10:03] thought about that, give 22V a call. I'm
[10:06] happy to come in and do what I'm doing
[10:07] for many many places at this point,
[10:09] which is preparing them not just for the
[10:11] AI productivity side and showing them
[10:12] how to incorporate in their business,
[10:14] but also the changes that are coming
[10:15] into measurements that are no longer
[10:17] relevant like GDP. So you can sit there
[10:19] and pay attention, but the reality is
[10:20] nominal GDP just grew to 8% in an
[10:23] annualized basis in the prior quarter.
[10:26] Look at how many times that happened
[10:28] this century outside of the postcoavid
[10:30] bounce. You're dealing with the largest
[10:32] annualized nominal GDP. U Atlanta Fed
[10:36] GDP now and I should remind you all that
[10:38] is during a time period where everyone
[10:40] was worried about the tariffs causing a
[10:42] recession. So, not only is the data
[10:45] saying something different is going on
[10:46] with AI, intelligent people are having a
[10:49] really hard time thinking about how AI
[10:52] is incorporating itself into the market.
[10:54] You cannot capture it. go back and read
[10:56] the Allen Gr the Greenspan pieces back
[10:58] around 1999
[11:00] on why we can't cover intangibles in
[11:03] this measurement which was not created
[11:05] for it and why this will continue to be
[11:07] a situation where profit margins grow
[11:10] stocks reflect the productivity GDP will
[11:13] start to reflect it now because of what
[11:15] I showed which is there is a
[11:16] relationship between how many hours
[11:18] worked for the job for the work that
[11:20] you're getting and the transactions that
[11:22] are happening Atlanta Fed GDP now for Q4
[11:26] is currently at 5%. That has some noise
[11:29] in it because of the trade situation,
[11:31] but the reality is currently we're five.
[11:33] We'll see where it ends up. We got the
[11:35] ISM services number. The only reason I
[11:37] want you to look at this again, if you
[11:38] look at the granular level, the new
[11:40] orders component went up to the highest
[11:41] level since 24. So, you're starting to
[11:46] get the new orders coming in. On the PMI
[11:48] side, we didn't get the ISM PMI to break
[11:51] above 50, but again, we've been above 50
[11:53] for almost a year now on the global one.
[11:57] The ISM survey, which is more dependent
[12:00] on the foreign side and what the tariffs
[12:02] did, uh people, it seems like
[12:05] corporations still haven't adjusted.
[12:06] That'll all change as we move forward.
[12:08] So again, if you're focused on the fact
[12:10] that the PMIs are not there, what
[12:12] matters is what the market sees. The
[12:14] market always gets ahead. This was my
[12:16] outlook piece for this year, the
[12:18] physical world upgrade. Why AI's next
[12:20] chapter demands a hardware renaissance.
[12:22] There are many, many trade ideas. If you
[12:24] guys did not get it, I'm going to try to
[12:26] do what I did on and I'll I'll reference
[12:28] this as we go on. Uh, but the PMI story
[12:32] is a massive story because this is not a
[12:34] one quarter thing. This is not a
[12:36] one-year thing. We will be building more
[12:38] cars, more phones, more appliances, more
[12:42] computers that all have intelligence in
[12:44] it. I'll talk more about that. But if
[12:46] you've read my pieces on edge devices,
[12:48] on NPUs, the Nvidia Grock deal, if
[12:50] you're not doing your homework on this
[12:52] and you're just sitting there reading
[12:53] research reports from cellside research
[12:55] people talking about what the economy is
[12:57] going to be without understanding AI,
[12:59] it's going to be a very challenging year
[13:01] to answer the question as to what's
[13:02] going on. I want to remind people, I
[13:04] wrote this piece in July 21st. The
[13:06] reason I'm bringing it up now in terms
[13:08] of PMI, paranoia motiv motivates
[13:10] investment. This is the most critical
[13:12] part because this was the best week to
[13:15] use as an example of this paranoia and
[13:18] why this drives the need for a race for
[13:21] commodities, a race for hardware, a race
[13:24] for chips, a race for countries.
[13:29] I believe the US manufacturing PMIS is
[13:31] poised to move significantly higher
[13:32] driven not by traditional demand cycles
[13:34] around real estate and autos but by a
[13:36] new and powerful force paranoia.
[13:40] governments and corporations. And that's
[13:42] the key thing. The environment of
[13:44] uncertainty and strategic anxiety is now
[13:46] driving investment at a scale and
[13:47] intensity that I believe will be showing
[13:49] up in the manufacturing data through a
[13:51] long overdue rise in PMIs. And soon,
[13:53] this paranoia won't just be confined to
[13:55] the boardrooms or defense departments.
[13:57] It will spread to the workforce as
[13:58] employees begin to grasp how rapidly AI
[14:01] is reshaping the nature of work and the
[14:02] security of their jobs. That is exactly
[14:04] what's going on later in the piece. The
[14:06] underlying shift in PMI should not be
[14:08] interpreted as merely whatever. Begin to
[14:09] recognize that this is not about a
[14:11] near-term stimulus or post-pandemic
[14:13] normalization. That's the inventory
[14:15] rebuild story that I'm saying, but
[14:16] instead a longduration investment cycle
[14:18] tied to AI, military modernization. We
[14:22] saw Trump say he's going to increase the
[14:25] military budget. We've seen this go on
[14:28] in Europe in terms of the breakdown of
[14:30] NATO. And the path to AGI will force
[14:33] corporations to continue to spend money
[14:35] on data centers regardless of any
[14:37] bottlenecks that appear. And regardless
[14:39] of the price that happens for the
[14:42] inputs, which I will talk about, this is
[14:44] the beginning of a secular shift. As
[14:45] Mark Andre put it, this shift is about
[14:48] soft isn't about software eating the
[14:50] world. It's about machines eating the
[14:51] world. Could not agree more. I have a
[14:53] big part on Mark Andre, an interview we
[14:56] did this week. Um, so I go through this.
[14:59] You got to look for different structural
[15:01] winners. Commodities usually have a huge
[15:04] impact on what goes on. The compute is
[15:06] real. The applications are scaling. The
[15:08] infrastructure is being stretched. And
[15:09] financing is coming due to paranoia
[15:12] around military supremacy at the country
[15:14] level and obsolescence from the
[15:15] hyperscalers. The world is not simply
[15:18] prepared from an investment standpoint
[15:20] for the electricity, hardware, and
[15:22] commodity needs of the AI revolution.
[15:24] This will be a return to physicality.
[15:26] This is what the future is. The past is
[15:29] about software.
[15:31] Here is the proof. Venezuela, Greenland,
[15:34] let's go give money to defense names as
[15:37] well. These all have something to do
[15:39] with what I said. They're all part of
[15:41] the bigger picture of commodities, rare
[15:44] earths, China, Russia. This is a
[15:47] strategic change because the military
[15:49] situation of the world is in a race. And
[15:52] if you don't believe that, you're not
[15:54] paying attention to what happened with
[15:55] the tariffs last year when I started
[15:57] saying very early on. Rare earth will be
[16:00] the deciding factor. We cannot put
[16:02] tariffs on China. The tariff situation
[16:04] is over. The only way it was going to be
[16:06] a big thing is if it was China and China
[16:08] had something above them. We're still
[16:10] playing the game of rare earth and
[16:11] commodities. Guys, you can see it in
[16:14] silver. Prices are soaring just as China
[16:16] moves to restrict exports, prompting
[16:18] fresh worries about supply of metal.
