YT YouTube2Text

YouTube2Text → Transcripts → Situational Awareness, Hyperscaler Earnings and Warsh: A Bottom for AI Stocks?

Situational Awareness, Hyperscaler Earnings and Warsh: A Bottom for AI Stocks? — Transcript

by Jordi Visser · 11,231 words · 1,754 segments · language en · Watch on YouTube

Open in transcriber Summarize with AI Download SRT/VTT

Full transcript

0:01All right, continue through the summer
0:03and end a uh
0:06a quiet month of July for the S&P, uh
0:09brutal market for factors, for tech, for
0:14the AI trades,
0:16and for hedge funds. Um
0:18I'll go through all of it.
0:20Try to give you uh both some
0:23experiential perspective from having
0:25been through uh
0:26a bunch of what situational awareness
0:29went through.
0:30Uh starting early in my career. So,
0:33hopefully everything I'm going through
0:34will help to some degree.
0:36I'm going to talk a lot about is that
0:38the end of what went on and what to look
0:41for. I do think there's a high
0:43probability that this was a a cleansing
0:46event. Uh
0:48I don't see everything just racing back
0:50up, but I do think that when you hit
0:52those types of events, and I'll go
0:54through the reasons why uh
0:57it's probably more important than you
1:00realize just because of what happens
1:02when a hedge fund actually goes under of
1:04that size. Uh
1:06Number eight, compute demand is
1:08insatiable. I just want to make sure
1:10that
1:12if I'm wrong about everything that you
1:14guys watch on a weekly basis,
1:17number eight's the most important thing
1:19because I say it on every interview. I
1:21say it every week here. Compute demand
1:24is insatiable. There is no way for the
1:27supply side to keep up with the demand
1:29of Q compute currently. And there are no
1:32indications that that's case, but most
1:35of the bearish rhetoric comes from the
1:37fact that people are worried that we're
1:39overbuilding, we're not going to get the
1:40revenues. I'm going to cover all of that
1:43this week based on the earnings, based
1:45on podcasts again at a time when people
1:48are so negative on AI from so many
1:50different directions. And as someone who
1:52puts their um neck out on the line
1:55believing and using and talking about
1:58this.
1:59The sentiment is very easy for me to
2:00hear based on the emails I'm getting
2:02from subscribers. Um yes, people are
2:04losing money. I've lost money this month
2:07and and in in a big way and I my
2:08positions are reduced. So, I get the
2:11fact that when positions go against you,
2:14it's tough. But, that's the market and
2:16that's what goes on. Uh I'm not here to
2:18tell you everything is guaranteed.
2:20Everything I do is based on
2:21probabilities. The one thing that I do
2:23think is an absolute guarantee
2:26is that the market structure has changed
2:27forever and I'm going to cover that as
2:29well and I think it's important
2:30particularly for institutions,
2:32uh hedge funds, mutual funds.
2:35Let's get into it.
2:37So, this is something that I ran. Now,
2:40this was through
2:42uh
2:43Wednesday.
2:44So, this is the Goldman Sachs
2:48H or VIP Hedge Fund Index. So, they
2:52created an index
2:53of the most crowded positions or the
2:55most owned positions in the hedge fund
2:57world based on their data.
3:00I created took that index and the dot GS
3:03Hedge is relative to the S&P 500. This
3:06is a monthly lookback since the index
3:08was created, which goes all the way back
3:10to 2001.
3:14Down 12% for the month. So, that would
3:16mean that in a month that the S&P
3:18finishes unchanged. Now, again, this is
3:20through Wednesday. The S&P was down
3:21about 2% through Wednesday.
3:24Um this was by far the worst relative
3:27performance of that
3:29crowded hedge fund positioning, which
3:31again, hedge funds are smart. They're
3:33doing their homework. If it's crowded
3:35position, that means there's got to be
3:37good fundamentals to it. Uh
3:40worse than during the month of Lehman.
3:43That's what that month is there to do.
3:44And but
3:45aside from that, doubled any other
3:48month.
3:49So, this was a massive
3:52shock to the hedge fund side in terms of
3:54crowded positions.
3:56I did a rolling four-day.
3:58Seven of the 12% occurred in the four
4:01days from Friday of last week until
4:05Wednesday of this week. So, Friday,
4:07Monday, Tuesday, Wednesday, that
4:08four-day rate of change on that
4:11was the worst ever. And that includes
4:13during COVID when everything was
4:15collapsing at the market level. Again,
4:17markets going down. This is happening in
4:19a tape where the S&P
4:21is not moving and the VIX is low. This
4:23will get into the market structure side.
4:26So, we started getting these stories.
4:27This actually came out on Thursday
4:30morning. Hedge funds face demands to
4:31stump up collaterals as all stocks
4:33tumble. So, at this point
4:35news was starting to circulate that
4:39there was a problem. So, this is
4:41Thursday morning before the market. This
4:43is in the FT, so this is during European
4:46time.
4:47And clearly something is happening. And
4:49then
4:50you in the article, it says Goldman
4:52Sachs and JP Morgan were among the banks
4:54that had asked clients to stump up
4:55additional collateral, according to
4:57several people familiar with the matter.
5:01The additional collateral were
5:03automatically triggered by market
5:04volatility.
5:06Goldman said in a recent note that the
5:08build-up in gross leverage in the first
5:09five months of the year was the largest
5:11cumulative increase it had recorded
5:13since it began tracking the data in
5:142016.
5:16And this was already at all-time highs,
5:18according to data which I'll show you
5:19guys later.
5:21In a sign of the pain for hedge funds
5:24long-short hedge fund strategies were
5:26down over 1%. Multi-strats were down
5:28over 1.7.
5:29Um basically, the last time all of these
5:31strategies were down more than 1% a
5:32single day was during COVID.
5:35However, and this is probably the most
5:37important line in the first thing I want
5:40you to think about, which I did in my
5:42midweek market update this week for
5:44subscribers, which is hedge fund
5:46strategies were still on average up more
5:48than 10% for the year. Now, I don't
5:51think this was through Wednesday. I
5:52think this is through the Friday before.
5:54As I told you, it was a brutal week, but
5:57then you got the bounce back. So, figure
5:59they're down they're up less than 10,
6:01but that's still
6:03up a lot. And I talk about this often.
6:06During these situations, chip stack
6:08matters. So, when your chip stack is
6:11still positive, you're a lot more likely
6:13to just hedge your stuff up and then
6:15come back to fight another day. Now,
6:17the situational awareness Leupold
6:21>> [clears throat]
6:22>> seeks to raise capital after the AI
6:24route. So, again, this is in the FT.
6:29And we find out
6:32on Thursday at 11:30-ish that Citadel
6:35buys situational awareness a stock
6:37portfolio after big losses in AI.
6:41So,
6:42there's a lot to talk about on this. Um
6:45I just want to bring this up that
6:49this is the kind of thing that does mark
6:52bottoms normally.
6:53>> [clears throat]
6:54>> Do I think there's a lot of other funds
6:55that are
6:57possibly in a similar situation? Yes,
7:00they're probably are.
7:01The difference is, number one,
7:04this fund is very large. And number two,
7:06as I go through this, I want you to just
7:08pay attention. Citadel is obviously much
7:10larger.
7:11Situational awareness is reported to
7:14have had 45 billion worth of positions.
7:17Um I'll go through some of the details,
7:19but these are the articles or two
7:21separate articles that I just want to
7:23make sure you hear.
7:24Because based on my experience, this is
7:26the reason why I think there's a high
7:27probability that this was a market
7:29cleansing event. Even if there's, say,
7:32another 100 hedge funds that are in the
7:34same situation as Situational Awareness,
7:36but the question is, if you take 100
7:39hedge funds of a billion dollars, you go
7:41through what they've hedged up already,
7:42doesn't really matter that much and I'm
7:44going to say no.
7:46Um
7:47the hedge fund is working with prime
7:48brokers including Bank of America,
7:50Goldman Sachs, JP Morgan just to meet
7:52margin calls or wind down positions in
7:53an orderly way. Key thing in this,
7:56multiple prime brokers know what's going
7:58on.
