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

[0:01] OK. We continue
[0:04] the summer, completing a
[0:05] quiet July for the
[0:07] S&P index. A
[0:11] brutal market for
[0:12] factors, for
[0:13] technology, for AI
[0:14] trading, and for hedge
[0:15] funds. I'll walk you through
[0:19] all of this. I will try
[0:23] to give you some
[0:24] experience gained
[0:25] after going
[0:26] through a bunch of similar
[0:27] situations. Starting
[0:31] from the beginning of my
[0:32] career. So,
[0:33] hopefully, everything I
[0:34] break down will help in
[0:35] some way. I
[0:36] will talk a lot
[0:38] about whether this is the end of what
[0:39] happened and
[0:40] what to look out for
[0:41] . I really
[0:42] believe there is a high
[0:44] probability that
[0:45] this was a "cleansing" event
[0:46] . I don't see everything
[0:51] skyrocketing right away, but I
[0:52] think when
[0:53] these
[0:54] things happen, and I'll explain why
[0:56] , it's probably more
[0:56] important than you
[0:57] think, given
[0:59] what happens when a
[1:00] hedge fund of this
[1:01] size fails.
[1:05] Number eight: the demand for
[1:06] computing
[1:07] power is
[1:08] insatiable. I just
[1:11] want to make sure that
[1:13] if I'm wrong about
[1:14] everything that you
[1:15] follow every week,
[1:16] number eight—this is the
[1:17] most important thing, because I
[1:18] say it in every
[1:19] interview, I say it
[1:20] here every week. The demand for
[1:23] computing
[1:24] power is
[1:25] insatiable. Currently,
[1:27] supply can't
[1:28] keep up with
[1:29] demand for
[1:29] computing, and there's no
[1:31] sign that's
[1:32] going to change, but
[1:33] most of the bearish
[1:34] rhetoric comes
[1:35] from
[1:36] people's concerns that we're building
[1:38] too much and won't
[1:39] get the expected
[1:40] returns. I'll cover all of that
[1:42] this week,
[1:43] drawing on
[1:44] earnings reports, on
[1:45] podcasts, again
[1:46] , at a time when
[1:48] people are so
[1:49] negative about AI from so
[1:50] many different
[1:51] angles. And as someone
[1:53] who takes risks by believing in
[1:55] it, using it,
[1:56] and talking about it, I can
[1:57] feel
[1:59] these sentiments very clearly, judging by the
[2:00] emails I receive
[2:01] from subscribers. Yes,
[2:04] people lose money.
[2:05] I lost money this
[2:06] month, and quite
[2:07] seriously, so I
[2:08] reduced my
[2:09] positions. So I
[2:11] understand that when
[2:12] the odds are against
[2:13] you, it's hard. But this is
[2:15] the market, and that's how it
[2:16] works. I'm not here to
[2:18] say that
[2:19] everything is guaranteed. Everything
[2:20] I do is
[2:21] based on
[2:22] probabilities. The only thing
[2:23] I am
[2:24] 100 percent sure of is that the
[2:25] market structure
[2:27] has changed forever, and
[2:28] I will also talk about
[2:29] this because I think
[2:30] it is important, especially
[2:32] for institutions, hedge
[2:33] funds, and mutual
[2:34] funds
[2:35] . Let's begin. So, this is what
[2:38] I launched. This is data
[2:41] as of Wednesday.
[2:46] So, this is the
[2:48] Goldman Sachs H or VIP hedge fund index
[2:50] . They create
[2:52] an index of the most
[2:53] popular or
[2:54] most
[2:55] widely held
[2:56] positions in the hedge
[2:57] fund world based on
[2:58] their data. I took
[3:01] this index, and the
[3:03] GS index hedged against the S&P
[3:05] 500. This is a monthly
[3:07] review since the
[3:08] index was created,
[3:09] which goes back to 2001.
[3:13] Minus 12% per month.
[3:15] So, that means that
[3:16] in a month where the S&P
[3:18] remains unchanged.
[3:19] Again, this is data
[3:20] as of Wednesday. The S&P
[3:21] is down about
[3:23] 2% as of Wednesday.
[3:25] This was by far the
[3:26] worst relative
[3:27] performance for such
[3:28] popular hedge
[3:29] fund positions, which
[3:30] , again, are
[3:31] reasonable. They do
[3:33] their homework.
[3:34] If it is a popular
[3:35] position, it means
[3:36] it must have
[3:37] good
[3:38] fundamentals. Worse than
[3:40] during the Lehman month.
[3:43] That's what this
[3:44] month is for. Also, it's
[3:45] twice as bad as any
[3:47] other month.
[3:50] So this was a huge
[3:52] shock to hedge funds
[3:53] regarding popular
[3:54] positions. I did a
[3:56] four-day
[3:57] calculation. Seven of that
[4:00] 12% occurred over four
[4:01] days from Friday
[4:02] last week to
[4:04] Wednesday this week.
[4:06] So, Friday,
[4:07] Monday, Tuesday,
[4:08] Wednesday—the four-day
[4:10] shift was the worst ever
[4:11] . And this
[4:12] includes the COVID period,
[4:14] when everything collapsed
[4:15] at the market level.
[4:16] Again, markets are
[4:18] falling. This
[4:19] occurs in
[4:20] conditions where the S&P is
[4:21] flat and the VIX is
[4:22] low. This
[4:23] concerns the
[4:25] market structure. So, we started
[4:26] getting these
[4:27] stories. This actually
[4:28] came out on Thursday
[4:29] morning. Hedge funds
[4:30] are facing
[4:31] collateral requirements
[4:32] as
[4:33] their stocks fall.
[4:34] So, at this point,
[4:37] news began to circulate that there
[4:38] was a problem. So, it's
[4:41] Thursday morning
[4:42] before the
[4:43] market opens. This is in the FT, so it
[4:44] was European
[4:46] time. And, obviously, something is
[4:48] happening. And then the
[4:49] article said that
[4:51] Goldman Sachs and JP Morgan were
[4:52] among the banks
[4:53] asking customers
[4:54] to provide additional
[4:56] collateral, according to
[4:57] several people
[4:58] familiar with the
[4:59] situation. Additional
[5:02] collateral was
[5:02] automatically
[5:03] triggered by
[5:04] market volatility.
[5:06] Goldman noted in a
[5:07] recent note
[5:08] that the build-up in
[5:08] gross leverage
[5:09] in the first 5 months of
[5:10] the year was the largest
[5:11] cumulative increase
[5:13] they have
[5:13] recorded since they
[5:14] began tracking
[5:15] the data in 2016. And it
[5:16] was already at
[5:17] historic
[5:17] highs, according to the
[5:18] data I'll
[5:19] show you later. In a
[5:21] sign of trouble for
[5:22] hedge funds,
[5:23] long- and
[5:24] short-term
[5:25] investment strategies fell
[5:26] more than 1%.
[5:26] Multi-strategy
[5:28] funds fell by more than
[5:29] 1.7%. Overall, the
[5:30] last time all
[5:31] strategies fell
[5:32] more than 1% in a single
[5:34] day was during COVID.
[5:35] However, and this is probably the
[5:36] most important point
[5:38] I want
[5:39] you to think about—
[5:40] something I told my
[5:41] subscribers about in my
[5:42] midweek market review
[5:43] —hedge
[5:45] fund strategies are
[5:46] still averaging
[5:48] over 10% growth per
[5:49] year. Although it seems that
[5:50] this data is not as of
[5:51] Wednesday. I think this is data
[5:53] as of the previous
[5:54] Friday. Like I
[5:55] said, it was a
[5:56] brutal week, but then
[5:57] there was a rebound. So,
[5:58] let's say they
[6:00] grew a little less
[6:01] than 10%, but that's
[6:02] still a significant
[6:03] figure. I often
[6:05] talk about this. In such
[6:06] situations,
[6:07] the size of your
[6:08] capital is important. So, when
[6:09] your capital is still in
[6:11] the black, you are much
[6:12] more likely to
[6:13] simply hedge your
[6:14] positions and
[6:15] come back to the game
[6:16] another day. Now
[6:18] about the situation with Leopold.
[6:22] They are looking to
[6:22] raise capital
[6:23] after the collapse of AI stocks
[6:25] . So, this is
[6:26] from the Financial Times again. And on Thursday at
[6:31] around 11:30 a.m., we
[6:32] learn that Citadel is
[6:34] buying a portfolio of
[6:35] Situational Awareness shares after
[6:37] big losses in the
[6:38] AI ​​space. There is a lot to
[6:42] talk about here. I
[6:45] just want to point out
[6:47] that it is these kinds of events that
[6:50] usually mark the
[6:51] bottom being reached. Do
[6:54] I believe there are many
[6:55] other funds that
[6:56] may be
[6:58] in a similar situation?
[6:59] Yes, there probably is.
[7:01] The difference is that,
[7:03] firstly, this fund is very
[7:04] large. And secondly,
[7:05] while I'm
[7:06] describing this, I want
[7:07] you to just pay attention
[7:08] . The Citadel is obviously
[7:10] much larger.
[7:13] Situational Awareness was reported to have
[7:15] $45 billion in positions. I'll
[7:18] go into some
[7:19] details, but these are two
[7:20] separate articles that
[7:21] I want you to
[7:22] know about. Because
[7:25] based on my
[7:26] experience, I believe there is a
[7:27] high probability
[7:28] that this was a
[7:31] market "cleaning" event, even if there are,
[7:32] say, 100 other hedge
[7:33] funds in the same
[7:35] situation as Situational
[7:36] Awareness. But the question is
[7:37] , if you take 100
[7:38] hedge funds worth a
[7:39] billion dollars
[7:40] each and
[7:40] analyze what
[7:41] they've already hedged,
[7:43] does it really
[7:43] matter that much? And I will answer:
[7:45] no. The hedge fund works with
[7:47] prime brokers
[7:48] including Bank of America,
[7:49] Goldman Sachs and JP Morgan to
[7:50] meet
[7:51] margin requirements
[7:52] or
[7:53] close out positions in an orderly manner.
[7:54] The main thing here is that
[7:55] several prime
[7:56] brokers know what is
[7:57] happening. Banks
[7:59] were
[8:00] marketing—that's the
[8:00] key word—
[8:01] marketing a portfolio of
[8:03] long and short
[8:04] hedge fund positions
[8:05] even before the market opened
[8:06] on Thursday. I was very
[8:10] closely involved with LTCM,
[8:11] both six months
[8:12] before and during the
[8:14] closure; I took over
[8:15] the management of the
[8:16] S&P portfolio before
[8:17] it
[8:18] was even formed. I've
[8:22] seen this since LTCM
[8:23] , but I was also
[8:24] heavily involved in the
[8:27] GARP quant fund sell-off in 2007
[8:29] before the great
[8:30] financial crisis, as well
[8:32] as the Archegos affair.
[8:33] There were many other
[8:34] cases, but these are the
[8:35] main ones. I don't
[8:38] think Melvin Capital is the same
[8:39] case, but
[8:40] you're faced with the fact
[8:41] that once the
[8:43] portfolio marketing starts and
[8:44] they're looking for
[8:45] buyers, you're essentially
[8:46] showing people someone else's
[8:48] poker cards, and they
[8:49] go into the market to
[8:50] hedge their
[8:51] positions, play
[8:53] ahead, or
[8:54] do whatever they want.