[16:20] This is not good. Elon Musk talks about
[16:22] how critical in this article silver is
[16:25] for AI. Big copper shortage to pose
[16:27] systemic risk to global economies, warns
[16:30] S&P. This is in the FT this week. I'm
[16:32] throwing this in there just because for
[16:34] the people who keep fading Trump's
[16:36] decision to make sure that the
[16:37] manufacturing side and that people can
[16:40] afford to buy houses. He's doing
[16:42] everything he possibly can because right
[16:44] now is an 80cent probability in
[16:46] prediction markets that the Dems win the
[16:48] house. So, whatever he's going to throw
[16:50] at this, it's credit card uh limiting
[16:53] credit card uh interest rates this
[16:55] morning. It doesn't matter. If you
[16:57] continue to fade this and think that
[16:58] shorting these things are going against
[17:00] the administration is a good idea, I
[17:02] think you're going to suffer. Here is a
[17:04] rate of change of silver going back to
[17:06] 1990, the last 100 days.
[17:10] Only time it was bigger was during
[17:13] again a supply shortage issue. Uh in
[17:16] terms of uh what happened coming out of
[17:18] COVID copper same chart 100 days you're
[17:23] getting rises in copper DRAM prices
[17:25] insanity you stay between two and a half
[17:27] and five for basically the entire time
[17:30] until when? September. Go look when the
[17:33] silver broke out. Look when I wrote the
[17:35] piece in July. Basically, the PMIs, the
[17:37] market started discounting that PMIs
[17:39] were going higher in September. What
[17:41] else blew out in September, something I
[17:42] wrote about last year. I am not a
[17:44] bottoms up semiconductor person, but I
[17:45] pitched and pounded the table on Micron
[17:48] all year. Made a lot of money on Micron
[17:50] personally. Here's what Micron has done.
[17:52] If you don't believe in Micron on this,
[17:54] here's Western Digital. You don't see
[17:56] charts like this with software companies
[17:59] unless it's a bubble. These have to do
[18:00] with shortages. These are all about
[18:03] shortages.
[18:04] Alon Musk, most important interview I've
[18:07] heard for this year in terms of an
[18:09] outlook piece. No one will listen to it
[18:11] because most people don't sit there and
[18:13] believe Alon Musk is someone you want to
[18:15] listen to when he's the guy that
[18:16] actually knows everything that's
[18:18] happening from an industrial and a power
[18:19] perspective. He was on moonshots for a
[18:21] three-hour podcast. I'm not going to
[18:23] read this whole thing. I'm just going to
[18:25] read you the front first line and a
[18:27] couple points in here. He frames the
[18:29] current mo moment as a near-term
[18:31] transition problem, not a long run
[18:33] scarcity problem. This is really the
[18:35] issue, and I talk about this all the
[18:37] time. The next five years is a dramatic
[18:40] issue for the world. We underinvested in
[18:42] power. We underinvested in hardware
[18:45] things like gas turbines, transformers.
[18:48] Our grid is old and and decaying. We are
[18:51] not ready for how quickly the needs for
[18:54] AI have come. And that's going to cause
[18:56] issues. the labor market who went to
[18:58] school who thinks looking for answers
[19:01] and doesn't know how to use AI and is
[19:03] resistant to use it. Companies whose
[19:05] structure does not allow them to thrive
[19:07] with AI, lack the curiosity, lack the
[19:10] people to do it. We're not ready for
[19:13] this. We're not ready to have an economy
[19:15] where people are going to lose their
[19:17] jobs because not because they're they're
[19:18] not capable of getting a job. They
[19:20] refuse to learn artificial intelligence.
[19:22] They refuse to read anything that says
[19:25] they should be using it. Instead, they
[19:27] just read what they want to read, which
[19:29] is it's a bubble. It's going to end. I
[19:31] don't need to to know anything. That's
[19:33] what the last 3 years has been about.
[19:35] And as someone who has to sit there in
[19:36] meetings and listen to people, usually
[19:38] above the age of 50, where I am too.
[19:41] Start the argument of it hallucinates.
[19:43] It does this. You guys are all wrong.
[19:45] You're falling behind by the second.
[19:48] Reach out to 22V. I'm happy to do
[19:50] presentations. I'm doing more and more
[19:51] of them. And I'm actually doing them
[19:52] now, which I'll show you later on how to
[19:54] get from point A to point C. And if you
[19:57] don't want to do it, you should be
[19:58] hiring people, young people, hoarding
[20:01] them. The biggest thing that I would be
[20:02] doing running a business right now is
[20:04] looking for all of the talent that are
[20:06] having trouble getting jobs out of
[20:08] universities that don't have the best
[20:10] grades. They have the highest curiosity
[20:12] and they're AI native already. you can
[20:14] make a lot of money because they can
[20:16] pick this stuff up in minutes, not days,
[20:20] not weeks, in minutes. So, all of this
[20:22] stuff on here leads to a violent regime
[20:25] shift, short-term social unrest,
[20:28] long-term abundance, and a narrow window
[20:30] where infrastructure, not ideology, der
[20:33] determines outcomes. You've got to
[20:35] generate the electricity. You need
[20:36] transformers. You've got to convert that
[20:38] voltage to something that computers can
[20:40] digest. You've got to cool the
[20:41] computers. Electricity generation and
[20:43] cooling are the limiting factors of AI.
[20:45] This is the most important point of the
[20:46] three-hour podcast. We are at the point
[20:50] where all of these bottlenecks are now
[20:52] going to impact investors because we
[20:55] will have delays in data centers. We
[20:57] will not have delays in progress in AI
[21:00] and I'll go through the reasons why. But
[21:02] if you've invested in the data centers
[21:04] and you're long GE Vernova and you're
[21:07] long I'm not going to name all of them
[21:09] because you guys are long all of them.
[21:10] Everybody is. It's not that these
[21:13] companies are not going to make money.
[21:14] The question is over the next two years
[21:16] are they going to be able to surprise on
[21:19] the upside. And the reason that becomes
[21:21] important if your multiple is now 30 40
[21:24] 50 and you've built in the next three
[21:26] years that's fine except for the fact
[21:28] that for the next two years there's
[21:30] going to be a lot of headwinds to
[21:32] actually get the data centers built to
[21:34] actually find the power and can get it
[21:36] in there because of these issues. This
[21:38] is a major major issue because the grid
[21:41] can't give everybody what it needs as
[21:43] quickly as people want. It's an issue
[21:45] not just because of the bottlenecks
[21:46] here. Labor is a bottleneck. The
[21:49] governments are becoming a bottleneck at
[21:51] the state level. All of these things
[21:52] will cause delays and as I'll highlight
[21:55] and as you know whenever we've had delay
[21:57] fears in data centers or efficiency
[21:59] gains on chips which are accelerating
[22:01] dramatically based on what Nvidia said
[22:03] this week, you will get pullbacks in the
[22:05] marketplace. They will be painful. This
[22:08] year is about finding adopters. This
[22:10] year is about finding commodity stocks.
[22:12] Those things have not built in the next
[22:15] three years of demand. What has built in
[22:17] the next three years of demand already
[22:19] is not ready for bottlenecks, which I
[22:21] think are going to come. So just be
[22:23] prepared as I go through this. Look at
[22:25] your portfolio. I'm going to show you
[22:26] charts at the very end that highlight
[22:28] that these things are already starting
[22:30] to underperform. Why is Nvidia unchanged
[22:33] and and and Broadcom unchanged while
[22:36] many semi-names are up 25%. I put a list
[22:38] out with 22V that I only showed two
[22:40] names on last week. I'm going to show
[22:41] you guys the entire list this week just
[22:43] because the payw wall is not done yet.
[22:45] And for those of you who keep reaching
[22:46] out, the payw wall will be done this
[22:47] month. I want to make sure and so does
[22:49] 22V that everything is functioning and
[22:51] everything is good. There's a lot of
[22:54] stuff to go up on it and I want to make
[22:55] sure it works so we don't have to put it
[22:57] up, pull it down, put it up, put it
[22:58] down. So unfortunately it hasn't been
[23:00] finished yet. These are the main points
[23:01] he makes in there. And what I do is I
[23:03] have a prompt that just says go through
[23:05] and give me the single ex the signal
[23:07] extraction. Eliminate the noise and just
[23:09] give me the most important things that
[23:10] he mentioned. White collar cognitive
[23:12] labor is the first category to be
[23:13] displaced. Absolutely the case. But if
[23:15] you want to train your white collar
[23:16] labor, it can be done. Progress is
[23:18] capped by power generation, power
[23:20] conversion, and cooling, not model
[23:22] design. This becomes important if you
[23:24] made money on Bloom Energy this year.