7:59The banks had been marketing, it's a key
8:01word, marketing
8:02a portfolio of hedge funds long and
8:04short holdings before the market open on
8:06Thursday.
8:07I was involved in LTCM
8:11in a very close manner both before
8:14uh in the 6 months before, during the
8:16close down, I took over the S&P book
8:19when it wasn't even completed yet.
8:22Um I've seen this from LTCM back, but I
8:24was also very involved in the uh GARP
8:28quant unwind of 2007 before the great
8:31financial crisis and also Archegos. Um
8:34there's been other many ones, but those
8:35are the main ones. I don't consider
8:37Melvin Capital to be in the same side,
8:39but what you're dealing with is
8:41once the marketing of the portfolio
8:44begins and they start looking for buyers
8:47basically you're showing people po-
8:49people other people's poker hands and
8:51they're going going to go out there and
8:52they're going to hedge their book
8:55or front run it or do whatever.
8:57In another one, multi-strategy hedge
8:58fund firm Millennium Management also had
9:00a bid. So, when you combine
9:02Millennium and Citadel
9:05these firms both combine trillions of
9:08dollars
9:10of
9:10of positions.
9:12So, of course if they've seen what's
9:15going on and to the extent there's a
9:17variety of things that are going to
9:18happen, but the reality is this fall
9:21that we saw here, which is really those
9:23last 4 days. And the reason I have this
9:25chart up, this is Bloom Energy relative
9:27to Adobe.
9:29And effectively when you go through
9:31their positions, the biggest ones they
9:32have, the short positions that are
9:35rumored, it's is pretty good proxy for
9:37their book. The reason I bring this up
9:39is
9:40they go around shopping things cuz
9:42they've given back a sizable portion.
9:43And when you give back a sizable portion
9:45of your gains, even if they are up 80%
9:47year-to-date, with half or 3/4 of that
9:50being their private Anthropic position,
9:53you're still getting very close to the
9:55fact where you can't meet cash cuz the
9:56Anthropic's not going to go through. So,
9:58they have to find someone to
10:00buy their book and give them the cash
10:02necessary to meet the collateral calls.
10:04So, you get this massive sell-off and
10:07then
10:08immediately at that day, you've got a
10:10change that's gone on. You've had a
10:11cleansing event. So, in my mind,
10:14given the fact that Bloom Energy gave
10:15back everything that it had to the point
10:18where we got the last 13F holdings,
10:20which is right here, which is where we
10:23saw Situational Awareness's positions,
10:25I do think that you have to start from
10:28the perspective of is the unwind over?
10:31It's definitely not over from people
10:33that need to get out. They never are,
10:35but the question is are there more
10:36people in a position now that have
10:38already hedged their portfolio,
10:40especially when you're dealing with the
10:41prime brokers and the largest banks out
10:43there who have been the banks were
10:45already out there asking for more
10:46collateral, and then you have the two
10:48biggest hedge funds or two of the
10:50biggest hedge funds in the world
10:52also seeing the situation and having
10:54positions on internally as well. So,
10:57I did this during the week and I just
10:58want to bring this up. This was an
10:59especially important risk management
11:01week with the Federal Reserve, major
11:03hyperscaler earnings, and the have a
11:05brutal month for momentum. That's what
11:06made this situation worse. And honestly,
11:08this is what obviously took
11:10Leopold over the edge is that going into
11:13the week from last Friday when there was
11:15clearly a problem,
11:17you had a big week coming up with a Fed,
11:20and I'm not a tin foil person, but the
11:24fact that Citadel
11:26came out with a news story saying
11:27surprise rate hike was their call for
11:31Kevin Warsh, and you have the hyper
11:33scalar earnings, which could cut CapEx,
11:35and that rumor was flying around, and
11:37you've already had a brutal month, which
11:39means people are playing with a short a
11:40short stack. That's why I went through
11:42this to kind of go through midweek. So,
11:44for those of you who are subscribers,
11:47I've done a midweek, I think, every week
11:49this week in July, or at least the last
11:513 weeks.
11:53Um and the reason is because there was a
11:54lot going on in this, but also a lot
11:56going on in the thematic portfolio,
11:58which
11:59uh I've obviously put on the the
12:01website. It's part of this because it's
12:03correlated to everything that's gone on.
12:06So, this was the Citadel Securities
12:07seeing Warsh delivering surprise rate
12:09hike. So, when you put it all together,
12:12you're back with I saw LTCM,
12:16I saw the quant unwind in 2007, was
12:19involved in it, had positions on in both
12:21of those. By Archegos, I was not
12:24involved in it other than having a
12:26variety of
12:2830-some-odd portfolio managers reporting
12:30to me, and all of them end up in the
12:32same situation where you get the point,
12:35and then margin calls from several
12:36global investment banks. They all know
12:38what's going on. Block trade started to
12:40happen in this case. It's a little bit
12:42of all of them to me. I had said before
12:44that the market movements were reminding
12:47me more of LTCM than anything, and the
12:52quant unwind as well. LTCM is different
12:54in the fact that that leverage was
12:55impacting the world, and all of the
12:57markets around the globe except for the
12:59US markets were already falling. In this
13:01case,
13:02almost all the markets around the world
13:04are above their 200-day moving average
13:05and continuing to move higher.
13:07So, one of the elements that came on
13:10again last week that was an indication
13:12is
13:13someone, and it was either Leopold or
13:15preparing for while they're going out to
13:18go out and figure out how to market
13:19this,
13:20there were multiple billion dollars of
13:22SMH puts that were bought
13:24uh
13:26to hedge things. So,
13:28SMH falls violently in the four days,
13:31and guess what?
13:32You have this massive volume spike
13:35basically on the lows.
13:37Um
13:39I I you know, for everyone who's looking
13:41for things in here. And again,
13:43this is actually an awesome-looking
13:45chart for the size of the unwind that's
13:47happened. I said this before, this is an
13:50ABC at this point. Um the lows to me are
13:53critical, and I'll go through that as
13:54well. Uh but, I mentioned Micron last
13:57week. There's a neckline of a head and
13:59shoulders formation right here that we
14:01broke below, and that we've broken back
14:05above now at this point.
14:07This was the most interesting part. Uh
14:09the day before the news articles came
14:11out,
14:12uh this was the second day in a row that
14:15uh people had called me up saying that
14:17someone was being unwound, and something
14:19big was happening. So, this is the day
14:21after Kevin Warsh.
14:23So, this is Wednesday.
14:25The market sold off. The S&P was down
14:26120 points, which it made all back the
14:29day the next day. This was clearly the
14:32day that the positions were taken on
14:34hold by, I'm assuming, Citadel. And they
14:38take these positions on, and then
14:40there's a massive
14:42futures position that goes on. And the
14:44reason that you can see this is this is
14:47the total year-to-date volume of the
14:51E-minis, the S&P electronic futures,
14:54from 2:00
14:56to 4:00. So, into the close, except for
14:59one day, which was June 4th, which
15:01happened to be the gap down here, which
15:03was a big volume day.
15:05This was the biggest volume for E-mini
15:07futures in that 2-hour period into the
15:09close of the entire year. So, that's how
15:13you know that a lot of the selling was
15:14done in a pukage basis by just hedging
15:18what had gone on. So, when someone comes
15:19to the market, and let's assume that the
15:22reports are correct and he sold about 15
15:25to 20 billion dollars of positions at
15:27the market bid down a certain amount,
15:30well, the futures, not surprisingly, get
15:32knocked down one one and change percent.
15:35Uh
15:36they probably bought everything down as
15:38an index 3% down, 4% down, 5% down. So,
15:42I used to do this when I traded the S&P
15:44book. Um these types of things go on and
15:47you make markets on them and that's what
15:48happens, and I think that's what went
15:49on. Now, it doesn't get past the fact
15:51that
15:52as of Thursday morning
15:55before we got the bounce, this is where
15:57the momentum factor was for technology.
16:01Down 40% year-to-date.
16:04Worse than anything during the dot-com
16:06bubble by almost
16:0820% except for this one period here.
16:11I I that one month there. So, again,
16:14historic month
16:16for
16:17industrial momentum.