[8:56] In another case,
[8:57] multi-strategy
[8:58] hedge fund Millennium Management
[9:00] also filed.
[9:01] So when you
[9:02] combine Millennium and
[9:03] Citadel, these firms together
[9:07] hold trillions of
[9:08] dollars in various
[9:09] positions. So, of course,
[9:13] if they saw what was
[9:15] happening, and
[9:16] given that
[9:17] a lot could happen,
[9:18] the reality is this: this
[9:20] drop that we
[9:21] saw happened in just the
[9:22] last 4 days. And the
[9:24] reason I brought up
[9:25] this graph is because
[9:26] it's Bloom Energy
[9:28] relative to Adobe. And, in
[9:29] fact, when you
[9:30] analyze their
[9:31] positions, the largest of
[9:32] them, those rumored short
[9:33] positions
[9:34] , are a
[9:35] pretty good
[9:36] indicator of their
[9:37] portfolio. I mention
[9:38] this because they
[9:39] go around offering
[9:40] assets because they have already
[9:41] lost a significant
[9:42] portion of their profits. And
[9:43] when you give away a
[9:44] significant portion of your
[9:45] profits, even
[9:46] if they're up 80%
[9:47] year-to-date,
[9:48] with half or
[9:49] three-quarters of that
[9:50] coming from their
[9:51] private stake in
[9:52] Anthropic, you're still
[9:53] getting close to the point where you
[9:54] can't
[9:55] find
[9:55] the cash because the
[9:56] Anthropic deal isn't going to happen.
[9:58] So they are forced
[9:59] to look for someone to
[10:00] buy their
[10:00] portfolio and provide the
[10:01] money needed to
[10:02] cover
[10:03] margin requirements.
[10:04] So you get
[10:05] this massive
[10:06] sale, and right
[10:07] after that, on the
[10:08] same day,
[10:09] there's a reversal of the
[10:10] situation. You have experienced a
[10:12] cleansing event. So, in
[10:14] my opinion, given
[10:15] that Bloom Energy lost
[10:17] everything it had by the
[10:18] time it filed its
[10:19] last
[10:20] 13F—and this is where
[10:21] we saw the
[10:23] Situational Awareness positions—I
[10:24] think you have to
[10:25] start with the question: is
[10:27] this
[10:28] sell-off process complete? This is
[10:31] definitely not the end
[10:32] for those who
[10:33] need to get out. They
[10:36] never
[10:37] end, but
[10:38] the question is whether there are
[10:39] more people now
[10:40] who have already insured
[10:41] their portfolios,
[10:42] especially when you are
[10:43] dealing with prime
[10:44] brokers and the
[10:45] largest banks,
[10:46] which have already demanded
[10:47] more collateral, as well
[10:48] as the two
[10:49] largest hedge
[10:50] funds in the world, who
[10:51] also see the situation and
[10:52] have corresponding
[10:53] positions. So I did
[10:56] this for a week and I
[10:57] just want to
[10:58] mention it. This was a
[10:59] particularly important
[11:00] week for
[11:00] risk management: the
[11:01] Federal
[11:02] Reserve, reports from
[11:03] major
[11:03] hyperscalers, and
[11:04] the end of a
[11:04] brutal month
[11:05] for impulse
[11:06] investing. This
[11:07] made things
[11:08] worse, and to be honest, that's what
[11:10] probably
[11:11] pushed
[11:12] Leopold over the edge: a big week with the Fed was coming up at
[11:13] the beginning of the week, after
[11:14] last Friday
[11:15] when the problem became
[11:17] apparent
[11:19] . I'm not a
[11:23] conspiracy theorist, but the
[11:24] fact that Citadel
[11:25] released the news of a
[11:26] surprise
[11:27] rate hike as
[11:28] their forecast for
[11:29] Kevin Varsh, plus
[11:30] reports of hyperscalers
[11:32] potentially cutting
[11:33] capital expenditures—
[11:34] these rumors were spreading, and it
[11:35] was already a
[11:36] tough month, so people are
[11:37] playing with limited
[11:39] resources. That's why I
[11:42] analyzed it
[11:43] midweek. So,
[11:44] for those who are
[11:45] subscribed, I've
[11:46] been doing a
[11:47] midweek review,
[11:48] it seems, every week in
[11:49] July, or at least for the
[11:51] last 3 weeks. And
[11:54] the reason is that there was
[11:55] a lot going on
[11:56] , and also
[11:57] a lot in the thematic
[11:58] portfolio, which I,
[11:59] of course, posted on the
[12:01] website. It's part of it
[12:02] because it
[12:03] correlates with everything that's
[12:04] happening. So
[12:05] it was Citadel Securities
[12:06] that expected
[12:07] Warsh to announce a
[12:08] surprise
[12:09] rate hike.
[12:10] So, putting it all
[12:12] together, we
[12:12] go back to what
[12:14] I saw with LTCM. I
[12:16] saw the quantum
[12:17] rollout in 2007
[12:19] , was involved in
[12:20] it, and had positions in
[12:21] both cases.
[12:23] I wasn't involved with Archegos,
[12:24] except that I
[12:25] had about 30
[12:26] portfolio
[12:27] managers
[12:28] reporting to me, and
[12:29] they all ended up in the
[12:31] same situation where
[12:32] you get a call and
[12:33] margin demands from
[12:34] multiple global
[12:35] investment banks
[12:37] . They all know what's
[12:38] going on. In this
[12:39] case,
[12:40] block deals began to take place
[12:41] . For me, it's a
[12:42] mixture of all of these. I've
[12:45] said before that the
[12:46] market moves
[12:47] remind me
[12:48] more of LTCM than anything
[12:49] else, as well as
[12:50] quantum rollover
[12:52] . LTCM is different
[12:54] in that
[12:55] leverage affected the
[12:57] world, and all markets around the
[12:58] world, except for the
[12:59] US markets, were already
[13:00] falling. In this
[13:01] case, almost all
[13:02] markets in the world
[13:03] are above
[13:04] their 200-day
[13:05] moving average and
[13:06] continue to rise
[13:07] . So, one of the
[13:11] elements that
[13:11] came out last
[13:13] week that was a bit of
[13:14] an indicator was that
[13:15] someone—whether it was
[13:16] Leopold or in
[13:17] preparation for
[13:18] how they were going to
[13:20] sell it—bought
[13:22] billions of dollars of put options on SMH as a
[13:23] hedge. So, SMH
[13:28] plummets in
[13:29] four days, and
[13:30] guess what? You
[13:32] see this massive
[13:34] spike in volume
[13:34] actually at
[13:35] the lows. Um, I,
[13:38] you know, for anyone
[13:39] looking for any
[13:40] signs here, and again
[13:42] , this is actually a
[13:43] great chart,
[13:44] given the magnitude of the
[13:45] reversal that has occurred.
[13:48] I've said it before,
[13:49] at this point it's ABC.
[13:51] Um, minimums are critical for me
[13:52] , and I'll
[13:54] talk about that too. But I
[13:55] mentioned Micron
[13:56] last week. Here is
[13:59] the neck line of the
[14:00] head and shoulders pattern,
[14:02] below which we fell and
[14:03] now have broken through
[14:04] it again. That was the
[14:09] most interesting part, the
[14:10] day before
[14:11] the news came out, it was the second
[14:12] day in a row that
[14:14] people called me and said that
[14:15] someone was closing
[14:16] positions and
[14:17] something
[14:18] big was happening. So, it's the day
[14:21] after Kevin Warsh.
[14:23] So, it's Wednesday. The market
[14:25] was selling out. The S&P
[14:26] fell 120 points,
[14:27] which it fully
[14:28] recovered the next
[14:30] day. This was clearly the
[14:32] day the positions
[14:33] were taken over, I
[14:35] assume, by the Citadel. And
[14:38] they take these
[14:39] positions on themselves, and a
[14:41] massive futures
[14:42] position opens up. And the reason
[14:46] you can see that is because
[14:49] this is the total
[14:50] trading volume of E-mini,
[14:51] electronic
[14:52] futures on the S&P since
[14:54] the beginning of the year, from 2:00 to
[14:55] 4:00, which is the
[14:56] close, except for
[14:58] one day, June 4th,
[14:59] when there was a big gap
[15:00] down and a lot of
[15:02] trading volume. This was the
[15:06] highest volume for
[15:07] E-mini futures in this 2-
[15:08] hour period
[15:09] before the close
[15:10] all year. That's how you
[15:12] know that a large
[15:13] portion of the sales
[15:14] were made
[15:15] by dumping
[15:16] assets simply because of
[15:17] hedging what
[15:18] happened. So when
[15:20] someone comes into the
[15:21] market, and let's say the
[15:23] reports are correct, and
[15:24] they sold about $15-20 billion worth of positions in
[15:26] the market
[15:28] , driving the price down by a
[15:29] certain amount, well, it's no
[15:30] wonder that futures
[15:32] fell by a little over one
[15:33] percent. Well, they
[15:36] probably
[15:36] bought everything up
[15:38] when the index dropped
[15:40] 3%, 4%, 5%. So, I did the same thing
[15:42] when
[15:43] I traded the S&P portfolio
[15:44] . Um, these things
[15:45] happen, you
[15:46] create markets for them
[15:47] , and that's what
[15:48] happens, and I
[15:49] think that's what
[15:50] happened. Now, that doesn't
[15:51] negate the fact
[15:52] that as of
[15:53] Thursday morning, before
[15:55] we got the
[15:55] rebound, that's where the
[15:57] momentum factor was for the
[15:58] tech
[15:59] sector. A 40% drop
[16:02] since the beginning of the year. Worse
[16:05] than anything during the "
[16:06] dot-com bubble" by
[16:07] almost 20%,
[16:08] except for this
[16:09] one period here. I,
[16:11] I mean that
[16:12] one month there. So, again
[16:14] , a
[16:15] historic month
[16:16] for industrial
[16:17] momentum. Again
[16:20] , by the
[16:22] market opening on Thursday, a
[16:23] 32% drop. I
[16:24] mean, twice as much
[16:25] as any other
[16:26] month. This is the
[16:28] most important point
[16:30] , especially for you
[16:31] multi-strategists, but
[16:32] also for anyone who
[16:33] focuses on
[16:34] factors. If you
[16:36] remove factor
[16:37] risk from your
[16:37] portfolio, you
[16:38] are managing a factor-
[16:39] neutral
[16:40] portfolio, which
[16:41] will also be sector-
[16:41] neutral. This is
[16:42] sector-
[16:43] neutralized
[16:44] momentum. Again
[16:46] , as of
[16:48] Thursday morning, a
[16:50] 21% drop. Um, now again,
[16:51] when you look at the
[16:52] growth months, these
[16:53] were unprecedented
[16:54] growth months. This is an
[16:55] unprecedented
[16:56] month of decline. So again
[16:57] , we
[16:58] give away profits, and that's how it
[16:59] is. If you
[17:01] stretch the rubber band too much in
[17:01] one direction, you
[17:02] get a backlash. This is the
[17:03] reason why I
[17:04] did a review of the
[17:04] AI ​​slowing down
[17:05] mid-cycle,
[17:06] because we
[17:07] stretched the rubber band too much in
[17:08] one direction. I
[17:08] got it for getting
[17:10] out of Micron too
[17:11] early, and then I
[17:12] got it last
[17:13] week for
[17:14] saying I was buying Micron
[17:15] below 900. I
[17:17] 'll go through
[17:18] this as it
[17:19] goes along, guys. When you
[17:20] see movements like that,
[17:21] if you think you
[17:22] can guess the
[17:22] tops and bottoms, I've
[17:24] never been able to
[17:25] do that in my life. What I
[17:26] try to do
[17:27] is increase my
[17:28] positions based
[17:29] on a Bayesian
[17:30] philosophy of the news
[17:31] coming to me.