[23:26] We're entering a new stage. I'll go
[23:27] through the the the bring your own or
[23:30] bring your own generation, bring your
[23:31] own power, whatever you want to call it.
[23:33] Uh it's a major theme going forward.
[23:34] I'll probably write something on it next
[23:36] week. Prices, not wages, collapse in an
[23:38] AI economy. You're going to see massive
[23:40] deflationary pressures. If your business
[23:43] gets hurt by deflation, anything built
[23:45] on code gets hurt by deflation. Get
[23:48] ready. Data centers, grids,
[23:50] transformers, and cooling systems are
[23:51] the choke points. investors
[23:53] underestimate
[23:56] the inflection point and this is where
[23:58] he brings up two important points.
[24:00] Number one, China will exceed US
[24:02] electricity output by multiples by 2026.
[24:05] So if energy efficiency or if energy and
[24:08] power is the main point,
[24:11] China will exceed electricity. I
[24:13] actually don't think I'm not in
[24:15] agreement completely on this. I think
[24:17] the compute side will be okay and I'll
[24:20] go through why. But his point, chip over
[24:23] supply can coexist with an inability to
[24:26] deploy. This is the most important
[24:28] thing. Cyclicality appears artificial.
[24:32] It's caused by power, not demand. AI
[24:36] demand is infinite. I'll keep saying
[24:37] that. Every single month, every single
[24:40] day now that I use it, it's better than
[24:42] it was the day before. It's no longer
[24:44] every release. It's every day. Every
[24:46] single day, I do something differently.
[24:48] I learned something new that it couldn't
[24:49] do the day before, two weeks before,
[24:52] because a lot of these models,
[24:54] particularly Grock, they're updated
[24:55] regularly just like your iPhone would,
[24:57] except it's happening without needing to
[24:59] do a big update. It's from the cloud.
[25:01] It's I'm doing this most of this stuff,
[25:03] not on my uh application. I'm doing it
[25:06] on Chrome, as you can see. All right.
[25:09] The t the mismatch decides the winners.
[25:12] Semiconductor supply is no longer the
[25:14] dominant limiter. AI development cycles
[25:16] are now measured in months while the
[25:18] physical systems required to support
[25:20] them power generation grid are measured
[25:24] in years. This is the mismatch. Chips
[25:26] without power are stranded assets. He
[25:29] makes the point that at some point this
[25:31] year within 12 months we will have a
[25:33] bunch of GPUs just sitting around
[25:35] because we can't get the power to turn
[25:37] them on. Think of having a billion light
[25:38] bulbs but not having enough sockets to
[25:42] plug the light bulbs in or enough power
[25:44] to get them going. He's saying we're
[25:45] going to be there this year. That's how
[25:46] big the problem is in terms of the uh
[25:49] power. So he was arguing that when asked
[25:51] the question, is Taiwan semi holding
[25:53] things up? He said they might be worried
[25:55] about the past in terms of cyclicality
[25:57] and overbuilding compute, but the
[25:59] reality is they're right. We're going to
[26:01] run out of power. Uh we're not going to
[26:03] be able to plug them in. They're going
[26:04] to be stranded. You're going to purchase
[26:05] them, but they're not going to be doing
[26:07] anything. So just think about that when
[26:09] you're going through. That is a problem
[26:10] that Elon Musk thinks will happen this
[26:12] year. He's already seeing the
[26:14] bottleneck.
[26:16] He goes through these components again
[26:18] that we've talked about. Compute is
[26:20] already power constrained at the margin.
[26:22] This is why hyperscalers are building
[26:24] off-rid. So this is where we start
[26:26] getting into Alon Mus saying you have to
[26:28] bring your own generation. You can't
[26:30] depend on the grid to do that. You
[26:33] obviously have to find a way to build it
[26:35] out yourself which is getting harder and
[26:37] harder. Um the real problem is the
[26:40] infrastructure lag. compute will will
[26:42] appear cyclical on paper while demand is
[26:44] infinite in reality. You're going to get
[26:46] GPUs uh this year into next year sitting
[26:50] idling waiting for transformers,
[26:52] training clusters limited by cooling not
[26:53] chips. Capital markets are misreading
[26:56] underutilization as over supply and
[26:59] that's what will happen this year. there
[27:00] will be effect uh uh at least 70% of the
[27:04] people purely based on the AI bubble um
[27:07] Deutsche Bank thing that I showed last
[27:08] week that the number one risk for this
[27:10] year was that there's a bubble in AI. It
[27:12] was over 50% and the next thing below it
[27:14] I think was 20 or something along those
[27:16] lines. AI growth versus power
[27:18] availability a perfect storm in 2026.
[27:22] The reason this is important and the
[27:24] reason Alon Musk is the single person to
[27:26] know he has already built the largest
[27:28] data center. He's on his way to it being
[27:30] 1 gawatts. There are no other one
[27:32] gigawatt data centers. He's structure or
[27:35] he's his goal is to for it to be two
[27:37] gigawatts soon enough. He's done a
[27:40] cluster that's already over 100,000. I
[27:42] think the largest cluster of GPUs out
[27:44] there currently is still only 50. We
[27:46] don't have Stargate online yet. We don't
[27:48] have any of these other ones. Elon Musk
[27:50] did a Colossus, the first version in 122
[27:53] days. He is the person that has the only
[27:56] experience of the difference between
[27:58] doing a 50 megawatt data center and a 1
[28:02] gigawatt until you've done it which
[28:04] nobody thought you could have clusters
[28:06] that big. He's the one that started this
[28:08] whole thing. So he should be the one
[28:11] since he now can talk about batteries
[28:14] became operational and mandatory not
[28:16] optional. That was based on his
[28:17] exposure. And the reason is there's all
[28:20] kinds of instability that are going on
[28:22] because power demand is not linear. So
[28:26] it's very volatile. He exposed how AI
[28:28] training creates highly volatile
[28:30] nonlinear power demand. Vertical
[28:32] integration was forced by execution
[28:34] speed. They did the whole thing.
[28:38] Colossus reframed cyclicality is
[28:40] stranded compute risk. Again, he's the
[28:42] expert on this. If you want to go spend
[28:44] time, spend time with Alon Musk on what
[28:47] it was. He's not Satcha Nadella. He's
[28:50] not Sundar Pachai. He's not. Go through
[28:53] the list of them. Demis Sabis, all of
[28:55] these guys, they're all really, really
[28:57] good technology people. Elon Musk is an
[29:00] engineer. He is a mass scale person.
[29:04] That's what we're at now. That's what
[29:06] PMIs are about. That's why Tesla, in my
[29:08] opinion, will be the best performing
[29:09] one. But it's not just Tesla. his entire
[29:12] business, XAI, SpaceX, Tesla, they're
[29:15] all interconnected into his long-term
[29:17] vision as a systems thinker. The winners
[29:19] in the AI arms race will not be the
[29:21] firms with the most advanced chips, but
[29:23] those who can energize and deploy them
[29:25] fastest. Power, not silicon, is now the
[29:28] pacing item and execution. We are at the
[29:30] inflection point for where software did
[29:33] not need energy. We got fracking. We got
[29:36] software that knocked oil prices down.