16:20Again, before the market opened on
16:22Thursday, down 32%.
16:24I mean, double any other month.
16:28This one's the most important,
16:29especially for you multi-strats, but
16:31also for um anyone who's factor uh
16:34focused. If you strip out the factor
16:37risk of your portfolio, then you're
16:38running a factor neutral portfolio
16:40that's also going to be sector neutral.
16:42This is sector neutralized momentum.
16:46As of again, Thursday morning, down 21%.
16:50Um now again, when you look at the
16:52months that were up, these were
16:53unprecedented up months. This is an
16:56unprecedented down month. So, again,
16:57we're giving back gains and this is the
16:59way it goes. You stretch the rubber band
17:01too far one way, you get levered. This
17:03is the reason why I did the AI mid-cycle
17:05slowdown because we had stretched the
17:06rubber band too much one way.
17:08I got crap for getting out of Micron way
17:11too early, and then I got crap last week
17:13for saying I was buying Micron below
17:15900.
17:17Guys, I'm going to go through this as it
17:19goes on. When you see these types of
17:20moves, if you think you're going to be
17:21able to pick tops and bottoms, I've
17:23never in my lifetime been able to pick
17:25tops and bottoms. What I'm trying to do
17:27is increase positions based on a
17:29Bayesian philosophy of the news that's
17:31coming for me. If you're watching this
17:33to try and pick tops and bottoms,
17:35especially in the speed chess world,
17:37it's not going to happen. This down here
17:40is already extended. It was extended
17:41down here. It was extended here. You're
17:44trying to stay solvent by waiting until
17:46the panic subsides, and we saw a panic
17:49situation that historically signals the
17:52end. In terms of those four days, this
17:55is more shocking. This is the four-day
17:57rate of change, again,
17:59into the close on Wednesday for
18:02sector neutralized momentum. So, again,
18:07is a long-short basket down 17%
18:10that is sector neutral. So, this is
18:12momentum across all sectors. It just
18:14shows the extent of the damage. Now,
18:16I've talked about this before, and this
18:17is why
18:20everything has changed going forward. Um
18:23this is as of uh
18:25the close on Friday.
18:28This is the current 30-day and 60-day
18:31realized vol of the Morgan Stanley Tech
18:33Momentum Factor.
18:35There is no way that this doesn't keep
18:38the risk at lower levels. So, even if
18:41people have de-risked, they're not going
18:43to re-risk anywhere near where they were
18:45before.
18:46Um partly [clears throat] because of the
18:48drawdown, partly because some of them
18:49are going to be in the hole, but a lot
18:51of it has to do with until this
18:52volatility comes down, and I don't
18:54expect it to come down quickly, and I
18:56still believe structurally it's going to
18:58be higher than it's been in the past.
19:00This is the main problem. This is
19:02something which is beyond belief. To be
19:03up at 110 when the prior highs were
19:07around 70, which means a normal move
19:10right now in the market,
19:12Taking the square root of 252, 108
19:15divided by divided by the square root of
19:17252, which is 16,
19:19gives you about 6 to 7% a day.
19:23And that's market or you know, that's
19:25that's market neutral in the fact that
19:26it's dollar long, dollar short. These
19:28types of moves are huge. Now, if you
19:30take it a step further, what I did is I
19:32took all of the market neutralized
19:36factors. So, forget just the tech
19:38momentum side, which is tech, and I took
19:40it for beta, I took it for quality, I
19:43took it for momentum, and I took it for
19:44value, and I just went through and gave
19:47it an equal weight. So, divide each of
19:49them by four, they're uh
19:52uh 60-day volatility.
19:54And here's where we are, 38. We're
19:55above. So, this gets into any kind of
19:58optimized portfolio where you're risk
20:00weighting even your factors.
20:03Th-
20:04This is something that has been trending
20:06higher for some time. Just look, you
20:08don't see anything with this stuff
20:09trends higher for a long period of time.
20:12This is the impact of AI, and it is
20:14never going away. And I'm going to go
20:16through the reasons why.
20:18We Here's more um on the liquidations
20:20that went on and just some more of the
20:22stats. Hedge funds just sold more tech
20:23than any point in Goldman's 10-year
20:25record. Momentum factor volatility at
20:26its highest in 45 years outside a
20:28recession. So, I don't even have data
20:30going back 45 years. Gross and net
20:33exposure is off five and four-year
20:35highs, now in the middle of the pack.
20:37Six weeks of that, and hedge funds are
20:38still up 9% through June. Again, through
20:41June, one of the best first halves in 20
20:43years. So, again, that was through June.
20:46So, I'm going to remind you guys, I
20:48wrote this paper back in May.
20:51Bubbles, parabolas, and speed crashes.
20:52This is exactly what we saw. So, I wrote
20:54this before the peak. We're going to see
20:56more bubbles, we're going to see more
20:58para- parabolas, and what comes with
21:00that are speed crashes. So, it's going
21:02to look like a bubble. Basically, we're
21:04compressing time.
21:06And the reason we're compressing time is
21:07cuz AI is compressing time. The model
21:09capabilities are going so fast and we're
21:11entering to a world of AI agents which
21:14are on 24/7. It's the end of human
21:17beings. Now, this has been happening
21:18anyway. If you guys have been down in
21:19the New York Stock Exchange anytime
21:22uh since the 1990s,
21:24it looks a lot different today than it
21:26did back then. We've already been in
21:28this situation. Machine learning has
21:30already been a part of our world. It's
21:32just that now it is ubiquitous and
21:33everyone has it. You have to incorporate
21:35that into what's going on. Now, we not
21:37only saw the liquidations for hedge
21:40funds
21:41in Korea, we've basically been in a
21:433-week unwind of positions there. So,
21:46again, we've cleansed positions,
21:49but you're not going to be able to take
21:51that kind of size again. So, don't
21:52expect this to go back to the same
21:54direction cuz what chased it, as someone
21:56who focuses a lot on Bitcoin and on uh
21:58crypto, you need the energy of retail
22:01and we're just not going to get back to
22:02that anytime soon. I talked about that
22:04with silver or gold and Bitcoin. The
22:06difference is with the AI trades for
22:08those of you, I say it every week,
22:11the majority of positions are above the
22:12200-day moving average still. So, of the
22:15AI positions, I think 74% are still
22:18above and 84%
22:20of my basket
22:21are still the 200-day moving average is
22:23still going higher. So, we're not in any
22:26position where any damage was done.
22:28Hedge fund leverage.
22:30This was as of the third quarter of 2025
22:33and this is data from the Securities and
22:34Exchange Commissions Form PF.
22:37You can see how much it had gone up
22:39since COVID.
22:41Hedge funds have been having
22:42performance.
22:44This has gone higher. It's not just the
22:46basis trade, it is also equities. So,
22:49it's not just fixed income, it is across
22:51a range of places. Now,
22:54I want to bring this up cuz I wrote this
22:56paper on the art of unlearning at the
22:58Fed, which we'll get back to, but I just
22:59want to read this line cuz this is what
23:01goes for everyone who's a risk manager.
23:03If you're at a fund and your risk
23:05manager doesn't watch my videos, I would
23:06give them this because this is a
23:08structural change. AI is progressing at
23:10an unprecedented pace and in a world
23:12defined by exponential innovation,
23:14relying on backward-looking models is
23:16too slow. This was me writing about what
23:19Scott Bessent had said
23:22in an interview on the All-In Pod when
23:25he was going through what was going to
23:26have to happen with the Fed.
23:28Here's the way that it's doing it. First
23:30of all, AI is compressing economic time.
23:32The model capabilities are improving
23:34rapidly. So, what used to take 10 years
23:36in innovation is now taking a year.
23:38We went through this in the first
23:39quarter. Terminal values are harder to
23:41estimate.
23:42You don't know if you're going to be in
23:44business in 3 years if you're built on
23:45code. Think Adobe. Think Salesforce.
23:48People still have no idea. So, there was
23:50multiple compression cuz we were
23:53taking off terminal value. All this
23:55feeds into the factors.
23:57If all of a sudden, we don't know if a
23:59company is going to go, they can go down
24:0040% without their earnings changing.