[17:32] If you are watching this
[17:33] to try to
[17:34] guess the tops and
[17:35] bottoms, especially in the world of
[17:36] fast trading, it
[17:37] won't happen.
[17:38] It's already
[17:39] overstretched down here. It
[17:40] was overstretched
[17:42] down here. It was
[17:43] overstretched here.
[17:44] You try to
[17:45] stay
[17:45] solvent
[17:47] while waiting for the panic
[17:48] to subside, and we saw a
[17:49] panic situation that
[17:50] historically signals the
[17:52] end. From the
[17:53] perspective of those four
[17:54] days, it's even more
[17:55] shocking. This is a
[17:57] four-day
[17:58] rate of change,
[17:59] again, to
[18:00] Wednesday's close for
[18:01] sector-
[18:02] neutral
[18:03] momentum. So,
[18:05] again, this is a "long-
[18:07] short" basket with a
[18:08] 17% drop that is
[18:10] sector-neutral
[18:11] . So, this is momentum across
[18:13] all sectors. This
[18:14] simply shows
[18:15] the extent of the damage. I've
[18:16] talked about this
[18:17] before, and that's why everything
[18:21] has changed for the
[18:22] future. This is data
[18:24] as of Friday's close
[18:25] . This is the current 30
[18:29] -day and 60-day
[18:30] realized
[18:31] volatility of the
[18:33] Morgan Stanley Technology Momentum Factor.
[18:35] There is no
[18:36] way this won't
[18:37] keep risk at
[18:38] lower levels. So,
[18:40] even if people
[18:41] have reduced their risks, they
[18:43] will not
[18:43] return to
[18:44] previous levels.
[18:46] Partly because of
[18:47] the drawdowns, partly
[18:48] because some of
[18:49] them will end up at
[18:50] a loss. But a lot of it is
[18:52] because until
[18:53] volatility comes
[18:54] down, and I don't expect it to
[18:55] come down quickly,
[18:56] I still believe that
[18:57] structurally it will be
[18:58] higher than in the past.
[19:00] This is the main problem.
[19:02] This is something incredible.
[19:04] Being at 110 when
[19:05] previous highs
[19:07] were around 70
[19:08] means that
[19:09] the normal movement
[19:11] in the market right now, if you take the
[19:13] square root of
[19:15] 252, 108 divided by the
[19:16] square root of
[19:18] 252, which is 16,
[19:19] gives about 6-7% per day
[19:21] . And it's market
[19:23] neutral in the
[19:25] sense that it's a dollar
[19:26] long and dollar
[19:27] short position. Such
[19:28] movements are huge.
[19:31] Now, going further,
[19:32] I took all the market-
[19:33] neutralized
[19:34] factors. Forget
[19:37] just the
[19:38] technology
[19:39] momentum side of things
[19:40] , I took
[19:42] beta, I took quality, I
[19:43] took momentum and I took
[19:44] cost. And I just
[19:46] went through them and
[19:47] gave them equal weight.
[19:48] So, divide each
[19:50] of them by four, that's
[19:51] their 60-day
[19:52] volatility. And that's
[19:54] where we are now, 38. We're higher
[19:56] . So this applies to
[19:57] any
[19:58] optimized
[19:59] portfolio where you
[19:59] weigh the risk of
[20:00] even your
[20:01] factors. This is something that has been
[20:04] trending
[20:05] upwards for some
[20:06] time now. Just
[20:07] look, you won't
[20:08] see anything where
[20:09] these things have been growing for a
[20:10] long
[20:11] time. This is the impact of AI, and it's
[20:13] not going away,
[20:14] and I'm going to
[20:15] break down the reasons
[20:16] why. Here is more
[20:18] information about
[20:19] the liquidations that
[20:20] have occurred and some more
[20:21] statistics. Hedge
[20:22] funds just sold
[20:22] more technology
[20:23] assets than at any
[20:24] time in Goldman's 10-year
[20:25] history.
[20:26] Momentum factor volatility is at its
[20:27] highest level in 45
[20:28] years,
[20:28] excluding recessions.
[20:29] I don't even have
[20:30] data for the last 45
[20:32] years. Gross and
[20:32] net exposure have
[20:33] fallen from five- and
[20:34] four-year
[20:35] highs, and are now in
[20:35] the middle.
[20:36] Six weeks of this,
[20:37] and hedge funds are still up
[20:38] 9% as of
[20:39] June. Again,
[20:41] as of June, this is
[20:42] one of the best
[20:42] first half years in 20
[20:43] years. So, again
[20:44] , this was as of
[20:45] June. I want
[20:46] to remind you, friends,
[20:48] that I wrote this
[20:48] article back in May.
[20:50] Bubbles, parabolas
[20:51] and rapid collapses. That's
[20:52] exactly what we
[20:53] saw. So I
[20:54] wrote this before the peak.
[20:56] We will see even more
[20:57] bubbles. We
[20:58] will see more
[20:59] parabolas, and with
[21:00] them, rapid collapses
[21:01] . So, it will
[21:02] look like a
[21:03] bubble. Essentially, we are
[21:04] compressing time. And
[21:05] the reason we
[21:06] compress time
[21:07] is because AI
[21:08] compresses time.
[21:09] The capabilities of models
[21:11] are growing so rapidly,
[21:12] and we are entering a world of AI
[21:14] agents that
[21:14] work 24/7. This is
[21:16] the end for people.
[21:18] However, this is what
[21:18] happens; If you've
[21:20] been to the New
[21:21] York Stock
[21:22] Exchange at any time since the 1990s,
[21:25] it looks very
[21:26] different today. We have already
[21:27] found ourselves in such a
[21:28] situation. Machine
[21:30] learning has already become a
[21:31] part of our
[21:31] world. It's just that now
[21:33] it's everywhere, and
[21:34] everyone has it. You
[21:35] have to take into account
[21:36] that this is happening.
[21:37] Now we
[21:37] have seen not only
[21:39] liquidations for hedge
[21:40] funds. In Korea, we
[21:42] actually
[21:43] saw a
[21:44] three-week rollback
[21:45] . So, again
[21:47] , we have cleared
[21:48] the positions, but you will
[21:49] not be able to
[21:50] take that volume
[21:51] again. So don't
[21:52] expect everything
[21:53] to go back in the same
[21:54] direction, because as
[21:55] someone who focuses a lot
[21:56] on
[21:57] Bitcoin and crypto,
[21:58] I will say: you need
[21:59] the energy of retail
[22:00] investors, and we are
[22:02] not going back to that anytime soon.
[22:03] I've talked about this in relation to
[22:04] silver, gold, and
[22:05] Bitcoin. The difference with
[22:06] AI deals, for
[22:08] those of you who I tell
[22:09] this to every week, is this
[22:10] . Most positions are still
[22:12] above the 200-day
[22:13] moving average.
[22:14] So, from the AI ​​positions,
[22:16] I think 74% are still higher,
[22:18] and 84% of my basket still
[22:20] has a 200-day moving
[22:21] average that
[22:22] continues to rise.
[22:24] So we are not in a
[22:25] situation where
[22:26] any harm has been done.
[22:27] Hedge
[22:29] fund leverage. This is data
[22:30] as of the third
[22:32] quarter of 2025 from the
[22:34] Securities and Exchange Commission's Form PF
[22:36] . You can see
[22:38] how much it has
[22:39] grown after COVID. Hedge
[22:41] funds showed
[22:42] good results. It
[22:44] went up. It's not
[22:46] just fundamental
[22:46] trading, it's also
[22:48] stocks. So it's not just
[22:49] fixed income, it
[22:50] applies to a wide
[22:52] range of markets. I want to
[22:53] mention this because I
[22:54] wrote an article about
[22:55] the art
[22:55] of unlearning at the Fed, which
[22:57] we'll
[22:57] come back to later, but I
[22:58] just want to read
[22:59] this line because it
[23:00] applies to anyone
[23:01] who works as a risk
[23:02] manager. If you're in a
[23:04] fund and your risk
[23:05] manager doesn't watch
[23:06] my videos, I would give it to
[23:07] him, because these are
[23:08] structural changes. AI
[23:09] is evolving at an
[23:10] unprecedented
[23:11] rate, and in a world
[23:12] defined by
[23:12] exponential
[23:13] innovation,
[23:14] relying on
[23:16] backward-looking models is too
[23:17] slow. This is what I wrote
[23:19] about what
[23:21] Scott Bessent said in an
[23:22] interview on the
[23:23] All-In podcast when he was
[23:24] dissecting what should
[23:25] happen to the Fed. Here's how
[23:28] it happens.
[23:30] First, AI compresses
[23:31] economic time.
[23:32] The capabilities of models are
[23:33] rapidly improving
[23:34] . So what used to take
[23:35] 10 years of
[23:36] innovation now
[23:37] takes a year. We
[23:38] went through this in the
[23:39] first quarter.
[23:40] Final costs
[23:41] are becoming more difficult to estimate.
[23:42] You don't know if
[23:43] you'll be in business
[23:44] in 3 years if you're
[23:45] built on code.
[23:46] Think Adobe.
[23:47] Think Salesforce. People
[23:48] still have no
[23:49] idea. So,
[23:50] there was a compression
[23:51] of the multipliers because
[23:53] we were subtracting the final
[23:54] cost. All these
[23:55] factors influence.
[23:57] If suddenly we don't
[23:58] know whether a company will survive
[23:59] , it can
[24:00] fall by 40% without any change in
[24:02] profits. This is bad
[24:03] for
[24:04] risk management in terms of
[24:05] such volatility
[24:06] . Hedge funds operate
[24:07] on a much shorter
[24:08] time scale than the
[24:09] assets they
[24:10] own. This will be a
[24:11] big deal,
[24:11] especially when we
[24:12] move to
[24:13] tokenization, where everything
[24:14] will be traded 24
[24:15] hours a day. Will
[24:16] hedge funds be
[24:17] able to trade this 24
[24:18] hours a day?
[24:19] How much more
[24:20] miserable would
[24:21] life become if you
[24:22] fell hard over the weekend?
[24:23] The longer the markets
[24:24] are open or
[24:25] trading, the more
[24:26] likely you are to
[24:27] react emotionally
[24:28] . This is bad.