[29:39] That knocked natural gas down. We are at
[29:42] the early stages of a new commodity boom
[29:45] because commodities are not a big cost
[29:48] in this situation. And unlike when gas
[29:51] at the pump would triple and oil would
[29:54] triple and we'd see the economy go down,
[29:56] this is not an oil sensitive thing. The
[29:59] energy power is being bottlenecked by
[30:02] the things going on. And we have enough
[30:04] energy in the world. We have enough
[30:07] enough natural gas. So this is a very
[30:09] unique situation. But we do not have
[30:12] enough copper. We do not have enough
[30:15] silver. If you want to read something
[30:17] good on the uh bring your own power semi
[30:21] analysis did a very very long piece.
[30:23] It's free. It's in Substack. How AI labs
[30:25] are solving the power crisis. Highly
[30:27] recommend the semi analysis guys during
[30:29] this. I use them all the time. Um,
[30:32] remember when Musk bought a power plant?
[30:36] Semi analysis was the first to say they
[30:38] just bought a power plant from overseas
[30:40] and are shipping it for the US because
[30:42] they couldn't get a new one built in
[30:43] time. They're doing all this crazy
[30:46] to get the compute. So again,
[30:49] very different time period. Is Colossus
[30:51] the largest? I'm just going through this
[30:52] so you guys can get the numbers on it
[30:54] and just see the scale and the size of
[30:56] this of what he's doing. Does he mention
[30:58] what can be done to speed up power
[31:00] generation? I'm only bringing this up
[31:01] because again, if you want to find
[31:04] investments, this is where you want to
[31:06] spend time on. Okay, I'll eventually put
[31:09] together a list as I've done before. Um,
[31:12] this is what I included. BYOG and behind
[31:14] the meter power. If you didn't read this
[31:16] week, the physical uh upgrade, I go
[31:18] through a bunch of places in there where
[31:20] you guys should be focusing your
[31:21] attention. edge devices for the computer
[31:24] and and phone and auto upgrade cycle on
[31:29] premise is going to be a big theme this
[31:30] year because you can't depend on the
[31:32] cloud. So every major big company is
[31:35] going to have to have the AI with inside
[31:36] their walls to deal with the amount of
[31:38] bandwidth that's going to happen but
[31:40] also cyber and everything else. This is
[31:42] the paper I'm working on for next week
[31:43] just to show that I'm focusing my
[31:45] attention on where how you can make
[31:47] money on bring your own generation and
[31:49] just look this is what happened again.
[31:51] And look at when this started.
[31:53] I remember having conversations in over
[31:55] the summertime about Bloom Energy. And
[31:57] at that point, most energy people that I
[31:59] talked to thought it was kind of a scam
[32:00] and magic and, you know, Jack and the
[32:03] Beanstalk beans stuff. Again, I think
[32:07] the only reason people thought that is
[32:09] because they didn't recognize the AI
[32:11] problem and the fact that you needed to
[32:13] have your own generation. And this is
[32:14] not a h ah, you know, we'll do we'll
[32:16] wait for the data center to come on.
[32:18] That's the whole point. you need. The
[32:20] reason the US is being so aggressive in
[32:22] terms of doing things around the glo
[32:23] geopolitically is because there's no
[32:25] time left every day in AI is like a
[32:29] year. And that's the part you have to
[32:32] start getting in your mind is the reason
[32:33] the charts look like this is because the
[32:35] demand is insatiable and the need is
[32:38] insatiable and the speed is insatiable.
[32:42] Just another one gigawatt. You're going
[32:44] to have more of these um these stories
[32:47] out there. I have these set up on
[32:48] ChatgPT Pulse, so I get every single one
[32:51] sent to me. If you haven't used Pulse
[32:53] before, I would start using it and just
[32:55] telling it what you care about. Chevron,
[32:57] I wrote a piece about Chevron. Um,
[33:00] people reached out and said they didn't
[33:01] see it. And I realized I got to make
[33:02] sure I time these things when the actual
[33:05] care goes because everyone on the hedge
[33:07] fund world is very much about what's
[33:09] going to make my make me money the next
[33:10] month. Uh, Chevron's going to be a great
[33:12] stock this year. I'm not going to read
[33:14] you all the stuff, but let's just say
[33:15] this is the critical point. AI companies
[33:17] are going faster and faster to deploy
[33:20] this. They're building bigger and bigger
[33:22] data centers. If you're going to do a 1
[33:25] gigawatt data center, you need the help
[33:28] of massive companies. You cannot go farm
[33:30] this out to Bloom Energy with that
[33:33] rally. Before that rally, it was less
[33:35] than a $10 billion market cap company.
[33:38] You can go through, you know, they can't
[33:40] do that much. Well, when you start
[33:42] getting into one gigawatt, 2 gawatt, 3
[33:44] gawatt big data center things, who are
[33:47] you going to bring in? Well, the energy
[33:50] sector is not that big. And that means
[33:51] Exxon and Chevron are going to be part
[33:53] of this. And plus, you need the natural
[33:55] gas. They own the natural gas. So,
[33:58] again, whether you've done the work on
[34:00] this, whether you think of them as an
[34:01] old school company, commodities are
[34:04] going higher. We're at the beginning of
[34:05] a commodity cycle. Go look at all the
[34:06] energy names. Energy and basic materials
[34:09] will be the two perform best performing
[34:11] sectors in my opinion this year. Chevron
[34:14] been consolidating for a while, but more
[34:16] importantly, since fracking became a
[34:19] thing, we went sideways. We did nothing.
[34:21] Now we've got another again. I'm telling
[34:23] you, this is the beginning of something
[34:24] different. We are short power in the
[34:27] world and however we have to get it,
[34:28] generators, whatever. We are not waiting
[34:31] and we are price insensitive. Price
[34:34] insensitive. That's the thing you have
[34:35] to go. That has never been the case
[34:37] before. But this is controlled by a few
[34:39] people and their cost. They're selling
[34:42] something distributed to 8 billion
[34:44] people. It's a odd lot in the cost. I'll
[34:46] show how this is for silver as we go.
[34:48] Exxon just about to make new all-time
[34:51] highs. Very close at this point. The XLE
[34:54] consolidating here. These are John Rog
[34:56] specialties. If he's not already
[34:58] positive on them right now for the
[35:00] breakout, it will come. Copper and
[35:01] silver play critical roles in AI data
[35:03] center infrastructure, but they
[35:05] represent small percentages of total
[35:07] costs. typically low single digits for
[35:10] copper and negligible for silver. The
[35:13] reason this is important is whether this
[35:15] price is 12,000 or 100,000 it is not
[35:17] going to change the copper and silver
[35:20] needs. That's the thing you have to go
[35:22] and that's why it's price inelastic. We
[35:24] haven't been in this situation. They
[35:25] need what they can get. Now the metals
[35:27] are leading energy. So here's the CRB
[35:30] raw industrials. Look at how good this
[35:32] looks. I've shown this thing many times.
[35:34] I've shown John Ro showing other
[35:36] versions of it in terms of the metals.
[35:37] So this is the raw industrial
[35:39] commodities. Just nice, beautiful
[35:41] rounding bottom. When these things tend
[35:43] to go, they go. Here's the overlay with
[35:46] the six contract of crude. Obviously,
[35:48] we've been breaking away over the course
[35:50] of the last two years. And again, they
[35:52] were all lined up together. And then
[35:53] chat GBT comes out. We have the metals
[35:57] kind of sitting here. This goes down.
[35:59] Everyone knows that right now there's a
[36:01] shortage in metals relative to the
[36:03] needs. And so there's hoarding going on.
[36:05] Gavin Baker put this out on December
[36:08] 2nd. The next 12 months might be
[36:10] different as the GB300 which is
[36:12] Blackwell finally ramps and then Reuben.
[36:15] I want you to pay attention to this.
[36:16] This is critical. If you don't
[36:18] understand what this means, this is
[36:20] really important. If we don't have any
[36:22] more data centers built this year other
[36:24] than the ones that are near finished,
[36:26] like the Colossus Next Stage, the
[36:28] Blackwells will be ramping. Right now,
[36:31] hoppers make up about 80% of the compute
[36:36] from the data centers. Now, Hopper is a
[36:40] far less efficient chip than Blackwell.