24:03This is not good for risk management in
24:05terms of having this instability.
24:07Hedge funds operate on a much shorter
24:08clock than the assets they own. This is
24:10going to be a big deal, especially as we
24:12get into tokenization, where things are
24:14going to trade 24 hours a day.
24:16Are the hedge funds going to be in a
24:17position to trade this stuff 24 hours a
24:19day? How much more miserable is life
24:21going to get if you're down big over the
24:22weekend? The more that markets are open
24:24or trading, the more likely you are to
24:26make emotional response. That is bad.
24:30Leverage turns valuation uncertainty
24:32into liquidity risk.
24:35That is important because
24:38when you get falling confidence in
24:39terminal value, it raises volatility,
24:41tightens financial conditions. We're
24:42seeing all of these play out. So, I
24:44don't think the prime brokers are going
24:46to allow leverage to go to the levels
24:47they have because I don't think the
24:49volatility that we saw is coming down.
24:51Ubiquitous AI risk tools create a false
24:53sense of confidence. This is like
24:54watching people drive down the street
24:56with all-wheel drive when it's icy and
24:58snowy out thinking that they're
25:00protected. Everyone right now has AI
25:03risk tools. You can just say optimize
25:06this portfolio, what is a way to reduce
25:08my risk?
25:10It
25:11You're just at a point where there's a
25:12bubble in optimization and thinking that
25:14your portfolio is hedged because it's
25:16using as Scott Bessent alluded to,
25:19the way that you come up with an
25:20optimized portfolio is that you're using
25:22past correlations and past volatility.
25:25As I've shown you in the factor stuff,
25:27this is no longer relevant unless you
25:29believe mean reversion's happening. And
25:30I'm telling you the market structure has
25:32changed forever because of AI. It is
25:34never going back. Tokenization is here.
25:37Terminal value is gone forever. All of
25:40this stuff is happening. You're seeing
25:41things like companies that are private
25:43like Anthropic grow so fast, the fastest
25:46in the history of companies,
25:49but we're also seeing companies go out
25:50of business faster than they've ever.
25:52The greater terminal value uncertainty
25:54leads to multiple compression.
25:56That's what we're seeing. So, the S&P
25:58500 earnings are growing at over 20% a
26:00year,
26:01but the multiple's compressing cuz the
26:03stocks are not going up to match it. If
26:06you don't believe the AI agents are
26:07here, Robinhood keeps releasing things.
26:09Here's an AI trading agent that reads
26:11charts, RSI, MACD, and 16 more. This
26:13means retail has access to the same
26:15tools
26:16that every quant hedge fund has.
26:18Now we're going to get through the
26:19changes that are happening and how
26:21disastrous this is, especially for the
26:23month of July, because I am sure there
26:26are lots of people. I get so many people
26:29that reach out saying they've been
26:30buying S&P puts, they've been buying
26:32Nasdaq puts. This is going to crash.
26:35I'm not kidding you when I say that very
26:37smart people who I respect keep
26:39constantly calling up with this stuff.
26:42Here's what's going on. This is tech
26:45momentum, the one-year realized vol at
26:4867 and a half.
26:50Here is the NDX realized vol for the
26:52same time period. Here is the
26:54relationship over time.
26:57This is insanity.
27:00You're at one of the lowest levels of
27:02realized wall in the NDX and you are
27:04parabolic on the volatility.
27:08This is where RIAs, mutual funds,
27:10everyone who doesn't have the same time
27:12horizon as the hedge fund
27:15should be having a conversation with me.
27:17We're getting those conversations. This
27:19is about market structure. You can
27:21benefit significantly from this and from
27:23the people that I talk to regularly,
27:25they did very well reducing risk back in
27:28May even though it was early because
27:30they're in a position right now to start
27:32building up some of the risk on
27:33fundamentals that they like. There's
27:35going to be more speed crashes, guys,
27:37than there ever have been. If you don't
27:39believe that chart, here is that factor
27:41neutral wall that I said, which is at
27:4336, which again takes an equal weight of
27:46quality beta,
27:48momentum, and value
27:50sector neutral factors.
27:53Equal weight relative to the six
27:55contract of the VIX.
27:57So, for everyone who's bought S&P wall
27:59or six month,
28:01this thing tends to go higher and we
28:04could be an event right now where
28:05multiple hedge funds are going under. If
28:07you ask me, as I'll get through, the
28:09only two areas when I've spoken to the
28:11risk group at the Fed that I'm worried
28:13about, I'm worried about insurance
28:15companies and I'm worried about hedge
28:16funds. Those are the only groups that I
28:18think have the leverage capable of this
28:21going up here. That would be a credit
28:23event. I think that needs to involve
28:25either a hedge funds, but since Citadel
28:28Millennium are up big for the year and
28:30they're involved in this stuff and
28:32they're the best risk managers in the
28:34world,
28:35we're probably in a situation where this
28:37is more likely to decompress over time
28:39and that's my baseline. If you ask me
28:41what's more likely to happen, I don't
28:42think there's going to be a major event.
28:44I think there's going to be multiple
28:45compression and I think there's going to
28:46be a lot of disappointment, but I think
28:49deleveraging is going to happen for a
28:50long time because of this staying up
28:52here. So, if you ask my gut, I don't
28:54think we're mean reverting anymore.
28:56We've already seen a structural higher
28:57level here than it was all back here.
29:00And this is not a short time period.
29:02This basically stayed between 5 and 15
29:05from 2000 to 2020. Since then, it's been
29:09from 15 to 30.
29:11We're in a structurally high volatility,
29:13so you're just going to have to deal
29:14with it. Here's another one. This is
29:16taking the VIX, which is the index
29:19implied vol, and this is taking the VIX
29:21EQ, which is the equity
29:24vol. So, equity vol has broken away.
29:27This is the long-term relationship
29:29between the two. And we've got a huge
29:32difference here, and effectively it
29:34started right around ChatGPT.
29:37So, I just think that equity vol,
29:39because you can go out of business much
29:41very quickly,
29:42is going to stay at the high side, and
29:45index vol is going to stay on the lower
29:47side relative to it.
29:49Um again, if you want to reach out,
29:51especially if you're an RIA, where I'm
29:53having these conversations, you can do
29:54it weekly, you can do it monthly, but I
29:56think being on top of the factor side,
29:58being on top of the AI from a bigger
30:00perspective, asking the questions,
30:02that's where the alpha is going to be
30:03generated. So, you can reach out on
30:06there.
30:07Um
30:08This is a bull market still. 67 new
30:11highs in the S&P yesterday, the most
30:12since February. So, as this whole thing
30:14has been going on in July,
30:16we've got that. We got 72% of the S&P
30:19500 greater than their 50-day, 200-day,
30:2268% above the 20-day. I mean, again,
30:26everything's been fine during this.
30:28Here's where we finished the month.
30:31Unchanged.
30:32Four sectors down,
30:35seven up.
30:37This is not normal compared to what went
30:39on in LTCM and what went on during the
30:41quant unwind. It is similar to Archegos.
30:44Archegos ended up going higher from
30:46there. But, I think you just have to pay
30:48attention. Here's a chart on the S&P.
30:50And remember, that was the day where we
30:53sold off at the end of the day and we've
30:55immediately not only gone back up. The
30:57next day we went higher and now we
30:59closed way above it. I'm going to get
31:00into some of the technicals on that, but
31:02just be aware that every time we kind of
31:05push this ball underneath the water, it
31:07pops out.
31:08Here's equal-weight S&P monthly returns.
31:11Look at how many months we've been up
31:14with only one month down, which was when
31:15the invasion or the bombing of of Iran
31:18occurred and oil spiked higher.
31:22This is a broad market
31:24in terms of the rally that's going on.
31:27Here's the NDX. So, we hit the lowest
31:29RSI in the last year. You look at what
31:32happened the last time we got down to
31:33this level and we had a similar type day
31:36where we were down and the next day we
31:38gapped higher. Down, gapped higher.
31:40We'll see what happens. Uh I'm going to
31:42go through some of the technicals, but
31:44remember, this is very correlated with
31:46my thematic portfolio, the AI trade in
31:48general, semiconductors.