[24:30] Leverage converts
[24:32] valuation uncertainty into
[24:33] liquidity risk. This is
[24:35] important because when
[24:36] confidence in the
[24:37] ultimate value falls, it
[24:39] increases
[24:39] volatility and
[24:40] tightens financial
[24:41] conditions. We see how
[24:43] all this unfolds.
[24:44] So I don't think
[24:45] prime brokers
[24:45] will allow
[24:46] leverage to reach
[24:47] previous levels because I
[24:48] don't think the
[24:49] volatility that we've
[24:50] seen is decreasing. The
[24:50] ubiquitous AI
[24:52] risk assessment tools create a
[24:53] false sense of
[24:54] confidence. It's like
[24:55] watching
[24:56] people drive down
[24:56] the street in
[24:57] four-wheel drive cars
[24:58] during an icy weather,
[24:59] thinking they're
[25:00] protected.
[25:01] Everyone now has AI
[25:02] tools for
[25:03] risk assessment. You
[25:05] can simply
[25:06] say, "optimize
[25:07] this portfolio, how can I
[25:08] reduce the risk?" You
[25:11] are now at a
[25:12] point where there is an
[25:13] optimization bubble and you
[25:14] think your
[25:15] portfolio
[25:16] is insured because, as
[25:17] Scott
[25:18] Bessent mentioned, the way
[25:19] to create an
[25:19] optimized
[25:20] portfolio is to
[25:21] use past
[25:22] correlations and past
[25:23] volatility. As I
[25:25] showed you with the
[25:25] example of factor
[25:26] analysis, this is no longer
[25:27] relevant
[25:28] unless you believe in
[25:30] mean reversion,
[25:31] and I am telling you that the
[25:32] market structure
[25:33] has changed forever
[25:33] because of AI.
[25:35] There is no turning back.
[25:36] Tokenization is here,
[25:37] terminal value
[25:38] is gone forever. All
[25:39] this is happening
[25:41] now. You see
[25:42] private companies
[25:43] like Anthropic growing
[25:45] very quickly, the
[25:46] fastest in the history of
[25:47] business. But you also
[25:49] see companies
[25:50] disappearing from the market
[25:50] faster than ever.
[25:52] Greater
[25:52] uncertainty about
[25:53] terminal
[25:53] value leads
[25:54] to compression
[25:55] of multipliers. This is what
[25:56] we are
[25:56] observing now. So,
[25:58] S&P 500 earnings
[25:59] are growing at over 20%
[26:01] per year. But
[26:01] the multipliers are
[26:02] getting squeezed
[26:03] because the shares are not
[26:04] growing
[26:04] proportionately to that.
[26:05] If you don't believe
[26:06] AI agents are here yet,
[26:07] look at how Robinhood is
[26:08] constantly releasing
[26:09] new tools. Here is an
[26:10] AI trader that
[26:11] reads charts, RSI, MACD,
[26:12] and 16 other indicators. This
[26:13] means that retail
[26:14] investors get
[26:15] access to the same
[26:16] tools as
[26:17] any quant
[26:18] hedge fund. Now we'll
[26:19] look at
[26:20] the changes that
[26:21] are happening and why
[26:22] it's catastrophic,
[26:23] especially for July,
[26:25] as I'm sure there are a
[26:26] lot of people. I get
[26:27] so
[26:28] many people coming to me
[26:29] saying they're buying
[26:30] put options on the S&P,
[26:31] buying put options
[26:32] on the Nasdaq. This all has to
[26:34] collapse. I'm not
[26:35] kidding when I say that
[26:36] various people I
[26:38] respect call me all the time
[26:39] with
[26:40] this question. Here's what's
[26:43] happening. This is a
[26:44] technology
[26:45] momentum, annual
[26:46] realized
[26:47] volatility at
[26:49] 67.5. Here is the
[26:50] realized
[26:51] volatility of the NDX over the
[26:52] same time period
[26:53] . Here is their
[26:54] relationship in time.
[26:58] This is madness. You
[27:00] are at
[27:01] one of the lowest
[27:02] levels of realized
[27:03] volatility in NDX,
[27:04] while
[27:04] exhibiting a
[27:05] parabolic
[27:06] rise in
[27:06] volatility. This is where
[27:08] independent
[27:09] investment advisors
[27:10] (RIAs), mutual funds, and
[27:12] anyone with a different
[27:13] time horizon than
[27:14] hedge funds need to have a
[27:15] dialogue with me.
[27:17] We have these
[27:18] conversations. This
[27:19] concerns the
[27:20] market structure. You can
[27:21] benefit a lot from
[27:22] this, and people
[27:24] I
[27:24] talk to regularly have done very well by
[27:26] reducing
[27:27] risk back in May,
[27:28] although it was early, because
[27:29] now they are able to
[27:31] increase risk again in
[27:32] fundamentally
[27:33] attractive assets.
[27:35] Guys, there will be more
[27:36] high-speed
[27:37] market crashes than ever
[27:38] before. If you don't
[27:39] believe this chart,
[27:41] here's that factor-
[27:42] neutral
[27:42] volatility
[27:44] I was talking about, it's at
[27:45] 36, and again
[27:46] , it takes into account
[27:47] equal weighting of
[27:48] quality,
[27:49] beta,
[27:50] momentum, and value factors, which are
[27:52] sector-neutral. Equal weighting
[27:54] to the sixth
[27:55] VIX contract. So for
[27:57] anyone who bought
[27:58] S&P volatility or 6-
[27:59] month contracts,
[28:00] that figure is
[28:01] usually going up, and
[28:02] right now we could
[28:03] be in a
[28:04] situation where
[28:05] multiple
[28:06] hedge funds are going bankrupt. If
[28:07] you ask me, as I've
[28:08] already learned from my
[28:10] conversations with the
[28:11] Fed's risk group, the only
[28:12] two areas that worry me are
[28:14] insurance companies and
[28:16] hedge funds. These are the only
[28:17] groups that I
[28:18] think have the leverage to
[28:20] drive this kind of
[28:21] growth. This would be a
[28:22] credit event. I
[28:24] think this should be
[28:25] about hedge
[28:26] funds, but because
[28:27] Citadel and Millennium have had
[28:28] significant growth in the
[28:29] year, and they are involved in
[28:31] these things and are the
[28:31] best risk
[28:32] managers in the world,
[28:34] we
[28:34] are probably in a
[28:35] situation where this will tend to dissipate over
[28:36] time
[28:37] , and that is my
[28:38] base case scenario.
[28:39] If you ask
[28:40] me what is most
[28:41] likely to happen, I
[28:42] don't think there will be
[28:43] any major event.
[28:44] I think there will be a
[28:45] compression
[28:45] of multipliers and
[28:46] a lot of disappointment,
[28:48] but I think
[28:48] deleveraging will last
[28:49] a long time because this
[28:50] indicator remains
[28:52] high. So, if
[28:53] you ask my
[28:53] intuition, I don't think
[28:54] we're going back to the
[28:55] mean.
[28:56] We are already seeing a
[28:57] structurally higher
[28:58] level than
[28:59] before, and this is not a
[29:00] short period of time.
[29:02] It practically
[29:03] remained within the range of
[29:05] 5 to 15 from 2000 to 2020. Since
[29:08] then, it
[29:09] has fluctuated between 15 and
[29:10] 30. We are in a
[29:11] structurally
[29:12] high
[29:12] volatility environment, so
[29:13] you will just
[29:14] have to
[29:14] accept it. Here is
[29:15] another example. This is the VIX, which
[29:17] is the implied
[29:18] volatility
[29:20] of the index, and this is the VIX EQ, which is the
[29:21] volatility of the stock.
[29:24] So,
[29:26] stock volatility has taken off. This is a
[29:27] long-term
[29:28] relationship between the
[29:29] two of them. And here we have a
[29:31] huge
[29:32] disagreement, which
[29:33] actually began around the
[29:35] advent of
[29:36] ChatGPT. So I just
[29:38] think that
[29:39] the volatility of stocks,
[29:40] because you can
[29:41] go bankrupt very quickly
[29:43] , will remain
[29:43] high, and the
[29:44] volatility of
[29:45] the index
[29:45] will remain
[29:46] relatively lower.
[29:49] Again, if you
[29:50] want to reach out,
[29:51] especially if you're the RIAs that
[29:52] I'm having these
[29:53] conversations with, you can
[29:54] do it weekly,
[29:55] monthly, but I
[29:56] think being up to
[29:57] date on factor
[29:58] analysis, understanding AI
[29:59] in a broader perspective,
[30:01] asking the right
[30:02] questions—that's where
[30:03] alpha is going to be generated.
[30:04] So you can
[30:05] contact me
[30:06] there. This is still a "
[30:08] bull" market. Yesterday, the
[30:10] S&P hit 67 new
[30:11] highs,
[30:12] the most since February.
[30:13] So, while all this was
[30:14] going on in July, we
[30:15] have this
[30:16] result. 72% of
[30:18] S&P 500 companies are above their 50-
[30:20] day and 200-day
[30:21] moving averages, 68% are
[30:23] above the 20-day. I
[30:24] mean, again
[30:26] , everything
[30:27] was fine this whole time. This is how we
[30:29] ended the month. No
[30:31] changes. Four sectors
[30:34] fell, seven rose. This is
[30:37] not like the
[30:38] LTCM situation or what
[30:39] happened during the
[30:41] quantitative fund sell-off. This is
[30:42] similar to the situation with
[30:44] Archegos. Archegos eventually
[30:45] went up after
[30:46] that, but I think you have to
[30:47] be careful.
[30:49] Here's the S&P chart, and
[30:50] remember, that was the
[30:51] day we
[30:52] sold off at the end of the
[30:53] day, and then not only did we
[30:55] recover right away,
[30:56] we
[30:57] went higher the next day and now
[30:58] closed well
[30:59] above that level. I'll
[31:00] go into some of the
[31:01] technicalities,
[31:02] but just keep in
[31:03] mind: every time we
[31:04] try to submerge
[31:06] this ball underwater, it
[31:07] pops back up. Here is
[31:08] the equilibrium S&P,
[31:09] monthly
[31:10] return.
[31:11] Look at how many
[31:13] months we went up,
[31:14] with only one
[31:15] month of decline—when there
[31:16] was an invasion
[31:17] or bombing of
[31:18] Iran and oil prices
[31:20] skyrocketed. This is a
[31:23] broad market in terms of the
[31:24] rally we are
[31:25] seeing. Here is the
[31:27] NDX index. We have reached the
[31:28] lowest RSI level
[31:30] in the last year.
[31:31] Look what
[31:32] happened last
[31:32] time we
[31:33] dropped to this
[31:34] level: we had a
[31:35] similar day, we
[31:36] fell, and the next
[31:37] day we opened with a
[31:38] gap up. They fell,
[31:39] opened up higher.
[31:40] Let's see what happens
[31:41] next. I'll go over
[31:42] some technical
[31:43] indicators, but
[31:44] remember that this is highly
[31:45] correlated with my
[31:46] thematic portfolio
[31:47] , AI trading in
[31:48] general, and
[31:49] semiconductors. The NDX
[31:50] really fell, and the NDX
[31:51] relative to the S&P actually
[31:53] had its worst
[31:54] month since the
[31:55] dot-com crash,
[31:57] believe it or not, guys.