[36:43] So, these are different generations of
[36:44] chips from Nvidia. So, if you remember
[36:47] DeepS last year, what we're saying is
[36:50] these are ramping now, which are about a
[36:52] 50%. So if you have equivalent of 10
[36:55] gigawatts with hoppers in them, you're
[36:57] now going to basically be equivalent to
[36:58] 5 gawatts of use. After that, it takes
[37:02] it down again and you're down to about 1
[37:04] gawatt with Reuben. That's how we're
[37:07] going to be able to continue to have the
[37:08] AI progress as the chip efficiency
[37:11] comes. So the last three years or two
[37:13] years while we waited for Blackwell, it
[37:15] didn't come from chip uh improvements.
[37:18] It came from algorithmic improvements
[37:20] like reinforcement learning, all the
[37:22] things we heard from Deep Seek. Well,
[37:23] now we're getting into the chip side.
[37:24] This is why the US can compete with
[37:26] China regardless of the fears that are
[37:28] happening because we're getting
[37:29] efficiency gains. And this week at CES,
[37:33] we heard that the Reuben is going to be
[37:34] the second half of the year. Models
[37:36] trained and then inferenced on these
[37:38] GPUs are likely to show a dramatic leap
[37:40] in capability. I think we see first
[37:42] significant AI scientific breakthrough
[37:44] and the first economically useful
[37:46] agents. That's the thing you have to
[37:48] realize. You if if you're waiting to get
[37:51] from from New York to uh uh to Boston on
[37:55] a train in an hour, the only way you can
[37:58] do that is to have an improvement in the
[37:59] tracks and improvement on all. That's
[38:01] what we're getting now. We're finally
[38:02] going to be able to move faster, which
[38:03] means we'll be able to have AI agents
[38:05] roll out, which will speed up the
[38:06] adoption side. This is what he talked
[38:08] about. And it's cheaper. So inference
[38:10] gets cheaper and it's efficient. It's a
[38:12] mega news story. It's really important.
[38:14] it continues to get through the fact
[38:16] that the next generation will allow us
[38:19] to do more. You'll have more
[38:21] breakthroughs. And that's what he talks
[38:22] about. The breakthroughs across
[38:24] healthcare, climate science, robotics,
[38:25] embodied intelligence, and autonomous
[38:27] driving are going to accelerate. This is
[38:28] why when you hear recursive
[38:30] self-improvement, you hear AGI. We can't
[38:32] get any of this without the models
[38:34] getting better. The models can get
[38:35] better through more power with older
[38:38] chips, which is what Elon Musk has been
[38:40] doing. But then when you say more power
[38:42] meaning more data centers, but then you
[38:44] also get the chips now you're really
[38:46] accelerating things and you can see
[38:47] where the compounding gets. So, I just
[38:50] went to saying that hoppers 10 gawatts
[38:52] now 50 megawws in terms of what it would
[38:54] be with Reuben. And then if you guys
[38:57] haven't done this and you want to go do
[38:59] the numbers in AI and then you're like,
[39:00] "Hey, take these numbers. Is this the
[39:01] right number?" Bring it to um Gemini.
[39:04] That's what this is. Gemini is my
[39:06] academic. I find it to be reminding me
[39:08] of school. I don't like to use it for
[39:09] general questions. I love to use it for
[39:12] correcting stuff or fact-checking. Um
[39:15] DeepSeek comes out, market freaks out.
[39:17] Microsoft drops some AI data center. TD
[39:20] Cowan market freaks out. Microsoft
[39:21] abandons data center. This was in April.
[39:23] Freaks out. Oracle delays some data.
[39:26] Remember all these things last year?
[39:28] Every time they happened, there was a
[39:29] fall-off in that's going to happen all
[39:32] next year. This this year, this is going
[39:34] to start being a real thing. You're
[39:36] going to have delays. You're going to
[39:37] have all these things show up. And you
[39:39] had Baird come out. And I don't I don't
[39:41] know anything about the person. All I
[39:42] know is that they started coverage, I
[39:44] think, in 24. They raised their price
[39:47] target up to 816 not too long ago and
[39:50] now they just cut it and again growing
[39:53] investor worries about potential power
[39:54] capacity over supplies. It's not an over
[39:56] supply issue but that fear is going to
[39:59] be there. Mark Andre 2026 outlook cannot
[40:02] highlight enough why this one is also
[40:04] what you need to do. I broke this down
[40:06] very simply. AI is already monetizing
[40:08] faster than any in history. If you guys
[40:09] are worried about the revenues coming
[40:11] in,
[40:12] listen to who you're reading and then
[40:14] listen to Mark Andre. This new wave of
[40:16] AI companies is growing revenue like
[40:18] actual customer revenue. Real demand
[40:20] translated into dollars showing up in
[40:22] banking accounts at an absolutely
[40:23] unprecedented takeoff rate. Again, I I
[40:26] this is why I find it ridiculous when
[40:28] people regurgitate to me that there'll
[40:30] be no revenues. there is going to be
[40:32] trouble for the hyperscalers to get
[40:33] enough revenues relative to the capex in
[40:37] the time period because the capex is
[40:38] going to come first. The bottlenecks are
[40:40] there. If they can't get the revenues
[40:41] because they can't build the capacity
[40:43] like Google and Amazon for the cloud,
[40:47] they're just not going to be able to get
[40:48] the revenues as fast as what people are
[40:49] going to want. It's not a collapse in
[40:51] the stock, but it is probably multiple
[40:53] compression. It is probably continued
[40:55] fears. We're still early in the AI
[40:56] cycle. AI is bigger than the internet.
[40:58] Think electricity or microprocessor.
[41:01] He says it's the biggest investment
[41:03] opportunity of his of his career.
[41:06] The price of AI will collapse and demand
[41:09] will explode. This is the Jevans paradox
[41:11] thing. This is what we're seeing and
[41:13] this is the case. Once AI capability is
[41:14] proven, others catch up shockingly fast.
[41:17] Once somebody proves something is
[41:18] capable, it doesn't seem to be that hard
[41:20] for other people to catch up. This is
[41:22] the no chance of a moat. You can do it
[41:26] with fewer people. This is my whole
[41:27] argument as to why eventually AI
[41:29] cannibalizes everything and capitalism
[41:31] is in a very late stage. Public fear,
[41:35] actual behavior is different. AI is
[41:38] early, monetizing faster than any prior
[41:39] tech wave, deflating in price, expanding
[41:41] in demand, geopolitically competitive,
[41:43] and structurally favors portfolios over
[41:45] single single monopolies. Um, on the
[41:48] payw wall, this is something else I'm
[41:50] going to do because I've had requests
[41:51] from people. Every time I show one of
[41:52] these, it kind of forces you guys to
[41:55] just watch the video, pause it, take
[41:56] things out. What I am going to do is put
[41:58] on the payw wall the name of the podcast
[42:00] that I think are the most important, the
[42:02] link if you want to go to it, a sum, a
[42:04] three paragraph summary, and then the
[42:06] timestamps underneath. So that way you
[42:08] guys can just go to that site, you can
[42:10] get them all once it's launched.
[42:11] Everything that I show in the video for
[42:13] that week, plus somes that that I don't
[42:15] go, you'll be able to do. Uh you guys
[42:17] realize that all software is about to be
[42:19] free, right? and software is presently
[42:21] most valuable capital asset. So two
[42:22] points in this. I completely agree and I
[42:24] still debate people that Adobe
[42:26] Salesforce.com they have moes they're
[42:28] blah guys I'm I'm telling you I I just I
[42:32] don't agree. I think you're you're
[42:34] trying to pick a bottom in something
[42:35] that could bounce for three months. It
[42:37] could be a shift. You could get short
[42:39] covering as people delever. But in terms
[42:41] of businesses and what they're fighting
[42:43] against in software, I I can build stuff
[42:45] so quickly. It's just not it's just not
[42:48] there. And then on this point and
[42:49] software software is arguably 30 to 40%
[42:54] of all the benchmarks that you're
[42:56] invested in.