31:50The NDX did go down and NDX over the S&P
31:54actually had its worst month since the
31:56dot-com bubble, believe it or not, guys.
31:58Uh worst month since the dot-com bubble.
32:04Here's SMH. As I mentioned, we had this
32:06reverse head and shoulders. I'm telling
32:08you um again, these lows need to hold.
32:12But, reverse head and shoulders where
32:15you break the neckline, everyone gets
32:17short. I'm sure a lot of people got
32:19short here and then we gapped higher and
32:21then we sit there.
32:23Technically,
32:25this is what I'd be focused on. If you
32:26guys aren't, you know,
32:29aware of follow-through days,
32:33a follow-through day is confirmation
32:34that an oversold bounce may be turning
32:36into a real investable rally rather than
32:38just a short covering pop. So, the
32:40reason I like this and the reason I used
32:42it, made a lot of money off this after
32:44bottoms.
32:45I preferred to find bottoms than tops
32:47because generally the sentiments there,
32:49you get these very noticeable signals. I
32:53was talking about all the things that
32:54bother me beginning in May and the
32:56market kept going higher. Situational
32:58awareness.
32:59Going out of business and all the things
33:01that I showed you in the beginning of
33:03this, that is very indicative of a
33:04cleansing event. Typically technicians
33:06look for it a few days after the low.
33:09Often four day day four through 10 of an
33:12of rally attempt. A major index rises
33:14meaningfully, it's common, volume is
33:16higher, blah blah blah. So, basically
33:18what you're looking for is give it time.
33:21Here's the low.
33:22The next day we have the biggest volume
33:25since June.
33:27This is on the Qs. So, we get this
33:30sell-off here. This is the day again
33:33that we get the panic from Leo Bolt. So,
33:35the positions trade, Worch speaks, we
33:38get this massive sell-off. The next day
33:40we trade on higher volume, we close
33:41above the highs of that day.
33:44Structurally a good thing. Do I think
33:46we're probably going to spend time here?
33:47They're talking about day four to 10.
33:49So, as we move out here, if we hold the
33:51lows and if we just kind of build this
33:53time in here, that's a good sign. That's
33:55what I'm looking for.
33:57Um
33:58we'll see what happens. If you're
33:59looking for fundamentals to support
34:01this,
34:03uh we're now through 60% of the S&P.
34:06Uh sales growth. Since earnings growth
34:08is a little bit is not a little, it's a
34:10lot distorted by
34:12Anthropic inside this, but when you go
34:15through each of these, just look how big
34:17these numbers are. It's every sector.
34:18More importantly,
34:20sales growth, 14%.
34:23I mean, guys,
34:24I ain't keep saying it. It's like it
34:26doesn't pay to be to be bearish stocks
34:28when this is going on. You have so much
34:31fundamental support and revisions as we
34:33go through earning season. Again, going
34:36higher. So, not only are they beating
34:37numbers, not only are they coming
34:39through on a historical basis,
34:41the revisions keep moving higher. You
34:43don't want to be bearish on stocks while
34:45this is going on. This is the longest
34:47period now in the last 5 years.
34:50It's very difficult. So, we'll see what
34:52happens. Here's my thematic portfolio.
34:54So, last week we were over here. I said
34:57we're at the 50. It wouldn't surprise me
35:00to see us go lower, blah blah blah. We
35:02did. We're back up in here. We're in the
35:0330 somewhat.
35:05Is that it?
35:06I'm like I said, I'm betting on the fact
35:09right now with my own portfolio off the
35:11lows of that panic low that that was it.
35:14If not, there's something bigger going
35:16on. Hedge funds could be massively
35:18unwinding. We'll see what happens, but
35:20that's why we trade these lows. We don't
35:22sit there and invest them. Um I have
35:24positions on in things I believe in,
35:26which I'll have. And then the positions
35:27like Micron and AAOI
35:31and
35:32uh Marvell and some of the other things
35:33I've added to. We'll just see where
35:35everything goes. Here's a thematic
35:36portfolio on an RSI basis. 14-day RSI
35:40hit the same level it did here on the
35:42lows here. Again,
35:44we'll see what goes on, but
35:47as of right now I'm leaning towards
35:48that. Now, this is the thing that I want
35:50you guys to understand in terms of
35:52adding.
35:54Go back to what I said how factors are
35:56changing and market structure's
35:57changing. Never before, in my opinion,
35:59has looking at the rate of change been
36:02more important.
36:03Most people I know don't use this. I've
36:05always used it since my days in Brazil.
36:08Uh I do believe markets move on a rate
36:10of change basis, and it pays to reduce
36:13risk when you get above historic levels.
36:15Now,
36:16in May and into here is when I was
36:19reducing my Micron position. So,
36:21beginning around the middle of May, a
36:23little bit all the time, a little bit of
36:25Marvell until we got up here. Now,
36:27this is the 63-day rate of change. So,
36:30this is 63 trading days, which is 3
36:32months.
36:33We made it all the way up to we were up
36:35over 50% on a look back of the prior 63
36:38days for the thematic portfolio.
36:42Now, over the last 63 days, we're down
36:441.7%. We have now retraced. We have now
36:49done what we needed to do in both price
36:51and time, and that's what it takes with
36:52rate of change. When I did the
36:54consolidation and I showed you the the
36:56consolidation patterns that are likely
36:58to happen,
37:00it's time and price. So, we've done
37:02that. So, this is just a metric for you
37:04guys to use. I provide it every single
37:06week in that technical sheet, and I'll
37:07show you some of the things in there.
37:09On the 30-day basis, so a month and a
37:11half or a month and a quarter, down 15%.
37:15So,
37:16from as my father taught me, the odds on
37:19the AI trade are much better than they
37:21were at here. This is on a 30-day basis
37:24when I started to reduce Micron. So,
37:26again, you're at a much better entry
37:28point in price and time. I'm not trying
37:30to pick the bottoms in this. There's a
37:31lot of new news that went on. Now, for
37:34those of you looking to trade the index,
37:36which I highly recommend, if you're
37:38going to pick names and trade the names,
37:40but if you're going to jump into things,
37:42the 10-name concentrated position, I
37:44just want to make sure you guys see that
37:46it was correlated before. I spent a lot
37:48of time to come up with 10 names that
37:49are highly correlated to the 100-name
37:51index, so you didn't have to do that. It
37:54stayed correlated with it and traded
37:55extremely well with it.
37:58One of the names in there is Corning. I
37:59highlighted this this week on the
38:01subscriber video. The thing about
38:03Corning, it hit the 200-day, it traded
38:05up. The reason I have this chart here
38:07from December 31st,
38:10this day here is when the announcement
38:12came from Meta for $6 billion.
38:16They had another day here where Nvidia
38:18partnered with them. So, they do an
38:20announce a partnership with Nvidia, they
38:22announce a partnership with Meta for $6
38:23billion, which according to every LLM
38:27that I use is equivalent to somewhere
38:28around $100 million
38:30100 million homes worth of fiber optical
38:33fiber.
38:35And the stock got all the way back to
38:37where it was basically on the close of
38:39that day. This is one of the
38:40concentrated names. Now, if you look at
38:43it on a rate uh
38:44Well, I guess I didn't put it in there.
38:46I thought I had a rate of change basis
38:47on it. Here are the uh percentage names.
38:50So, again, I've shown you guys this each
38:51week. When you upload the technical
38:53sheet, I highly recommend going into
38:55co-work and using this to look for this.
38:58The reason I highlighted these is July
38:5914th, we had 26% of the names above the
39:0320-day.
39:05Here we are now. We've had 2 weeks where
39:07even with all the carnage, it's starting
39:09to show signs here of positivity
39:11relative to these 2 weeks. I'm more
39:14interested in this one.
39:17This is somewhat similar to here. If we
39:19start to see this thing move higher
39:20again, I think that's a good sign. 84 of
39:24the names out of the 100 200-day slope
39:26rising. So, you're looking for signs on
39:28that.