[31:59] The worst month since the
[32:00] dot-com crash.
[32:04] Here is SMH. As I
[32:05] mentioned, we had
[32:06] this head
[32:07] and shoulders reversal. I tell you,
[32:09] again, these
[32:10] minimums have to hold
[32:11] up. But with an
[32:12] inverted
[32:13] head and shoulders pattern,
[32:14] when the neckline breaks through
[32:15] , everyone
[32:16] opens short
[32:17] positions. I'm sure
[32:18] many people here "
[32:19] shorted". And then we
[32:20] open with a
[32:21] gap up and
[32:22] stay there.
[32:23] Technically, that's what I
[32:25] would pay attention to. If
[32:26] you guys don't know
[32:28] about
[32:30] follow-through days, a
[32:33] follow-through day is a
[32:34] signal that a bounce
[32:34] after oversold conditions
[32:35] can turn into a
[32:36] real rally, not just a
[32:38] short-term
[32:38] closing of positions.
[32:39] So I
[32:40] like it, I
[32:41] used it and
[32:42] made a lot of
[32:43] money right after
[32:44] hitting rock bottom. It's
[32:45] easier for me to find
[32:46] the bottom than the top, because
[32:48] the mood is generally clearer there.
[32:50] You get very
[32:51] noticeable signals. I
[32:53] talked about everything that had been
[32:54] bothering me since
[32:55] May, and the market
[32:56] continued to rise
[32:57] . Situational
[32:59] awareness,
[32:59] bankruptcies, and everything
[33:01] I showed at the
[33:02] beginning are clear
[33:03] signs of
[33:04] market cleansing. Typically,
[33:05] technical analysts
[33:06] look for this a
[33:07] few days after the
[33:08] low is reached.
[33:09] This often
[33:10] occurs on the
[33:11] fourth to
[33:11] tenth day of a
[33:12] rally attempt. The main index
[33:14] is growing significantly. This is
[33:15] typical, volumes are higher and
[33:17] all that. So,
[33:18] essentially, you need to
[33:19] give the market time. That's the
[33:21] minimum. The next
[33:24] day we have the
[33:24] highest volume since
[33:26] June. This is by index
[33:29] Qs. So, we have
[33:30] a sale here. This is the same
[33:31] day we
[33:32] got the panic
[33:33] over Leopold. So
[33:35] , positions are traded,
[33:36] Varsh comes forward, and we
[33:37] get this
[33:38] massive sell-off.
[33:39] The next day we
[33:40] trade with higher
[33:41] volumes,
[33:42] closing above the
[33:43] highs of that day.
[33:44] Structurally it is good.
[33:45] Do I think we're
[33:46] likely to be
[33:46] stuck here?
[33:47] They say about the
[33:48] fourth to tenth
[33:49] day. So moving
[33:50] forward, if we
[33:51] hold the lows and
[33:52] just
[33:53] consolidate here,
[33:54] that's a good sign. This is
[33:55] exactly what I'm looking for.
[33:57] Let's see what happens
[33:58] next. If you're looking for
[34:00] fundamentals
[34:01] to
[34:02] support this, we've
[34:04] already gone through 60% of S
[34:05] &P reports on
[34:06] sales growth. Because the
[34:08] revenue growth is
[34:09] a little...no, not a little, but
[34:10] significantly distorted by
[34:11] the impact of Anthropic, but
[34:13] if you look at
[34:14] each of the companies,
[34:15] you can see how
[34:16] big these numbers are. This
[34:17] applies to every
[34:18] sector. More importantly,
[34:21] sales growth is 14%
[34:22] . I'm telling you guys, I
[34:24] keep
[34:25] repeating it. Being a "
[34:26] bear" on stocks
[34:27] when this
[34:28] happens is simply
[34:29] not profitable. You have
[34:30] strong
[34:31] fundamental
[34:31] support and
[34:32] forecast revisions
[34:33] throughout the
[34:34] reporting season. Again
[34:35] , the exchange rate is going up.
[34:36] So, not only do they
[34:37] outperform
[34:38] and
[34:39] demonstrate
[34:39] historical performance
[34:41] , but forecast revisions are also
[34:41] constantly
[34:42] increasing. It's not worth
[34:44] playing for a decline in
[34:45] stocks while this is
[34:46] happening. This is the
[34:47] longest period in the
[34:48] last 5 years. This is
[34:50] very difficult. So
[34:51] let's see what happens
[34:52] next. Here is my
[34:53] thematic portfolio.
[34:54] So, last week
[34:56] we were here. I said
[34:57] we were at the 50 mark.
[34:58] I wouldn't be surprised
[34:59] if we went lower,
[35:00] blah blah blah. And so it
[35:01] happened. We
[35:03] came up here again. We're
[35:04] somewhere in the 30
[35:05] something area. Is that all? As I
[35:07] said, right now in my
[35:08] own portfolio I'm
[35:09] betting that
[35:11] that panic low
[35:12] was the last one. If not,
[35:14] then something
[35:15] bigger is happening. Hedge
[35:16] funds may
[35:17] exit positions en masse.
[35:18] We'll see what happens, but that's why
[35:19] we're trading
[35:20] at these lows. We do
[35:21] n't sit back
[35:22] and invest
[35:23] in them. I have
[35:24] open positions and
[35:25] assets that I believe in, and
[35:26] they will remain. And
[35:28] also positions
[35:29] like Micron, AAOI, Marvell and
[35:31] some others that I
[35:32] added. Let's see how
[35:35] events develop. Here is a thematic
[35:37] portfolio based on RSI, the
[35:39] 14-day RSI has reached the same
[35:40] level as the
[35:42] lows here. Again
[35:44] , we'll see what
[35:45] happens, but right now I'm
[35:46] leaning towards
[35:47] this. Now, this is what
[35:50] you need to understand
[35:51] about building
[35:52] positions. Remember what
[35:55] I said about changing
[35:56] factors and
[35:57] market structure. In my opinion,
[35:59] analyzing the
[36:00] pace of change has never been
[36:01] more important.
[36:02] Most people
[36:03] I know don't
[36:04] use it. I've
[36:05] always used
[36:06] this since my days working
[36:08] in Brazil. I am
[36:08] convinced that markets
[36:10] move
[36:11] at the pace of change, and it is
[36:12] worth reducing
[36:13] risks when they
[36:14] exceed
[36:14] historical levels. In
[36:16] May and later, I was
[36:17] just reducing my
[36:19] position in Micron. So,
[36:21] starting in mid-
[36:22] May, a little bit all
[36:24] the time, a little bit of Marvell until
[36:25] we got here.
[36:26] This is a 63-day
[36:28] rate of change indicator. That's 63
[36:30] trading days, or 3
[36:32] months. We rose
[36:35] above 50% based on a
[36:35] retrospective
[36:37] analysis of the previous
[36:38] 63 days for the
[36:39] thematic portfolio
[36:41] . We are down 1.7% over the last 63 days
[36:44] . We
[36:46] have made a correction.
[36:48] We did what was
[36:49] needed, both in terms of cost and
[36:51] time, and that is what the
[36:52] analysis of the rate of
[36:53] change shows. When I
[36:54] was doing
[36:55] consolidation and
[36:56] showing you
[36:56] consolidation patterns
[36:58] that were likely
[36:59] to occur, it was a matter of
[37:00] timing and price. So, we
[37:01] did it. So this is
[37:03] just a metric for
[37:03] your use.
[37:05] I'll provide it every week in the
[37:06] technical report and
[37:07] show you some of the things
[37:08] that are there. On a 30-day
[37:10] basis, that is, over a
[37:11] month and a half or a
[37:13] quarter, a decrease of
[37:15] 15%. As my father taught me
[37:17] , the chances of making a
[37:18] deal with AI are much
[37:19] better than they were here.
[37:21] This is on a 30-day basis
[37:23] when I started
[37:23] to reduce my position
[37:25] in Micron. So again
[37:26] , you have a
[37:27] much better
[37:28] entry point in terms of price and
[37:29] time. I'm not trying
[37:30] to guess the minimums in
[37:31] this. There has been
[37:32] a lot of new news.
[37:33] Now, for those of you
[37:34] who want to trade
[37:35] the index, which I
[37:36] strongly
[37:36] recommend, if you're
[37:37] going to pick
[37:38] names, trade
[37:39] names, but if you're
[37:40] going to jump
[37:41] into things, into that
[37:42] concentrated
[37:43] position of 10 names, I
[37:44] just want to
[37:44] make sure you
[37:45] see that it was
[37:46] correlated before
[37:47] I spent a lot of
[37:48] time picking 10 names
[37:49] that are highly correlated
[37:50] with the 100-name index
[37:51] so you don't have to
[37:52] . It
[37:53] remained
[37:54] correlated with it and
[37:55] traded very
[37:56] well alongside it.
[37:58] One of the names there is Corning.
[37:59] I highlighted this
[38:00] in a subscriber video this week
[38:01] . Regarding
[38:03] Corning: it reached its 200-
[38:04] day moving average and
[38:05] went up. The reason
[38:08] I have this chart from December 31st is
[38:09] because that was the
[38:11] day Meta
[38:12] announced its
[38:13] $6 billion.
[38:16] They had another day
[38:17] when Nvidia became their
[38:18] partner. So, they
[38:20] announce a
[38:21] partnership with Nvidia,
[38:22] they announce a
[38:23] partnership with Meta for
[38:24] $6 billion, which
[38:26] , according to every LLM
[38:27] I use, is
[38:28] equivalent to
[38:29] about 100
[38:30] million
[38:31] households
[38:32] being provided with
[38:33] fiber. And the stock
[38:35] returned to the
[38:36] level where it
[38:37] actually was at the
[38:38] close that
[38:39] day. This is one of the
[38:40] concentrated names.
[38:41] Now, if
[38:42] you look at it from a
[38:43] course perspective...Well,
[38:44] I guess I didn't put
[38:45] that in there. I thought I had a
[38:46] calculation of the
[38:47] rate of change. Here are the
[38:48] percentage names.
[38:49] So, again, I
[38:50] show this to you
[38:51] every week. When you
[38:52] download the
[38:53] technical report, I
[38:54] highly
[38:54] recommend going into
[38:55] Co-Work and
[38:56] using it
[38:57] to search for it.
[38:58] The reason I
[38:59] highlighted this: On July 14th,
[39:01] we had 26% of the names above the 20-day
[39:02] average.
[39:04] Here we are. It's been
[39:06] two weeks, and even
[39:07] with all this chaos,
[39:09] we're starting to see
[39:10] signs of positivity
[39:11] compared to those
[39:12] two weeks. I'm more
[39:14] interested in this. It's a
[39:16] bit like this
[39:18] place. If we
[39:19] see it
[39:20] moving up again,
[39:21] I think that's a good sign.
[39:23] In 84 out of 100 names, the 200-
[39:24] day trend is
[39:26] upward. So you're
[39:27] looking for signs of this.