[42:58] That is a massive problem. So again, I
[43:00] think the S&P is going to be up 15%. I
[43:02] think the software companies, let's
[43:05] assume they're flat.
[43:07] 15% growth in the S&P while the top 40%
[43:10] is flat, just as a general thing, means
[43:12] the bottom 60 have to be up a lot. So
[43:15] just keep that in mind because if we get
[43:17] that kind of a dispersion when everyone
[43:19] is positioned in this and short the
[43:21] other names or underweight the other
[43:22] names, it is a massive change. And if
[43:24] you're a mutual fund, what's the size of
[43:26] your growth funds? What's the size of
[43:27] your value funds? What's the size of
[43:29] your commodity fund? What's your sign of
[43:30] your tech fund? This is going to be a
[43:32] massive rotation in my opinion. And Alon
[43:34] Mus seems to agree.
[43:37] I wrote this this week. Um I enjoyed
[43:39] writing this one. I wasn't planning on
[43:41] it. Opus 4.5 inflection point. So, I
[43:44] showed last week what I was able to do
[43:46] with it. I highlighted in here between
[43:48] December 26th and January 3rd, the span
[43:51] of a single holiday week, while most
[43:52] people were not paying attention to the
[43:53] markets, an XAI co-founder, an anthropic
[43:56] AGI researcher, a Google principal
[43:57] engineer, a formal Google Gemini lead,
[44:00] and the founder of Midjourney, one of
[44:02] the the the the image places, all
[44:05] independently concluded that Opus 4 and
[44:07] 5 plus Claude Code had fundamentally
[44:09] changed what was possible. This wasn't
[44:12] coordinated marketing and orchestrated
[44:13] hype. It was organic recognition from
[44:16] technical leaders who suddenly found
[44:18] themselves experience the capabilities
[44:20] they hadn't anticipated arriving for
[44:22] months or years. It was a massive
[44:24] moment. I felt it. I've been talking
[44:26] about how I've seen everything change
[44:28] since the summer, but this was
[44:29] breathtaking. This came out uh
[44:32] yesterday. Cloud code is changing how
[44:34] people hire and work. The transformation
[44:36] cloud code is making cannot be un
[44:38] understated. This isn't a fun tool for
[44:40] devs. This is an entirely new way to use
[44:42] computers, do work, market, sell, and so
[44:44] much more. AI agents,
[44:48] Claudecomb is a gamecher.
[44:51] Here's mine. For those of you who've
[44:53] ever walked by a desk of one of your
[44:54] data scientists or one of your codes and
[44:56] you see all this this stuff on all of
[44:58] the screens, I just want you to look.
[45:01] This is me and this is the output I
[45:04] built in terms of it's 800 lines of
[45:07] code. It took me to get this took me
[45:11] seven minutes.
[45:13] And I'm not kidding you. Seven minutes
[45:15] for me to verbally say what I wanted and
[45:18] then go through a few more iterations.
[45:20] And it took me a total of three hours to
[45:22] get to the point where I had this
[45:24] something I wanted to build for, as I
[45:26] wrote in my paper this week, a decade.
[45:29] This is my turbulence model. I wanted an
[45:31] immune system to warn me, to give me
[45:33] something that basically was like a card
[45:35] counting thing of just like, okay, the
[45:37] cards are no longer in your favor.
[45:39] Doesn't mean you're not going to win the
[45:40] next three hands. It doesn't mean the
[45:41] market's going to go not going to go up,
[45:43] but it does mean you should be paying
[45:44] attention for other metrics. So, you
[45:47] want a warning system that kind of says,
[45:48] "Hey, there's an increased probability
[45:50] of something happening." And the way the
[45:52] model is set up is it only gives these
[45:54] signals when we're trending above the
[45:55] 50-day moving average. So, you could add
[45:58] other things into it. you can use it
[45:59] yourself off the the payw wall. Uh but I
[46:02] just wanted to bring that up. Alon Mus
[46:04] said it best. Singularity has already
[46:06] arrived. It's the year of singularity.
[46:07] Singularity basically means that AI
[46:09] agents recursive learning, AGI, it's
[46:11] already here. That means the world has
[46:13] changed. Now, a visual for people that
[46:15] have had trouble keeping up with
[46:16] everything that I went through here. AI
[46:18] model demand just continues to increase
[46:21] exponentially. Chip efficiency and
[46:23] adoption continues to go. The problem is
[46:28] we've got issues here. So the
[46:29] hyperscaler expansion in data center
[46:31] upgrades, they need to increase the data
[46:33] centers to be able to allow more users.
[46:35] We don't have that. We've got a power
[46:36] and grid problem. We have a commodity
[46:38] supply ch problem. We have a memory
[46:40] supply problem. We're seeing it with
[46:42] SKHEX with all we have issues that won't
[46:45] allow us. The demand is there. It's the
[46:47] underinvestment in this side that's
[46:49] going to take time. Memory is going to
[46:51] take a long time to catch up. I had this
[46:54] discussion with people. I said
[46:55] repeatedly in the summer that said we
[46:57] were going to have a glut of memory.
[46:59] Micron is now up from 150 to 350 during
[47:02] that time period. I showed you Western
[47:04] Digital. There are tremendous
[47:06] opportunities if you can find the places
[47:08] where demand is massively above supply.
[47:11] Why now in terms of energy? Because AI
[47:13] demand is accelerating because adoption
[47:15] is finally beginning.
[47:18] It will only increase every single day
[47:19] from here. Computer is no longer the
[47:21] bottleneck. Okay, good. So, we have the
[47:23] chips. Efficiency gains paradoxically
[47:25] increase total energy demand. This is
[47:26] Jevans paradox. We're making faster
[47:28] chips, cheaper. It allows people to do
[47:30] things. It increases the energy demand
[47:32] because you get more and more users. AI
[47:34] agents are dramatically more. Ask
[47:36] yourself, how many hours a day to use
[47:39] artificial intelligence? I would say the
[47:41] average person is still less than an
[47:42] hour a day. An AI agent will be working
[47:44] 24 hours, 7 days a week. So every
[47:47] digital employee and AI agent that gets
[47:48] hired this year of which there will be
[47:50] millions is going to be working far more
[47:53] than an individual. Gets back to the
[47:54] productivity side, gets back to the GDP
[47:56] side. Data centers are no longer 20 to
[47:58] 50 megawatt facilities. This is a
[48:00] critical one too because the
[48:01] infrastructures entered the scale phase.
[48:03] We need bigger ones. We proved the
[48:04] point. That was not the case two years
[48:06] ago. We now need multi-gawatt AI
[48:09] factories. This change in both size and
[48:12] the speed requirements means energy is
[48:14] now an issue. The grid has run out of
[48:16] slack again. We've been using excess
[48:18] capacity. Now we're running out. Energy
[48:21] has become the binding and persistent
[48:22] constraint. This forces a structural
[48:25] shift in energy sourcing. Hyperscalers
[48:27] can no longer rely solely on the grid.
[48:29] Bring your own generation moves from
[48:30] optional to mandatory. Pulling AI
[48:32] directly into the domain of industrial
[48:35] scale energy production, fuel,
[48:37] logistics, and infrastructure buildout.
[48:38] Go through the entire uh chain. We need
[48:41] more energy and we will forever need
[48:43] energy until we can do this all off the
[48:45] sun, which isn't coming anytime soon.