39:30I'm uploading this to the website. This
39:33is combing through the rubble. So, I
39:35went through and I basically wanted to
39:37do something and again, part of this is
39:39to teach you guys how to do this. Um I
39:41highly recommend again, I do this
39:43presentation for RIAs as well in terms
39:45of how they can go through in someone's
39:48portfolio and be able to show the things
39:50that are technically still moving well
39:52and be able to do this. So, this is two
39:54conditions. I want to make sure that if
39:57I'm going through the rubble, number
39:58one, I want the uh price above the
40:00200-day moving average. Think Micron,
40:02the 200-day moving average is somewhere
40:03around 500. We're still around 850.
40:07And I want the 200-day itself still
40:08rising. Those are the names that
40:10structurally to me I want to be involved
40:12in. That's my favorite technical sign.
40:14So, I want to find things that have
40:15fallen back. And what I do in here is I
40:17go through and rank all of them on that.
40:19You guys will see
40:21the list. It has all the names in it.
40:24And so just to give you like a quick
40:25side, like I mentioned
40:27AAOI, I ended up buying some this week.
40:30And again, when I say I bought some,
40:31these are all small positions. If you go
40:32back to what I had in May, my Micron
40:35position right now is about 1/5 of the
40:37size that it was in May when I was
40:39getting out of it. So again, I'm not
40:40trying to pick the bottom of this. I'm
40:42trying to work into something where I
40:44don't think the upside is as good. At
40:46800, do I think it can get to 1,600 over
40:48the next year? Yes.
40:50At 500, it's a better risk reward. So
40:53I'm not convinced yet that we've made
40:55the absolute low, but I want to start
40:57buying some because I think we're
40:58eventually going back up to 1,500 to
41:001,600. Now, in the case of this where
41:03the slope is up on the 200-day and the
41:07percent above the 200-day is here and
41:09we're now
41:1131% down for the month, 27% down over
41:14the 3 months, verse the 50-day, it's
41:17below by 34. Like I'm trying to find
41:19names that are still above here, still
41:22positive here, but are way below in
41:24these things. So I go through the list.
41:26I also go through what they've said on
41:27their earnings as well to see if the
41:30fundamentals still line up.
41:32Here are the two things that I want you
41:33guys to start paying more attention to
41:35that I have one eye open. So I'm always
41:37as a risk manager looking for something
41:40that could
41:42change my mind as to what's going on. So
41:45the hyperscaler credit CDS does not
41:47bother me one bit. What does worry me is
41:51that the triple C market, the option
41:54adjusted spread,
41:56has this bad looking roll to it. And
42:00this is overlaid with high yield or the
42:02junk market. So whenever I see these
42:05divergences, the way credit usually goes
42:07is the weakest start going up and it
42:08eventually drags everything in. There is
42:10no doubt right now that the amount of
42:12capital that's being absorbed by AI is
42:14massive, and I do believe the disruption
42:17happening from AI is real. So, I think
42:18we're going to see more and more private
42:21credit issues. The question is, is this
42:23some of the private credit issues that
42:26have been in play really this entire
42:28year because this started right around
42:31the time that the software stuff fell.
42:33So, I wouldn't ignore it. And if you
42:35guys want to read, I I'm I love good
42:38conspiracy theory when it comes to
42:40credit. I like to stay on top of things
42:43because I do believe that they start
42:45with
42:46one-off things. And the Dodgers have
42:48bothered me as a sports fan spending
42:50money in any possible way. And so, this
42:54whole thing with the Dodgers,
42:57Guggenheim,
42:58Mark Walter, and private credit, I'm
43:02just I would go read this and just pay
43:04attention. I don't know what it means. I
43:06just know the insurance industry
43:08fairly well, and I think there's a lot
43:10of mispriced assets in there.
43:13Uh Google, I was shocked how many
43:15people, when I did my video last week,
43:18told me I was wrong on Google. This was
43:20going to collapse, and of course, this
43:22was here.
43:24Uh sorry, here. We closed below the
43:25200-day. Um
43:28I put this in last week. I'm shocked how
43:30many people
43:31don't care about the backlog of Google
43:33and tell me it doesn't mean anything.
43:36It's not real.
43:37I I couldn't disagree more, guys. Uh I'm
43:40going to go through a bunch of things on
43:41this. I think you guys need to stop
43:43reading some of the people who are
43:45basically just doing the math like I
43:46showed last week and saying these
43:48earnings sucked. Clearly, Google's
43:50earnings didn't suck. They've already
43:52gone back up. I expect fully this
43:54company to completely take off and keep
43:56going because the numbers are insane in
43:59terms of the backlog. And if you didn't
44:02believe them,
44:03Zuckerberg comments about compute demand
44:05from his call. There's just nowhere near
44:07enough compute for all the demand. We
44:09are getting a large number of offers for
44:10the compute that we have.
44:13Meta is not building computer ahead of
44:14uncertain demand. Zuckerberg said demand
44:16already exist available compute blah
44:18blah blah. If you don't believe them
44:19because their stock trade off.
44:24Sundar Pichai, we continue to be supply
44:25constrained a sign of momentum and rapid
44:27adoption. The demand still outpaces the
44:29investment.
44:30Meta, there's nowhere there's just
44:32nowhere near enough compute for all the
44:34demand.
44:35Microsoft which traded much higher.
44:38Customer demand continues to exceed
44:40available capacity. That additional in
44:42quarter capacity for Azure was quickly
44:44monetized. Can't tell you how many smart
44:46people told me that these guys were
44:48going to cut CapEx this quarter.
44:51Wrong.
44:53Every incremental unit of compute run
44:55online is being absorbed almost
44:57immediately while demand continues to
44:59outrun supply. If you don't like those
45:01hyperscalers, how about we go to Amazon?
45:03Even at that amount, we will still will
45:05not have enough capacity to meet all of
45:06the demand we have in 2026.
45:10I believe this dynamic will also be true
45:12in 2027, too.
45:14Guys, don't be short these things.
45:16Honestly, like what are you doing?
45:18I I don't like
45:21the demand was already we already have
45:23for 2028 is striking.
45:26This may be the strongest
45:27forward-looking comment from the call
45:29because Amazon is already seeing
45:30substantial commitment several years
45:31ahead.
45:33The lion's share of AWS 2027 compute
45:36capacity is already reserved. We have a
45:37clear sign of light to strong financial
45:39returns.
45:41Amazon also explained why free cash flow
45:43looks weak before the revenue of
45:44arrives. It begins spending on data
45:48So again, I want you to go back to what
45:49I said about the factor risk.
45:52The model capabilities are insane. They
45:55are moving faster and faster.
45:57This whole open source thing is stupid.
46:00Yes, it will be a part of it. Open
46:02router will go through it, but they need
46:04to have the cloud and they need to have
46:06these companies because they're the only
46:08ones with the compute. Nobody else is
46:11building the compute. I showed you guys
46:13last week that those Chinese companies
46:15are so tiny and they're building
46:17compute. They have to be run somewhere.
46:20So, where do you get the chance to use
46:23the compute of these Chinese models?
46:25It'll have to be because it's on bedrock
46:27or it's on wherever you want, but it has
46:30to be on somewhere where there's
46:31compute. Somewhere around the line there
46:33has to be hardware that has the compute.
46:35There isn't any compute.
46:38Here are the new backlog numbers.
46:40500 billion for Amazon, 500 billion for
46:43Google, 700 billion. This is 1.7
46:46trillion dollars between those three of
46:49a backlog. Our backlog stands at 496
46:52growing triple digits year over year. A
46:54large amount of it tends to be deals and
46:56agreements that you've signed. They
46:58talked about the fact that
47:00their new demand is not from Open AI and
47:03Anthropic.
47:05The majority of backlog is related to
47:07typical GCP contracts for a broad mix of
47:09customers.
47:11Oh, sorry. This is the one. All
47:13sequential commercial RPO growth was
47:14driven by commitments from customers
47:16outside of frontier model companies.
47:18Um if the people that have scared you
47:21are not writing about the facts in the
47:23earnings, then what they're saying is
47:25not facts. What they're saying is trying
47:27to
47:28use bear porn on you.