[39:30] I'm uploading this to the
[39:32] website. It's like
[39:33] sifting through ruins. So
[39:35] I went through it all
[39:36] , and I basically wanted to
[39:37] do something, and
[39:38] again, part of it is
[39:39] to teach you how to
[39:40] do it. Um, I highly
[39:42] recommend it, by the way, I also
[39:43] do this presentation
[39:44] for financial
[39:46] advisors so they
[39:47] can analyze
[39:48] someone's portfolio and
[39:49] show the assets that are
[39:50] technically still
[39:51] doing well.
[39:53] So, those are two conditions. I
[39:55] want to make sure that
[39:56] when I sift through the ruins,
[39:58] first, the price has to
[39:59] be above the 200-day
[40:00] moving average.
[40:01] Think about Micron, the 200-day
[40:02] moving average is somewhere
[40:04] around 500. We're still
[40:05] somewhere around 850. And I want the
[40:07] 200-day average itself to still be
[40:09] rising. These are the
[40:10] assets that I
[40:11] want to be structurally
[40:12] involved in. This is my
[40:13] favorite technical
[40:14] signal, so I want
[40:15] to find something that
[40:16] has pulled back, and
[40:17] I review and
[40:18] rank them all by this
[40:19] metric. You
[40:20] will see this list. It
[40:22] has all the names.
[40:23] So, to give you a
[40:25] quick note, as
[40:26] I mentioned AAOI, I
[40:27] ended up buying some
[40:28] this week. And again
[40:30] , when I say I
[40:31] bought a little, these are all
[40:32] small positions.
[40:32] Going back to what
[40:33] I had in
[40:34] May, my current
[40:35] position in Micron
[40:36] is about
[40:37] 1/5th the size it
[40:38] was in May
[40:38] when I exited it.
[40:39] So again, I'm not
[40:41] trying to catch the
[40:42] bottom. I'm trying
[40:42] to get into where
[40:44] I think the growth potential is
[40:45] good. Do I believe that
[40:46] at 800 it can go
[40:48] to 1600 within the
[40:48] next year? Yes.
[40:50] At 500, the risk and
[40:51] reward are better. So
[40:53] I'm not
[40:54] convinced yet that we've
[40:55] hit an absolute
[40:56] bottom, but I want to
[40:57] start buying because I
[40:58] think
[41:00] we'll eventually get back
[41:01] to 1500-1600. So in this
[41:02] case, where the slope of the 200-
[41:04] day average is
[41:05] up, and the percentage above the
[41:07] 200-day average is
[41:09] here, we're now
[41:10] down 31% for the month,
[41:12] down 27% for 3 months
[41:13] relative to the 50-day
[41:15] average, down
[41:17] 34. I'm trying
[41:18] to find names that are still
[41:20] higher here, still
[41:21] positive here, but
[41:22] significantly lower in these
[41:23] metrics. So I'm
[41:25] looking through the list. I
[41:26] also analyze what
[41:27] they said about
[41:28] their earnings to
[41:29] see if the
[41:31] fundamentals match up. Here are two
[41:32] things that I want
[41:33] you to start
[41:34] paying more
[41:35] attention to, and that I
[41:36] myself keep a close eye on.
[41:37] So, as a risk
[41:38] manager, I'm always
[41:39] looking for something that might
[41:41] make me change my
[41:43] mind about what's
[41:44] going on. So the
[41:45] credit default
[41:46] swaps (CDS)
[41:46] of hyperscalers
[41:48] don't bother me at all.
[41:50] What really
[41:51] worries me is the
[41:53] CCC bond market, where the spread,
[41:54] including
[41:55] options, looks
[41:56] very bad. And this
[41:59] is being imposed on the
[42:00] high-yield
[42:01] or “junk”
[42:02] bond market. So whenever
[42:03] I see these
[42:04] divergences, the
[42:05] credit market
[42:05] usually behaves
[42:06] like this: the weakest ones
[42:07] start to rise, and
[42:08] eventually that drags
[42:09] everything else down.
[42:10] There is no doubt that
[42:11] the amount of capital being
[42:13] absorbed by AI is
[42:14] enormous, and I believe
[42:15] that the changes brought about by
[42:17] AI are real. So I
[42:18] think we're going to
[42:19] see more and more
[42:20] problems with private
[42:21] lending.
[42:22] The question is whether
[42:24] these are the
[42:25] private
[42:25] lending issues that
[42:26] have been relevant this
[42:27] entire
[42:28] year, because it started
[42:30] right when the
[42:31] software sector collapsed.
[42:33] So I wouldn't ignore
[42:35] it. And if you want
[42:36] to read, I, I love a
[42:37] good conspiracy theory
[42:39] when it comes to
[42:39] loans.
[42:41] I like to keep my finger
[42:43] on the pulse because I believe that
[42:44] everything starts with
[42:45] isolated incidents.
[42:47] And the Dodgers bother
[42:48] me as a
[42:49] fan because of
[42:50] spending money in
[42:51] every
[42:52] way possible. And so
[42:55] this whole story with the Dodgers
[42:57] , Guggenheim, Mark
[42:58] Walter, and
[42:58] private
[42:59] lending—I would
[43:00] just read about it
[43:02] and pay attention. I, I
[43:04] know what that means. I
[43:06] know the
[43:07] insurance industry quite well and I
[43:09] believe that there are a lot of
[43:10] misvalued
[43:11] assets there. Uh, Google, I
[43:13] was shocked at
[43:14] how many people, when I
[43:15] made my video
[43:16] last week,
[43:17] told me I was
[43:18] wrong about Google.
[43:20] It was going to collapse, and
[43:21] of course it did here. Er
[43:23] , sorry, here. We
[43:25] closed below the 200-
[43:26] day moving average. Um,
[43:27] I added that last
[43:29] week. I'm shocked at
[43:30] how many people don't
[43:31] care about
[43:33] Google's order book and tell me
[43:34] it doesn't mean anything,
[43:36] that it's unrealistic. I, I
[43:37] couldn't
[43:38] disagree more, guys. Uh
[43:39] , I'm going to
[43:40] go over a bunch of things on
[43:41] this. I
[43:42] think you should
[43:43] stop reading
[43:44] some of the people who are
[43:44] just crunching the numbers,
[43:45] like I showed
[43:46] last week, and
[43:47] saying that these
[43:47] profits were
[43:48] terrible. Obviously,
[43:49] Google's profits
[43:50] weren't terrible. They have
[43:52] already gone up again
[43:53] . I fully expect
[43:54] this company to
[43:55] skyrocket and
[43:56] continue to grow
[43:57] because the backlog numbers are
[43:59] just insane. And
[44:00] if you didn't believe them,
[44:01] remember
[44:02] Zuckerberg's comments about the demand
[44:03] for computing
[44:04] power during
[44:05] his call. There is
[44:06] simply a
[44:07] catastrophic
[44:07] lack of computing power to
[44:08] meet all
[44:08] demand. We receive a
[44:09] huge number of
[44:10] offers for the
[44:11] capacities we have
[44:12] . Meta does not increase
[44:13] computing
[44:13] power in
[44:14] anticipation of
[44:14] uncertain demand
[44:15] . Zuckerberg said that
[44:16] demand already exists for
[44:17] existing capacity,
[44:18] blah blah blah. If you
[44:19] don't believe them because
[44:20] their stock has fallen,
[44:23] Sundar Pichai: We
[44:24] continue to be
[44:24] resource constrained
[44:25] due to significant
[44:26] momentum and rapid
[44:27] implementation. Demand
[44:28] still outpaces
[44:29] investment. Meta: There is
[44:30] simply,
[44:31] catastrophically
[44:33] not enough capacity for
[44:34] all the demand. Microsoft,
[44:35] whose shares
[44:36] were trading much
[44:38] higher. Customer demand
[44:39] continues
[44:39] to exceed
[44:40] available capacity.
[44:41] This additional
[44:42] quarterly
[44:43] Azure capacity was
[44:44] quickly monetized. I
[44:45] can't even
[44:45] tell you how many
[44:46] smart people
[44:47] told me that these
[44:48] guys were going to
[44:48] cut capital
[44:49] spending this
[44:50] quarter. Error.
[44:52] Every additional
[44:53] unit of
[44:53] computing
[44:54] power
[44:55] put into
[44:55] operation
[44:56] is absorbed almost
[44:57] instantly, while
[44:58] demand continues to
[44:58] outpace
[44:59] supply. If you
[45:00] don't like these
[45:01] hyperscalers, how
[45:02] about Amazon? Even with
[45:04] this amount, we
[45:05] still won't have enough
[45:06] capacity to
[45:07] meet all
[45:08] demand in 2026. I
[45:10] believe that this
[45:11] dynamic will continue
[45:12] in 2027. Guys, don't
[45:14] play the bearish game on
[45:15] these stocks.
[45:16] Honestly, what do you even
[45:18] do? Me, I don't like
[45:21] the demand we already
[45:23] have for 2028, it's
[45:25] impressive. This may be
[45:27] the strongest
[45:27] forward-looking comment
[45:28] from the call, because
[45:29] Amazon already sees significant
[45:30] liabilities for
[45:31] several years ahead.
[45:32] The lion's share of
[45:34] AWS's computing capacity for
[45:36] 2027 has already
[45:36] been reserved. We
[45:37] have a clear signal
[45:38] of strong
[45:39] financial
[45:40] results. Amazon
[45:41] also explained why
[45:42] free cash
[45:43] flow looks
[45:43] weak before
[45:44] revenue comes in. It
[45:45] starts to run out
[45:47] of data. So, again
[45:48] , I want you to
[45:49] go back to what
[45:50] I said about the
[45:51] risk factor.
[45:52] The possibilities of the models are
[45:53] crazy. They are
[45:55] developing
[45:56] faster and faster. This whole
[45:58] open source thing is
[46:00] stupid. Yes, that will be
[46:01] part of it. Open
[46:02] router will go through
[46:03] this, but they
[46:04] need the cloud, and they
[46:05] need these companies
[46:07] because they are the only ones with the
[46:08] computing
[46:09] power.
[46:10] Nobody builds
[46:11] computing
[46:12] power anymore. Last
[46:13] week I showed you
[46:14] that those Chinese
[46:15] companies are very small,
[46:16] but they are building
[46:17] capacity. They
[46:18] need to be
[46:19] launched somewhere. So where
[46:20] can you get the
[46:21] opportunity to
[46:22] use the
[46:22] computing
[46:23] power of these
[46:24] Chinese models?
[46:25] It should be on Bedrock, or
[46:26] wherever else, but it
[46:28] should be where there is
[46:29] computing
[46:30] power. Somewhere in
[46:31] the chain there has to be
[46:32] hardware
[46:33] with
[46:34] computing
[46:34] power. There is no
[46:36] computing
[46:37] power there. Here are the new
[46:38] numbers regarding the
[46:40] order book. 500
[46:40] billion for Amazon, 500
[46:42] billion for Google, 700
[46:44] billion. That's
[46:46] $1.7 trillion in
[46:47] order backlog
[46:49] for three. Our
[46:50] portfolio is 496
[46:51] , growing by
[46:52] triple digits year
[46:54] over year. A large
[46:55] part of it is the agreements and
[46:56] contracts you
[46:57] signed. They
[46:59] talked about the fact that the
[47:00] new demand is
[47:01] not coming from OpenAI
[47:03] and Anthropic. The majority of
[47:05] the portfolio
[47:06] is related to
[47:07] regular
[47:08] GCP contracts for a
[47:09] wide range of
[47:10] customers. Oh, sorry.