[48:47] The long power problem, a structural
[48:49] investment regime. We're at the
[48:51] beginning of a multi-deade power
[48:52] constraint. Efficiency gains are not
[48:55] reducing demand. They're accelerating
[48:57] it. Again, think about it. In between
[49:00] power, which normally historically if
[49:02] power prices went higher, we would have
[49:05] an issue and demand would go down. We're
[49:07] getting efficiency gains and we're
[49:09] getting a reduction in price on the
[49:11] things that are driving the demand. This
[49:13] is a very unique situation. You have to
[49:15] pay attention to you have to think
[49:16] outside the box and I understand no
[49:18] energy person understands this. I also
[49:20] understand no silo based person
[49:22] understands this. So it just is what it
[49:24] is. You can spend time with me if you
[49:26] want going through it but I try to make
[49:28] these as easy as possible. AI adoption
[49:30] is still early. I would say in the 1%
[49:32] we're in the 1% out of 100% eventually
[49:35] getting there right now. Power becomes
[49:37] the persistent bottleneck. It's not a
[49:38] transient one. It is forever
[49:41] for portfolios this year. You want to be
[49:43] long AI adopters. You want to be long
[49:45] healthcare companies. You want to be
[49:46] long insurance companies. You want to be
[49:48] long financial companies. You want to be
[49:49] long CH Robinson. Basically, you can
[49:52] find individual companies that are
[49:53] adopting Goldman Sachs, Morgan Stanley,
[49:55] all of the big banks, not just the
[49:58] enablers, not the ones giving it
[50:00] anymore. belong power providers and
[50:02] infrastructure. Be long commodities tied
[50:03] to the physical upgrade. Be cautious or
[50:06] short. Business is built primarily on
[50:08] code. Do your homework on anything
[50:09] you're invested on the software side
[50:11] that can be replicated very simply with
[50:13] what I showed you which now takes
[50:15] seconds. Because if something takes
[50:17] seven minutes to for me to build, it's
[50:20] seconds. For 2026, be wary of high
[50:22] multiple data center buildouts. Delays
[50:25] will cause multiple compression. Now,
[50:28] this is the bubble. So, the bubble again
[50:31] gets back to what I've just talked
[50:33] about. The top 10 companies have the
[50:34] highest weight in the index since 1929
[50:37] relative to the 75th percentile stock. I
[50:40] want you to pay attention to these
[50:41] points in time. So, this is just before
[50:44] 2000 and this is just before 1972.
[50:50] The reason I showed you those there,
[50:52] those inflection points of when we had
[50:54] concentration, it started to change.
[50:57] One of them lines up here. We've seen
[50:59] correlation as the concentration has
[51:02] risen, meaning the ability to make money
[51:04] has been driven by less and less things.
[51:06] The correlation with inside a
[51:08] hypothetical multiasset portfolio,
[51:10] you're either a winner or a loser and
[51:12] that's it. There's not enough breath
[51:13] going on. There's basically been a
[51:15] one-trick pony. The one trick pony since
[51:17] 2009 was exponential innovation. It was
[51:19] software. Software dominated. We are at
[51:22] the end of software. You're going to see
[51:23] a change in multiasset po portfolio. If
[51:26] I'm right and correlations go down,
[51:30] I think you're going to see a dramatic
[51:32] shift in leverage across the system.
[51:33] That's my big theme for this year is
[51:35] that leverage relative to the liquidity
[51:38] that's available for something like the
[51:40] Russell 2000. The reason I think it'll
[51:42] be up 60% is because every single
[51:44] portfolio manager on the planet's
[51:46] underweighted or short it even if it's
[51:48] just not through ETFs, it's through
[51:51] small cap names, it's through
[51:52] everything. These points in time are
[51:55] back to that graph of concentration. We
[51:57] were highly concentrated here. We were
[52:00] highly concentrated here. This is the
[52:02] log rhythmic scale of the Goldman Sachs
[52:07] commodity index or what was the Goldman
[52:09] Sachs commodity index. When commodities
[52:12] rise,
[52:13] it's a problem for concentration because
[52:16] historically when there's no movement in
[52:18] energy or commodities, technology wins.
[52:23] the ability for a very few amount of
[52:25] companies to win wins. But when energy
[52:27] becomes the dominant force again, a lot
[52:30] of a lot of places are needed. This is
[52:32] the broadening out of the market. This
[52:34] is the PMI. This is the MAG7 relative to
[52:37] the Russell 2000.
[52:40] We've stayed above the 200 day moving
[52:42] average for a while. We held it there in
[52:44] liberation day. It was a great buy.
[52:46] Every freakout we've had, I would watch
[52:48] this chart closely this year. It peaked
[52:51] right around the time early September
[52:52] when we started seeing micron change,
[52:55] silver ch, all of the PMI related things
[52:57] started to change and all of a sudden
[53:00] small cap started outperforming the mag
[53:01] 7. I think this is going down a lot.
[53:03] Look where it started
[53:06] right around chat GBT. That was that
[53:08] second thing. Here's another chat GBT
[53:10] one. This is the Goldman Sachs power
[53:13] uh customized basket. If you guys are
[53:15] long this, which I know a lot of you
[53:17] are, this is it relative to the S&P.
[53:19] Again, very similar since 2022. You
[53:22] wanted to be in the power names. This is
[53:24] going to have all of the names you guys
[53:27] have probably been in. Another way to
[53:29] play it, Nvidia verse the XSD, equal
[53:33] weight semiconductors. So, I want to be
[53:34] long equal weight semis. That's why I
[53:36] show you equal weight semis all the
[53:38] time. That's why I show you equal weight
[53:39] software. I don't think Nvidia is going
[53:41] down. Just like I'd be surprised if you
[53:44] know the Mag 7 got hit very hard. I
[53:46] don't think they're going down. I think
[53:48] Nvidia is cheap, but I do think Nvidia
[53:50] relative to the other names is an
[53:53] exposure because since chat GPT, this
[53:55] was about the data center buildout. Now
[53:57] we're talking about putting this in edge
[53:58] devices. I want to be long semis in
[54:00] cars, semis in phones, semis in uh in
[54:03] computers. I just recently got myself a
[54:06] new Mac. That's what I'm working on. I
[54:08] got myself a new iPhone for the first
[54:10] time in a long time. game changer for me
[54:12] in terms of its ability to do LLMs
[54:14] faster than what I do on my ThinkPad,
[54:16] souped-up ThinkPad. Uh I bring up the
[54:18] Nvidia verse uh XSD because I showed
[54:23] this last week. This is what I wrote at
[54:24] the end of the year. Why this deal
[54:26] mattered and how it basically says the
[54:28] architectural revolution has is now
[54:31] required over Moore's law. It's no
[54:33] longer about the compute itself. It's
[54:36] about the packaging. So here are the
[54:38] names that have happened so far this
[54:39] year. Nvidia unchanged,
[54:46] Broadcom unchanged.
[54:48] All of these names are up double digits
[54:51] and you guys can go through. This was
[54:53] the complete list of the names that I
[54:55] went that would benefit from the
[54:57] advanced packaging side. The names that
[54:59] I would want to be long this year, which
[55:01] didn't include Nvidia. It does have uh
[55:04] Broadcom because they benefit from this
[55:05] part. But as you go through it, this is
[55:07] the way I'll be doing things. And again,
[55:09] John Rog, we'll put the technical score
[55:11] if you know his work 0 to four, four
[55:14] being the best. One of the names that I
[55:16] love this year that is a bad one is
[55:19] synopsis. I'm not going to go through
[55:21] details why. You guys can do your own
[55:22] work. The commentary score, what the
[55:24] earnings commentary have been. I have an
[55:26] algorithm or prompt that will basically
[55:28] go through the last three uh earnings
[55:30] reports and basically look for
[55:31] progressions and go through it related
[55:33] to the theme that I'm doing. And then
[55:34] estimate revisions. because obviously
[55:36] all the estimate revisions are going
[55:37] higher almost across the board right
[55:39] now. But again, if you go through the
[55:40] names, you can go through and figure
[55:42] them out now and then go look and ask,
[55:44] "Hey, if I were you guys, just take a
[55:47] snapshot of this, put it into your LLM.