47:30You have to at least go through. You can
47:32say that these guys know nothing what
47:33they're talking about, that the Mag 7
47:35that are the proven track record
47:37winners, that their revenue is growing
47:39rapidly. They're the ones spending the
47:41CapEx and you're telling me the people
47:42writing these garbage newsletters are
47:45saying that Google, Amazon, and I that
47:47they're stupid.
47:49There is a risk that it's circular and
47:51the fact that Anthropic and Open AI go
47:53out of business, but we'll hit on that
47:55next. All in podcast, I thought it was a
47:57very, very good one. I give these guys a
47:59lot of crap because I think the thing
48:00has become way too political, but lately
48:04they're back on their their game. A lot
48:06of good things in there. I'm not going
48:08to go read all of it. Um but they go
48:10through the Kimi K3, they go through
48:13this. I thought David Sacks in
48:14particular had a ton of useful
48:16information on this.
48:17Just because he's connected to the White
48:19House and I think he was just very open
48:21and honest about all of this stuff.
48:23He
48:24pushes back against the panic and
48:26basically talks about how as much as he
48:28hates Anthropic,
48:31he talked about how they're the fastest
48:33growing tech company in history and that
48:35so far
48:37their annualized revenue
48:39is at $74 billion through the halfway
48:42mark of the year.
48:44So he's saying they're going to grow 10
48:45times again this year, over 100 billion,
48:49and he mentions that OpenAI is now on
48:51pace for about 75 billion this year.
48:54Guys, that's $175 billion in revenues
48:57from two companies that had zero two
48:58years ago.
49:00How can you be arguing about ROIC when
49:02these guys are printing money at that
49:04side and the cloud revenue guys are
49:06printing $2 trillion during the exact
49:08same period?
49:10So how people are coming up with this
49:12when the companies are getting the
49:13demand from the API side and from
49:16enterprises at the same time that the
49:18cloud guys are getting it. So it's just
49:21misleading completely. So compute demand
49:24versus supply Sacks goes through it. As
49:25far as I'm concerned,
49:28there is an unlimited appetite for
49:29on-demand intelligence. I don't think
49:31there's an upper bound for how much
49:32intelligence you can tap as long as it
49:34continues to get better. I completely
49:36agree. He goes through the numbers there
49:39on both these things.
49:42You have You still have stuff in the
49:44labs by Anthropic and OpenAI that is way
49:46ahead of the Chinese open source models.
49:49Sam is going to Washington over the next
49:51week to talk about GPT 6.0, which is
49:53blowing the doors off. Again,
49:56the models that are out, these guys have
49:57better. If you haven't seen Sam Altman
50:00on Invest Like the Best, I highly
50:02recommend going through it. I thought it
50:04was one of his better performances.
50:06Altman says OpenAI's conviction to
50:08secure massive compute came from seeing
50:10an exponential model improvement curve,
50:11especially around GPT-4.
50:14Basically, again, demand would be
50:16basically uncapped.
50:19He says the biggest near-term compute
50:21gains may come from software efficiency.
50:23Uh
50:24talks about the cyber incident that
50:26happened with hugging space.
50:30Uh I'll leave AGI alone and go through
50:32it. I'll summarize this in a little
50:34better ways if you guys care about, but
50:36um
50:38the fear flipped because demand
50:40exploded. This is the thing I want to
50:41make sure. As of
50:44October of last year,
50:46the probability of these companies being
50:48in trouble to me was
50:50much, much higher than at any time
50:53during this year.
50:55And the reason was because the
50:57the risk was these guys were spending
50:59money and the enterprises were never
51:00going to adopt it. That human beings
51:02were just going to be too slow. The only
51:04thing that changed it was that the AI
51:06agents came on much faster. And that's
51:08the point that a reason I did the
51:10portfolio is because the agentic world
51:13started and I started putting all these
51:15names together way before it went
51:17because I had already done the homework.
51:19The inference demands were going to be
51:21insatiable once AI agents reached. We
51:23just didn't know when it was going to
51:24happen and it happened at least a year
51:26faster. That was how Leopold invested in
51:29all these things as well. It just got to
51:31that point. What Sam does say is
51:33supply's the bottleneck today, compute
51:35bottleneck. The bottleneck has moved
51:37over time.
51:38He talks about research being a huge
51:40one. The amount of demand is just going
51:42to accelerate. Personal agents would
51:44massively increase demand. This is the
51:46meta side. This is why meta isn't
51:48selling their compute. They could have
51:49made investors very happy this week by
51:52saying we're going to rent out our
51:53compute, which they kind of leaked, and
51:56then they didn't do it. Why didn't they
51:57do it?
51:58Because if they do it, personal agents
52:00are going to massively increase demand.
52:02So, I'm sure they've run into this
52:03thinking this is coming faster than what
52:06we all know. And I believe that.
52:09The oversupply scenario would be models
52:11become so smart and efficient that they
52:13can do almost anything people need with
52:14current compute. Altman believes compute
52:17demand is structurally undersupplied
52:18because model quality unlocks new use
52:20cases. So, basically the Jevons paradox
52:22thing gets in.
52:24Is there a chance?
52:26So, what is the most likely scenario for
52:27supply to catch up demand? Based on his
52:30views. Software agents squeeze more
52:31intelligence out of compute. So, this is
52:33the efficiency side.
52:35Physical world robotics. So, humanoids
52:37eventually expands the entire compute
52:39supply chain.
52:41Neither of these is happening in the
52:42next year. It's just not happening. It
52:45It It It can't. We're not there yet to
52:47be able to go through it, and even if we
52:49are, the demand side is going to pick up
52:52the pace because of the personal agents.
52:54That's why I showed you guys the chart
52:55of what it means to have personal
52:56agents. So, agents make compute
52:59production dramatically more efficient.
53:00Models become so efficient that these
53:02tasks need less compute. Human attention
53:04becomes the bottleneck. So,
53:06good one to go through.
53:08Uh I talked about the revenues. They've
53:09seen a major uptick since May. This is
53:12the other reason why you have to be
53:13positive on these things. Now, this
53:15makes sense to me. I hate to say it.
53:17Where would you expect these to be?
53:18These guys are raising so much money. If
53:21credit spreads weren't widening and they
53:22weren't building in more risk as all of
53:25these guys are spending, of course
53:26there's a risk that any one of them
53:29could go out of business if they don't
53:31get the revenues in the door. Meta is
53:33making a huge bet. They don't have
53:35anything now. If they don't actually
53:36figure this out, they'll be in trouble.
53:39Gavin Baker, market is overreacting to
53:41hyperscaler credit spreads widening, the
53:43fact that spot prices for GPU rentals
53:45are at least two times higher than
53:46contracted rates is the missing piece. I
53:48could read all of this stuff. The
53:49reality is at 60
53:53billion per gigawatt, that's 1.5 to 2.2
53:56trillion in CapEx consensus estimates
53:57for hyperscaler cash flow are 1.3. He's
54:01just doing the same point I am, which is
54:04we don't have enough compute right now,
54:06guys.
54:08Dwarakesh put this out. It I think had
54:11over a million views um in terms of X
54:13going through it. I highly recommend
54:15going through to read it. It's a
54:18it's a very contrarian thought.
54:20Uh
54:21and it really says without going through
54:24it in a big way, the part that matters
54:26the most,
54:30his argument is that expensive compute
54:34converts token efficiency into pricing
54:36power, widens the frontier premium. So,
54:38this gets back to the fact that if these
54:41guys continue to move at this pace and
54:43their intelligent is better, but more
54:45importantly their token efficiency,
54:47which comes from the amount of compute
54:49they've built. So, the moat becomes the
54:51compute that they have access to. The
54:53fact that they can charge again, a
54:57reasonable price for token efficiency.
54:59So, to get this to a point where people
55:01can understand what it means, I had to
55:03kind of go through it. Um and again,
55:06Gavin spoke out about this, too.
55:11I wanted to give you guys an analogy
55:13that could work. Gas and cars. Why the
55:15best AI model wins when compute gets
55:17expensive.
55:19If gas is a dollar a gallon, nobody
55:21cares about the fuel efficiency.