[47:12] This is it. All of the
[47:13] consistent growth
[47:14] in
[47:14] RPOs has been
[47:15] driven by
[47:15] orders from
[47:16] customers who are not
[47:18] leading-edge companies. Um
[47:19] , if the people who
[47:20] scared you aren't writing
[47:21] about facts from
[47:23] earnings reports, then their
[47:24] words aren't facts. What
[47:25] they are saying is
[47:26] an attempt to manipulate
[47:27] you with
[47:28] pessimistic
[47:29] predictions. You should
[47:30] at least figure this out
[47:31] . You may
[47:32] say these guys
[47:34] don't understand anything,
[47:35] but the "magic seven"
[47:36] are proven
[47:37] winners whose incomes
[47:38] are growing rapidly. They are the ones who
[47:40] spend the
[47:41] capital investment
[47:42] , and you mean to say
[47:44] that the authors of these
[47:45] spam emails
[47:46] think Google, Amazon, and
[47:47] others are fools? There is a
[47:50] risk of cyclicality and that
[47:51] Anthropic and OpenAI will
[47:52] go bankrupt, but
[47:54] we will talk about that later
[47:55] . The All-In podcast, I
[47:56] thought it was
[47:58] really, really good. I
[47:59] often criticize them because I
[48:00] think the show has become
[48:02] too
[48:02] politicized, but
[48:03] they've been
[48:05] back on form lately. There are
[48:06] many interesting things there. I'm not
[48:07] going to
[48:08] reread everything. Um,
[48:10] but they're taking
[48:11] Kimi K3 apart, they're taking
[48:12] it apart. I think David
[48:14] Sacks, in particular, has provided
[48:15] a lot of useful
[48:16] information on this
[48:17] subject. Just because
[48:18] he's connected to the
[48:19] White House, and I think
[48:20] he's been very
[48:21] open and honest
[48:22] about all of these things.
[48:26] He counters the panic
[48:27] and essentially says
[48:28] that as much as he
[48:30] hates Anthropic, they
[48:31] are the
[48:32] fastest-growing
[48:33] tech
[48:34] company in history, with
[48:35] annual revenue
[48:37] already
[48:38] hitting $74 billion mid-year
[48:39] . So, he
[48:44] says
[48:45] they will grow
[48:47] 10x again this year, to over
[48:48] a hundred billion, and
[48:49] notes that OpenAI is
[48:50] currently on track for
[48:52] about 75 billion
[48:53] this year. Guys, that's
[48:55] $175 billion in
[48:56] revenue from two
[48:57] companies that
[48:58] had zero two years ago.
[49:00] How can you argue
[49:01] about return on
[49:02] investment (ROIC) when
[49:04] these guys are printing
[49:05] money on one hand, and
[49:06] cloud services are printing
[49:07] $2 trillion over the
[49:08] same period? So
[49:11] where do people get this from
[49:13] , when companies are
[49:14] getting demand from both
[49:15] API and
[49:16] enterprise,
[49:17] at the same time as cloud
[49:18] giants. This is just
[49:20] completely
[49:21] misleading. So, demand for
[49:23] computing
[49:23] power versus
[49:24] supply—he
[49:25] breaks it down. To
[49:26] me, there is an
[49:27] unlimited appetite
[49:28] for intelligence on
[49:29] demand. I don't think
[49:30] there is an upper limit to
[49:31] how much
[49:32] intelligence can be
[49:33] used as long as it
[49:34] continues
[49:35] to improve. I
[49:36] completely agree. He
[49:37] crunches the numbers on
[49:39] both of these things. You
[49:42] still have developments
[49:43] in the Anthropic
[49:45] and OpenAI labs that are far
[49:46] ahead of the
[49:46] Chinese
[49:48] open source models. Sam
[49:49] is going to
[49:50] Washington
[49:50] next week to
[49:51] talk about GPT-6.0,
[49:52] which is simply amazing in its
[49:53] results. Again
[49:55] , the models that are
[49:56] already available are better from these guys
[49:57] . If you haven't
[49:58] seen Sam Altman
[50:00] on "Invest Like the Best," I
[50:01] highly
[50:01] recommend
[50:02] watching. I think it was
[50:03] one of his
[50:05] best performances.
[50:06] Altman says
[50:07] OpenAI's confidence in
[50:07] the need for
[50:08] massive
[50:08] computing
[50:09] power came
[50:10] after observing the
[50:11] exponential
[50:12] curve of
[50:12] model improvement, especially
[50:13] around GPT-4. Essentially,
[50:15] again, demand would be
[50:16] virtually
[50:17] unlimited. He says the
[50:19] biggest
[50:19] short-term
[50:20] gains in
[50:20] computing may
[50:21] come from
[50:23] software efficiency. Er,
[50:24] talking about
[50:25] the cyber incident that
[50:26] happened with Hugging Face. Uh, I'll
[50:30] leave AGI aside and
[50:31] go through the points.
[50:33] I'll summarize it a little
[50:34] better if you're
[50:35] interested, but, uh... the
[50:38] fear is gone because
[50:39] the demand has exploded. This is what
[50:41] I want to
[50:42] make sure of. As of
[50:44] last October,
[50:45] the likelihood of
[50:46] these companies
[50:48] getting into trouble
[50:49] was
[50:49] much, much
[50:51] higher to me than at any
[50:52] time this year. And
[50:55] the reason was that the
[50:56] risk was that
[50:57] these guys
[50:58] were spending money, and
[50:59] businesses were
[51:00] never going to
[51:00] implement it. That
[51:01] people will simply be
[51:02] too slow. The
[51:04] only thing that changed
[51:05] the situation was that AI
[51:06] agents appeared
[51:07] much faster. And that's why
[51:09] I built a
[51:10] portfolio—because the
[51:11] era of agents had begun,
[51:13] and I had collected all these
[51:14] names long before it all
[51:15] took off, because
[51:16] I had already done my
[51:18] homework. The demand for
[51:19] inference
[51:20] will become insatiable
[51:21] once AI
[51:22] agents appear. We just didn't
[51:24] know when it
[51:24] would happen, and it
[51:25] happened at least a
[51:26] year early. That's how
[51:28] Leopold invested
[51:29] in all these things. It just
[51:30] got to this
[51:31] point. Sam says that the
[51:32] main
[51:33] bottleneck today is the
[51:34] supply of
[51:34] computing
[51:35] power. Over time, the
[51:36] bottleneck
[51:37] moved. He
[51:38] says research
[51:39] is one of the main
[51:40] factors. The volume of
[51:41] demand will only
[51:42] grow.
[51:43] Personal agents will
[51:44] significantly increase
[51:45] demand. This is the view from
[51:46] Meta's side. That's why Meta
[51:47] doesn't sell its
[51:48] computing
[51:49] power. They
[51:50] could have made
[51:51] investors happy this
[51:51] week by saying they
[51:52] would
[51:53] lease out the capacity
[51:54] they supposedly had
[51:55] tapped, but they didn't
[51:56] . Why did
[51:57] n't they do this
[51:58] ? Because if they
[51:59] do,
[52:00] personal agents will
[52:01] dramatically increase
[52:02] demand. So I'm
[52:02] sure they
[52:04] understood: it's coming
[52:05] sooner than we all
[52:06] know. And I believe in it. The
[52:09] oversupply scenario
[52:09] is when
[52:10] models become
[52:11] so smart
[52:12] and efficient that
[52:12] they can do almost
[52:13] everything
[52:14] people need at current
[52:15] capacity. Altman
[52:16] believes that demand for
[52:17] computing exceeds
[52:18] supply because the quality of the
[52:19] models opens up
[52:19] new
[52:20] use cases. So,
[52:21] essentially, Jevons' paradox comes into play
[52:22] . Is
[52:24] there a chance? So, what is the
[52:25] most likely
[52:26] scenario of supply
[52:26] catching up with
[52:27] demand
[52:28] according to his
[52:29] views? Software
[52:30] agents squeeze
[52:31] more intelligence out of
[52:31] computing
[52:32] power. This is an
[52:33] aspect of efficiency.
[52:34] Robotics in the
[52:35] physical world. So
[52:36] , humanoids ultimately
[52:37] expand the entire
[52:38] supply chain of
[52:39] computing
[52:40] resources. Neither will
[52:41] happen
[52:42] next year. This is
[52:43] simply impossible. This
[52:44] ...this can't happen.
[52:47] We are not ready for
[52:48] this yet, and even if we
[52:49] were, demand
[52:50] would grow even faster
[52:51] due to personal
[52:52] agents. That's why I
[52:53] showed you a graph of what the
[52:55] emergence of personal
[52:56] agents means. So, agents
[52:58] make
[52:58] computation production much more
[52:59] efficient. Models
[53:00] are becoming so
[53:01] efficient that these
[53:02] tasks require
[53:03] less computation.
[53:03] Human attention is becoming a
[53:04] bottleneck. So, it's
[53:06] good to
[53:06] sort this out. I was talking
[53:08] about income.
[53:09] They
[53:09] have shown
[53:10] significant growth since May. This is
[53:11] another reason why it is
[53:12] worth having a positive
[53:13] attitude towards these
[53:14] things. Now that makes
[53:15] sense to me. And I
[53:16] hate to say it,
[53:17] but where would you expect
[53:18] to see them? These
[53:19] guys are raising so
[53:20] much money. If
[53:21] credit spreads weren't
[53:22] rising and they weren't
[53:23] accumulating more
[53:24] risk while all these
[53:26] guys are spending
[53:27] money, of course there's a
[53:28] risk that any of
[53:29] these companies could
[53:30] go bankrupt if they
[53:31] don't make a profit.
[53:32] Meta is making a huge
[53:33] bet. Now they have
[53:35] nothing. If
[53:36] they really don't
[53:37] deal with this, they'll be in
[53:38] trouble. Gavin
[53:39] Baker believes that
[53:40] the market is overreacting
[53:41] to the rise in
[53:42] hyperscaler credit spreads. The
[53:43] fact that spot
[53:44] GPU rental prices are
[53:45] at least twice as
[53:45] high as contract
[53:46] rates is the missing
[53:47] link. I could
[53:48] read all of this.
[53:49] The reality is: $60
[53:51] billion per gigawatt
[53:52] is $1.5 to $2.2
[53:53] trillion in
[53:54] capital expenditures,
[53:55] while the consensus
[53:56] forecast for
[53:58] hyperscaler cash flows
[53:59] is $1.3. It
[54:01] just confirms
[54:02] my point: we
[54:03] don't have enough
[54:04] computing
[54:05] power right now, guys.
[54:08] Dwarkesh posted
[54:09] this. I think it's gotten
[54:11] over a million
[54:12] views on X. I
[54:13] highly
[54:14] recommend
[54:15] checking it out.
[54:17] This is a very contradictory
[54:19] opinion. And without going
[54:22] into detail, the
[54:23] most important part is
[54:25] that
[54:30] his argument is
[54:31] that expensive computations
[54:32] turn
[54:33] token efficiency
[54:34] into a price advantage,
[54:35] expanding the premium of
[54:36] advanced models.