[55:49] How do these names benefit from advanced
[55:51] packaging? You'll get details on all of
[55:53] them. You don't need me to go through
[55:54] it. But if you guys want to figure out
[55:56] how I can do that, go from a podcast to
[55:59] those names. all the prompts and the
[56:01] algorithms. Call up 22V and you guys can
[56:03] figure out in terms of how I can do that
[56:05] presentation for you. Um, I wrote this
[56:07] on Bitcoin. I'm gonna finish up this
[56:09] week with Bitcoin. Again, for those of
[56:11] you frustrated by Bitcoin and going
[56:12] through it, I will continue to say
[56:14] Bitcoin in my opinion based on
[56:16] everything I showed you today based on
[56:17] what I showed last week will be the best
[56:19] performing asset. It is currently held
[56:21] down by the technical levels. Technicals
[56:23] matter a lot for it until we start
[56:25] seeing it break out, which it did
[56:27] finally accomplish something. thing I
[56:28] highlighted last week we had closed
[56:30] below the 50-day moving average for a
[56:33] record time period or one of the longest
[56:35] in terms of the last 5 years. The last
[56:37] one we saw in a similar one was here and
[56:39] then we eventually broke out. We did
[56:41] have a crossover point in terms of the
[56:43] 20-day moving average getting above the
[56:44] 50. We've continued to make higher highs
[56:46] and higher lows. As long as we can keep
[56:48] this up, we're going to get up here.
[56:51] Then we'll spend some more time and then
[56:53] I would think we would eventually break
[56:55] out once we get back above this trend
[56:57] line and back above the 200 day moving
[56:59] average. They're probably going to meet
[57:00] before that happens. Uh I think this
[57:03] will soar for the rest of the year. We
[57:05] have MACD weeks that are down here. Only
[57:08] reason I'm showing this again doesn't
[57:09] mean we're at a breakout point yet. The
[57:11] technical chart is still brutal. When
[57:13] the technical chart is brutal, I found
[57:15] in in Bitcoin and all crypto, there's a
[57:17] lot of traders. They're good traders.
[57:18] They play from both sides. They're
[57:20] shorting it. For people who only buy it,
[57:22] that's great. They're waiting for this
[57:23] to go higher, but the dominant force
[57:25] right now is people selling into
[57:26] strength. The OGs are still selling.
[57:28] There seems to be buying underneath. But
[57:30] when we get the breakout and we get back
[57:32] above 100,000 and I've said, and someone
[57:34] called me this week, I've said on pump,
[57:37] I've said it on here, three daily closes
[57:39] above 92,000 in a row. And I believe the
[57:42] bottom is in. We did two. A bunch of
[57:45] people called up and said, "I hope we're
[57:46] going to get three. We didn't get three.
[57:48] I still think we're going to get there
[57:49] and go through it. I just want to
[57:50] highlight that one of the reasons that I
[57:52] think this will be the best performing
[57:53] stock is because right now it is being
[57:55] impacted by the Nvidia, the Mag 7 stuff,
[57:58] all the things I mentioned because
[57:59] people still view this. I hear this even
[58:01] from people who are focused and do
[58:04] weekly macro podcasts that involve
[58:06] Bitcoin where they're like, "Yeah, well,
[58:07] it's just the NASDAQ." It is not just
[58:09] the NASDAQ, guys. I can't believe I
[58:11] listened to this. But at some point,
[58:13] people have to understand why I wrote
[58:14] the piece that I wrote, that I just
[58:16] showed you guys, and why I've said
[58:17] repeatedly, Bitcoin is the only pure AI
[58:20] trade. You guys want it to be relative
[58:22] to now. Well, what you're going to see
[58:23] this year is the AI trade is power. The
[58:26] AI trade has nothing to do with
[58:28] technology. AI is not technology. A pure
[58:31] AI trade means when does AI start
[58:34] destroying all the tech companies and
[58:36] what actually doesn't need a moat in its
[58:39] business? Bitcoin doesn't need a moat.
[58:41] This will be a story by the end of this
[58:42] year and I promise you if you've never
[58:44] purchased Bitcoin and you've been
[58:45] doubting it by the end of the year you
[58:47] will start doing research. So I'm just
[58:48] going to try and help you get through it
[58:50] because you'll be staring at
[58:51] tokenization every single year, stable
[58:53] coins every single year. The banks
[58:55] offering it every single year. The next
[58:56] time Bitcoin is up at all-time highs at
[58:58] the end of the year, every single person
[59:00] that has doubted it will be questioning
[59:01] themselves. They'll never admit they
[59:03] bought it. They won't tell anyone, but
[59:05] they will.
[59:07] Here's the chart that is the most
[59:09] positive right now. Ethereum relative to
[59:11] Bitcoin.
[59:13] If we if I was worried right now, if I
[59:15] thought we were going to go
[59:16] significantly lower, I would think this
[59:18] would break down. I think this is like
[59:20] NASDAQ over Q's. I'm sorry, Q's over
[59:22] S&P, Q's over spies, this is the growth
[59:25] thing for it. This is the thing that
[59:27] matters. As long as Ethereum keeps
[59:29] hanging in there relative to Bitcoin,
[59:31] then for me, I think eventually these
[59:33] will all go up together on the PMI
[59:35] cycle. Final thing I just want to remind
[59:38] people because we're at this stage. The
[59:40] money is not the problem. AI is the new
[59:42] global arms race and capex will
[59:44] eventually be funded by governments. If
[59:46] you want to know why gold, silver,
[59:48] Bitcoin is soaring, it's the debasement
[59:50] to fund the AI arms race. But you can't
[59:52] print energy. Alon Mus that is why
[59:55] Bitcoin is based on energy. You can't
[59:57] issue fake fiat currency
[60:00] and every government in history has done
[60:02] so. But it is impossible to fake energy.
[60:05] Elon Musk has forever equated Bitcoin to
[60:09] energy. That is the reason why you need
[60:12] to do your homework because everything
[60:14] that was up here was on October 13th, 3
[60:16] days before we had an event where Trump
[60:19] came out said this. We ended up losing
[60:21] market makers in crypto and crypto's
[60:23] been kind of in a bare market phase
[60:24] since then, which people thought was
[60:25] some other magical thing. The reality is
[60:28] the next time we go higher and people
[60:29] realize that Bitcoin is energy. It's
[60:32] based on energy and energy is the new
[60:34] currency, I think the future currency
[60:36] will essentially just be wattage. That's
[60:39] it for me this week. Hopefully I
[60:40] provoked some thought. Reach out to 22V.
[60:43] I promise you I will get the payw wall
[60:45] up as soon as possible. Um I'm going to
[60:47] keep creating things. We're going to
[60:49] keep working through this. But the main
[60:50] thing for this year is if you've been
[60:52] spending the time putting AI in a
[60:53] bubble, start reaching out. Start
[60:55] finding ways
[60:57] to get me more involved in how you're
[60:59] doing things. Even if it's just a
[61:01] one-time show. It's a big year for AI
[61:03] and if you don't do something this year,
[61:05] I think your business, your children,
[61:06] the whole thing, you're going to fall
[61:08] way behind because it accelerating too
[61:10] fast. It is time to spend the time on
[61:12] it. I'll also release my HRV post this
[61:15] week. And just as a reminder for those
[61:17] of you in Miami or those of you in
[61:19] Florida who've been reaching out to me,
[61:21] Pomp and I will be doing a live
[61:22] broadcast from Real Vision. I'll be down
[61:24] there for at least four days. If you
[61:27] guys are covered by 22V down there and
[61:28] you want to see me, I'm going to be in
[61:29] the Miami area for a while. You can
[61:31] reach out to the sales force at 22V.
[61:33] Happy to spend some time with people who
[61:35] want to get to know me. All the best,
[61:37] guys. See you.