55:24When it's $10, everyone pays a premium
55:26for the car with the best mileage. This
55:28is what's going to end up happening is
55:30that you might be able to pay cheaper
55:31for Kimmy K 3.0, but it's not as
55:34efficient as the best models. And that
55:36situation gets worse and worse because
55:38you need more hardware compute. And if
55:40there isn't any hardware compute to run
55:42it, the only place that has it is the
55:45place with the compute.
55:48If Anthropic has
55:51an enormous amount of token efficiency
55:54and Disney wants to take a $5 billion
55:57movie cost, but make it for $100
55:59million, that
56:00that margin of being able to buy all of
56:03the compute that Anthropic has not only
56:05gives them an insane way to make tons of
56:08money, but it gets them to destroy their
56:10competition as well. So, we're getting
56:13to the point where you have to look at
56:14the token efficiency. This is the other
56:16reason why I've written so much about
56:17Vera Rubin. We're getting into a
56:19situation where token efficiency is the
56:21story, not token maxing. And just
56:23remember how much SpaceX was able to do
56:25for their compute. So, we have a
56:27shortage of compute, the models are
56:29going faster, they're bringing on more
56:31demand because they're getting better
56:32and better, but there isn't enough
56:34compute out there. So, the people that
56:35have already built the compute, those
56:37prices are going to get higher and
56:39higher as SpaceX saw, as Meta saw when
56:42they realized that they could sell this
56:43stuff out. So, just keep it in the back
56:45of your mind that we're reaching a point
56:47where it's great that there's all these
56:49competitive models, but you have to be
56:50able to run them on something. And the
56:52question is where is that compute going
56:54to come from? And if the models are more
56:56efficient in terms of the token usage,
56:58you're still going to go back to using a
57:00better model except for the stuff that's
57:02free that you download on there. So,
57:05we're not at a point yet where we have
57:06the hardware for everyone to be running
57:08open-source models. So, there's at least
57:11another 2 years in my opinion, and
57:13that's more than enough time based on
57:14what we heard from the hyperscalers this
57:17week as to when they're going to get the
57:18revenue in the door. It's not a given. I
57:21gave you guys a probability last week,
57:22but it is happening. Now, if you still
57:24don't believe that
57:26the data centers and the compute, then
57:28just look at DRAM prices. This is is
57:31prices of DRAM. And again, this is the
57:34oldest one. This goes back to 2016. This
57:37is not the newest memory.
57:40They're still going up every week. So,
57:42the cyclicality cyclicality green mainly
57:44red. Look how many weeks this is going
57:47on and it's not stopping. Yes, it's
57:49slowing down because the price has gone
57:50up so high.
57:52All right, let's bang through the the
57:53rest of these. I did want to do
57:55something on the Fed because again
57:57so many people talking about it this
57:59week. People telling me some of these
58:01strategists were saying it could be 50.
58:04So, he holds rates.
58:05Three dissent. The Fed did exactly what
58:07Wall Street was expecting. So, why are
58:09the markets so vexed by Kevin Warsh?
58:12Again, the art of unlearning the Fed if
58:14you didn't read it, go read it. This is
58:15the world that I think we're in again. I
58:17think Kevin Warsh is a smart guy. I
58:19think he has talked publicly about the
58:20fact that we can't be using the academic
58:23approach of the past. We got to have a
58:24task force in to figure out the new
58:25things to use. The data that we're using
58:27now is not only stuff that gets revised,
58:30it's old. We need to be more
58:32forward-looking.
58:33The Fed is not the same Fed in this
58:35whole dovish hawkish thing is not there.
58:37Now, 10-year rates
58:39are sitting up near 470.
58:42Here is again 10-year inflation swap
58:45rates. So, as of right now since Kevin
58:48Warsh went in, we're down here.
58:50I don't know how long this will be, but
58:51he was saying, "Well, the market did it
58:53for us. Real yields are up here.
58:55Inflation's come back down so we can sit
58:56and wait."
58:59Here are the inflation expectations for
59:01the August print
59:03around zero.
59:06Dollar positioning. So, this is an
59:08interesting one because when the dollar
59:10weakens
59:11it's usually a very bullish thing and
59:13it's good for gold, it's good for
59:15silver. Uh
59:18maybe Jackson Hole will have an impact
59:20on this. Maybe the dollar-yen
59:22intervention, but look how big the
59:24crowded positioning is on the dollar.
59:26It's at the highest level in the last 4
59:27years.
59:29And the dollar has come down somewhat
59:31this week, most of it because of this.
59:34Speculation on the BOJ.
59:37And I just want to highlight for those
59:39of you trading, let's see, end of April,
59:42end of July,
59:44end of January. So, this looks like 3
59:45months, 3 months. Regardless, it's not
59:48pegged, but it sure seems to be like
59:50it's on a
59:52Let's just make sure it doesn't go
59:53anywhere. One podcast for you guys to
59:55pay attention to to kind of figure out
59:57the fit finish out this week.
59:59Um
1:00:01I've heard a lot of people talk about
1:00:02China. Uh
1:00:03and they're not in the same plan on AI,
1:00:07and they're looking to just make it free
1:00:08for the rest of the world. This is a
1:00:10great interview. Uh Ezra Klein
1:00:12interviewed Kevin Rudd, who is the
1:00:14former Prime Minister of Australia and
1:00:15an expert in Xi Jinping.
1:00:18I loved listening to it. I'm not going
1:00:20to read all this stuff on there, but for
1:00:22those of you trying to understand better
1:00:25uh what China is trying to do, and where
1:00:29the
1:00:30balance of power is in terms of AI and
1:00:33how they're viewing it, I think this is
1:00:35a really interesting one. I think Kevin
1:00:37Rudd did a great job. For those of you
1:00:39care about things, the Marxist-Leninist
1:00:42lens combined with the Schumpeter side,
1:00:45it's really interesting to me. I put
1:00:47this in myself, so this is me kind of
1:00:49extrapolating what I heard, comparing it
1:00:51to my belief in Joseph Schumpeter. This
1:00:53is the reason why I care so much about
1:00:54Bitcoin.
1:00:56For those of you interested, I think
1:00:57it's a a great um interview worth.
1:01:01Couple things,
1:01:02the cyber stuff is getting real. OpenAI
1:01:04and Hugging Face, you had this hacking.
1:01:07You also had
1:01:10a Claude
1:01:11hacking as well.
1:01:14Just worth paying attention to.
1:01:16And the Clarity Act is sitting here.
1:01:19It's still below 40%.
1:01:21Um
1:01:22depending on which party you listen to.
1:01:25We'll see what happens. Uh again, I
1:01:27don't think this matters
1:01:29in the long term
1:01:30uh
1:01:31for Bitcoin and for crypto. I think it
1:01:33does matter obviously
1:01:36uh for the entrepreneurs that are
1:01:37involved in the space and for the short
1:01:39term, we'll see how it plays out. That's
1:01:40it for this week, guys. So, final thing,
1:01:44I do believe that we saw something
1:01:45really important this week and that
1:01:47based on my history of dealing with
1:01:48hedge fund unwinds,
1:01:50those lows matter a lot um and how we
1:01:53trade over the next 6 to 7 days, there
1:01:56should be back and forth, but use that
1:01:58follow-through
1:02:00day approach. See you guys next week.

About this transcript

This page contains the full transcript of Situational Awareness, Hyperscaler Earnings and Warsh: A Bottom for AI Stocks? by Jordi Visser, generated from the public captions YouTube serves with the video. The transcript has 11,231 words across 1,754 segments, with the original timestamps preserved so you can click any line to jump to that moment in the embedded player.

What you can do with it

Use the transcript to take notes, quote the speaker, build a study guide, generate a summary with ChatGPT or Claude via the YouTube Summary tool, or export it as a timed subtitle file with YouTube to SRT. You can also re-open it in the transcriber to translate the transcript into 100+ languages.

Free YouTube transcript tool

YouTube2Text is a free YouTube transcript generator — no signup, no daily limit. Paste any YouTube link and get the full transcript instantly, with timestamps, click-to-jump, translation to 100+ languages, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.