[54:38] So, this brings us back
[54:39] to the fact that if these
[54:39] companies
[54:40] continue to
[54:41] move at this
[54:42] pace, their
[54:42] intelligence
[54:43] will improve,
[54:44] but more importantly,
[54:45] the efficiency of the tokens will increase
[54:46] , which depends on
[54:47] the amount of
[54:48] computing
[54:49] power built.
[54:50] Thus, the computing
[54:51] power
[54:52] they have access to
[54:53] becomes their "moat."
[54:54] The fact that they
[54:55] can once again
[54:55] set an
[54:56] acceptable price for
[54:57] the efficiency of the tokens
[54:58] . So, to explain this in a
[55:00] way that people would
[55:01] understand what it
[55:02] meant, I
[55:03] had to do some
[55:04] analysis. And
[55:05] again, Gavin
[55:06] has also spoken out
[55:07] on this matter. I
[55:10] wanted to give you an
[55:11] analogy that might
[55:13] work. Gasoline and
[55:14] cars. Why the
[55:15] best AI model
[55:16] wins when
[55:17] computation becomes
[55:18] expensive. If gasoline
[55:20] costs a dollar a gallon
[55:21] , no one
[55:21] cares about
[55:22] fuel consumption.
[55:23] When it costs
[55:24] $10, everyone
[55:25] is willing to overpay
[55:26] for the car with the
[55:27] best mileage.
[55:28] Here's what will eventually
[55:29] happen: you can
[55:30] pay less for
[55:31] Kimmy K 3.0, but it's not
[55:33] as efficient as
[55:34] the best models, and
[55:35] the situation will
[55:35] get worse
[55:37] because you'll need
[55:38] more hardware or
[55:39] computing. And if
[55:43] there is no equipment or computing power to do the work, the only
[55:44] place they are is
[55:45] where
[55:46] the calculations are concentrated. If
[55:52] Anthropic has extraordinary
[55:54] token efficiency
[55:55] and Disney wants to reduce
[55:56] the cost of a film from
[55:58] $5 billion to $
[55:59] 100 million, then
[56:00] being able to buy
[56:01] all of
[56:02] Anthropic's computing power not
[56:03] only gives them a crazy
[56:04] way to make a ton of
[56:06] money, but also allows them
[56:07] to destroy the competition.
[56:11] So, we are coming to the
[56:12] point where you
[56:13] should pay
[56:14] attention to
[56:14] token performance
[56:15] . This is another reason
[56:16] why I have written so much
[56:17] about Vera Rubin.
[56:18] We find ourselves in a
[56:19] situation where the main thing
[56:20] is the efficiency of
[56:21] tokens, not their
[56:21] maximum
[56:22] number. And just
[56:23] think of how much SpaceX
[56:24] has been able to do
[56:24] thanks to its
[56:25] computing
[56:26] power. So, we
[56:27] have a shortage of
[56:27] computing
[56:28] resources. Models
[56:29] are evolving faster.
[56:30] They are creating
[56:31] more demand
[56:31] as they get
[56:32] better and better, but there is
[56:34] not enough computing power for everyone. So prices
[56:35] for those who have already
[56:36] built
[56:37] computing
[56:37] power will
[56:38] only increase, as
[56:39] SpaceX saw, and as
[56:40] Meta saw when it
[56:41] realized it could
[56:42] sell off these
[56:43] resources. So just
[56:44] keep in mind that
[56:45] we're reaching a point
[56:46] where it's great that there are
[56:48] a lot of competing
[56:49] models out there, but you have to be
[56:49] able to
[56:50] run them on
[56:51] something. And the question is
[56:52] , where will this
[56:53] power
[56:54] come from? And even
[56:55] if models become
[56:56] more efficient in terms of
[56:57] token usage
[56:58] , you will still
[56:59] end up back
[56:59] using the better
[57:00] model, except for the
[57:01] free
[57:02] content you
[57:03] download for yourself.
[57:04] So, we're not at the
[57:05] point where we have the
[57:06] hardware for everyone to be able to
[57:07] run
[57:08] open source models.
[57:09] So, in my opinion, there are at
[57:11] least 2 more years to go, and
[57:12] that's more than
[57:13] enough, based on what
[57:14] we've heard from
[57:15] hyperscalers this
[57:16] week regarding when
[57:17] they'll start
[57:18] making profits.
[57:19] This is not guaranteed.
[57:20] I
[57:21] gave you
[57:22] the probability last week, but it
[57:23] happens. Now,
[57:24] if you still don't
[57:25] believe in data centers
[57:26] and computing
[57:27] power, just
[57:28] look at
[57:30] DRAM prices. These are weekly
[57:32] DRAM prices. And again, this is the
[57:33] oldest indicator
[57:34] . It dates back to 2016. This is
[57:37] not the most recent memory.
[57:39] They are still growing every
[57:41] week. So,
[57:42] cyclicality,
[57:42] cyclicality: green,
[57:44] mostly red.
[57:45] Look how many
[57:46] weeks this has been going on and it has
[57:47] n't stopped. Yes,
[57:48] it is slowing down
[57:49] because the price
[57:50] has gone up very
[57:51] high. Okay,
[57:52] let's quickly
[57:53] go through the rest of
[57:54] them. I wanted to do
[57:55] something about the Fed because,
[57:56] again, a lot of
[57:57] people are discussing it
[57:59] this week. People
[58:00] tell me that some
[58:01] of the strategists
[58:02] even assumed 50.
[58:03] So he's holding his
[58:04] bets. Three
[58:05] dissenters. The Fed
[58:06] did exactly what
[58:07] Wall Street expected.
[58:09] So why are the markets so
[58:09] worried about
[58:10] Kevin Varsh? Again
[58:11] , "The Art of
[58:12] Unlearning What You've
[58:12] Learned About the Fed"—if
[58:13] you haven't read it,
[58:14] be sure to
[58:14] read it. I think
[58:15] we're back in that
[58:16] world again. I think
[58:17] Kevin Warsh is a
[58:18] smart person. I think
[58:19] he was publicly
[58:20] saying that we
[58:21] can't
[58:21] use the
[58:22] academic approach of
[58:23] the past. We
[58:24] need a working
[58:25] group to define
[58:26] new tools.
[58:27] The data we currently
[58:28] use is not
[58:29] only subject to
[58:29] revision, it is also
[58:30] outdated. We
[58:31] need to
[58:32] look more into the future.
[58:33] The Fed is not what it used to be,
[58:34] and this whole dichotomy of "
[58:35] doves" and "
[58:36] hawks" no longer
[58:37] works. Currently, 10-year
[58:38] rates are
[58:40] around 4.70. Here
[58:42] again are the 10-year
[58:44] inflation swap rates.
[58:45] So, at this point, since
[58:48] Kevin Warsh came in, we are here. I don't
[58:49] know how long it
[58:50] will last, but he
[58:51] said, "Well, the market
[58:52] did it for us. The
[58:53] real
[58:54] profitability is here.
[58:55] Inflation has decreased,
[58:56] so we can
[58:57] wait." Here are
[58:59] inflation
[59:00] expectations for
[59:01] August data.
[59:03] Close to zero.
[59:07] Dollar positioning. This is an interesting
[59:09] point, because when
[59:10] the dollar weakens, it is
[59:11] usually a very
[59:12] positive signal, which is
[59:13] good for gold and
[59:14] silver. Maybe
[59:18] Jackson Hole will affect
[59:19] this, maybe
[59:20] dollar-
[59:21] yen intervention, but look at
[59:23] how big this
[59:24] accumulation of
[59:25] dollar positions is. It is at
[59:26] the highest level in the
[59:27] last 4 years. And the
[59:29] dollar
[59:30] fell a bit this week,
[59:31] mostly because of that.
[59:34] Speculation about the
[59:36] Bank of Japan. And I
[59:37] just want to
[59:38] emphasize for those
[59:39] who are trading: let's
[59:41] see—end of
[59:42] April, end of July,
[59:43] end of January. So it
[59:45] looks like 3 months, 3
[59:46] months. In any
[59:48] case, he's not
[59:48] tied up, but it
[59:49] looks like he's on a
[59:50] leash: let's
[59:51] just make sure
[59:52] he doesn't
[59:53] get away. One
[59:55] podcast for you to
[59:56] pay attention to and
[59:57] somehow end this
[59:58] week. Hmm, I've heard a
[60:01] lot of people say
[60:02] about China that they have
[60:03] other plans for AI and
[60:04] they're looking to just
[60:06] make it
[60:07] free for the
[60:08] rest of the world. This is a
[60:10] great interview. Ezra
[60:11] Klein interviewed
[60:13] Kevin Rudd,
[60:13] former Australian Prime Minister
[60:15] and expert on
[60:17] Xi Jinping. I
[60:18] really enjoyed
[60:19] listening to him. I won't
[60:22] read everything
[60:23] that's written here, but
[60:24] for those of you who
[60:25] are trying to better
[60:27] understand
[60:28] China's plans, where the
[60:29] balance of power in the field of AI is now
[60:30] and how they
[60:31] perceive it. I
[60:34] think this is a very
[60:35] interesting point.
[60:36] I think Kevin Rudd
[60:37] did a great
[60:38] job. For those who are
[60:40] interested:
[60:40] the combination of Marxist-
[60:42] Leninist optics with the
[60:43] Schumpeterian side
[60:44] looks
[60:45] very interesting to me. This is my
[60:46] own addition, I'm kind of
[60:48] extrapolating what I heard,
[60:49] comparing it to
[60:50] my beliefs
[60:51] about Joseph
[60:52] Schumpeter. This is why
[60:55] Bitcoin is so important to me. For those interested,
[60:57] I think this is a great, uh,
[60:58] interview, worth paying attention to.
[61:01] A few points:
[61:02] cyber threats are becoming
[61:03] real. OpenAI and Hugging
[61:05] Face—you had those hacks.
[61:07] You also had a breakdown
[61:11] Claude too. It's just
[61:14] worth paying attention to.
[61:15] And here we have the
[61:17] Clarity Act. His
[61:19] support is still
[61:20] below 40%, depending on
[61:22] which side you
[61:23] listen to. Let's see
[61:25] what happens next. Again
[61:27] , I think this is
[61:28] important in the
[61:29] long
[61:30] run for
[61:31] Bitcoin and
[61:31] cryptocurrencies. This is
[61:33] certainly important
[61:34] for entrepreneurs
[61:35] working in this field,
[61:36] as well as for the
[61:37] short
[61:38] term.
[61:38] We'll see how it all
[61:39] works out. That's
[61:41] all for the week,
[61:42] guys. And lastly, I
[61:43] truly believe that
[61:44] we
[61:45] saw something very
[61:46] important this week, and based
[61:47] on my experience
[61:48] with
[61:49] hedge fund liquidations, these
[61:50] lows are
[61:51] very significant to how
[61:53] we
[61:53] trade
[61:54] over the next 6-7
[61:55] days. There will be
[61:57] fluctuations, but
[61:58] use the
[61:58] “
[61:59] trend confirmation day
[62:00] ” approach. See you
[62:01] next week.