Short-Term AI Fear, Long-Term Compute Scarcity

[0:00] in the heat of summer. A
[0:03] relatively quiet
[0:04] week,
[0:05] earnings reports continue to
[0:06] roll in and this will be a
[0:07] major theme for the
[0:08] rest of the year. Despite
[0:11] what you read or
[0:12] hear, profits
[0:13] will remain the main
[0:15] story. I'll focus
[0:17] on some specific
[0:19] things about AI and the
[0:20] issues I
[0:21] talked about earlier
[0:22] regarding profits
[0:23] and margins. China
[0:26] came to the market
[0:27] today, after
[0:28] I heard people
[0:29] say that China is not
[0:30] participating in the
[0:31] AI ​​race with the US.
[0:34] For those of you who
[0:35] listen to such statements, this is
[0:37] not true. And yes, I
[0:39] read the Deep Seek interview.
[0:41] I read an interview with Xi
[0:42] Jinping. Uh, they
[0:45] participate in
[0:46] the race for many
[0:48] good reasons. If
[0:51] you think that the
[0:52] dollar's status as a
[0:53] reserve currency was
[0:54] an advantage, then
[0:55] owning
[0:55] cryptocurrency is becoming
[0:57] perhaps the most important thing in the
[0:58] world in the future.
[1:02] So every country that
[1:03] can compete
[1:04] will do so, and that
[1:05] turns everything into a
[1:06] struggle between China and the
[1:08] US with two different
[1:09] strategic
[1:10] approaches. I'll go through
[1:12] some of them.
[1:13] Google reports. I'll
[1:16] get to the bottom of it
[1:17] and just
[1:18] lay out the facts, not
[1:19] just about the
[1:20] negative free
[1:21] cash flow
[1:22] that's been talked about on
[1:23] Twitter for the quarter, but also
[1:24] about the rapid
[1:25] growth in orders
[1:26] and the development of cloud
[1:27] technologies—
[1:28] short-term noise
[1:29] versus long-term
[1:30] trends. I'll try to
[1:35] give you some perspective
[1:36] on what to
[1:37] think about in the
[1:38] context of AI,
[1:39] whether
[1:40] this
[1:41] slowdown—that is,
[1:43] the consolidation after the
[1:44] parabolic move—continues for
[1:45] another six weeks or
[1:46] six months, it doesn't really
[1:48] matter. This does not
[1:50] change
[1:51] the long-term
[1:52] course of AI. A new request
[1:55] regarding Mosaic that I've
[1:56] been working on this week.
[1:59] Kimmy K3, I'll
[2:00] elaborate. I
[2:01] have published a new
[2:02] article about Vera Rubin
[2:03] , which is out now.
[2:05] Oil prices have
[2:06] risen again. So we
[2:07] will discuss this and the
[2:08] transparency law.
[2:11] But let's get started.
[2:13] So, in a week,
[2:14] despite all the events—
[2:16] big jumps up
[2:17] or down—the S&P index
[2:18] fell by 61 basis
[2:20] points. So, again
[2:21] , I like to
[2:22] look at the mass
[2:23] movement up, and we've now
[2:26] had, essentially, two
[2:28] weeks of growth,
[2:29] then two weeks of
[2:30] decline, two weeks of
[2:31] growth. We are now
[2:33] consolidating a significant
[2:34] upward movement. This is what the
[2:35] weekly
[2:37] changes look like. This may
[2:38] change, but for now
[2:39] I'm sticking to this. Q
[2:40] decreased by one and a half
[2:42] percent. IWM,
[2:44] fourth
[2:46] consecutive week with
[2:47] less than
[2:48] 1% fluctuation. So, it's been falling for
[2:50] four weeks in a row, by
[2:51] less than 1% each time.
[2:54] You won't see this kind of movement anywhere else. V
[2:57] is declining and we
[2:58] see
[3:00] small-cap companies
[3:01] outperforming the
[3:02] market, indicating
[3:03] strength in earnings and
[3:05] expansion. Impulse,
[3:09] technological
[3:10] impulse. After a
[3:12] terrible four
[3:14] weeks, we got an
[3:15] 8% rebound in a
[3:16] week. V per week
[3:19] was less. This is the
[3:22] movement. We went up and
[3:23] then
[3:24] back down again. This is
[3:25] a graph of
[3:25] technological
[3:26] momentum.
[3:27] Industrial momentum
[3:28] went straight back to
[3:29] the lows. The white line
[3:30] here is the momentum as a
[3:32] whole. After such a
[3:36] massive move with V,
[3:37] as I have shown, the chances of
[3:38] rapid growth are
[3:40] slim. Even if we
[3:43] go up, it will
[3:44] take some time.
[3:45] I think this will take the entire
[3:47] second half of the year.
[3:49] I believe we are in for a
[3:50] period of
[3:51] volatility, where companies
[3:52] that have exceeded
[3:53] expectations, like Intel,
[3:57] Micron, etc. They will find the
[3:59] bottom and be able to
[4:00] build something. And then there will be
[4:02] other companies
[4:03] that won't reach the
[4:04] bar. On the other hand,
[4:05] what helped
[4:06] this week's momentum
[4:07] in the tech sector was the
[4:09] sell-off in Google shares.
[4:11] This was the biggest
[4:12] part of the jump
[4:13] as
[4:14] hyperscalers were in the
[4:15] negative. If
[4:17] you look at the rest of
[4:18] the pulse, it's very
[4:19] quiet there, nothing is
[4:20] happening. This is
[4:21] truly the side of artificial
[4:23] intelligence. This is a
[4:25] balanced ETF, where,
[4:26] as I said, there is
[4:28] n't even any
[4:29] volatility visible. We're
[4:31] just hanging out
[4:32] . The thematic
[4:33] portfolio grew by one
[4:34] and a half percent.
[4:35] Again, a lot of
[4:37] volatility, a lot of
[4:38] stocks up, a lot of stocks
[4:39] down. I think we will continue to
[4:40] see this
[4:42] . As for
[4:43] hyperscalers, the
[4:44] thematic portfolio
[4:46] had its best
[4:47] week since late
[4:48] March. It was
[4:49] more than all the
[4:50] previous ones
[4:51] combined. This only
[4:52] underscores what a
[4:53] bad week it has been
[4:54] for hyperscalers.
[4:57] Thematic portfolio.
[4:58] I showed this
[4:59] last week. We've
[5:00] dropped a little
[5:02] below 60. Essentially, we're at
[5:03] about a 38%
[5:05] correction. Hmm, can
[5:07] it go any lower?
[5:08] Of course. I don't think that
[5:10] will happen at this point. I think we
[5:12] will spend some time
[5:13] fluctuating at this
[5:14] level. And if we go
[5:15] lower, it will be insignificant.
[5:16] Regarding technical
[5:17] analysis of indicators—
[5:18] again, this was a
[5:20] worrying signal that was
[5:21] starting to appear
[5:22] here. Well, we reached the
[5:25] highest level since
[5:27] July 10th. And in fact, with the
[5:30] exception of that
[5:31] single indicator for
[5:33] July 10, we see
[5:34] an improvement in the overall
[5:35] picture. This is worth
[5:36] monitoring,
[5:37] especially if
[5:38] we
[5:39] break above the 20-day moving
[5:40] average next week. This will be the
[5:41] first signal.
[5:43] Again, the 200-day
[5:43] moving average going
[5:44] up still
[5:45] looks great.
[5:46] Being above the 200-
[5:47] day is still
[5:48] good. We
[5:49] are still in a
[5:50] bull market.
[5:51] Last week I
[5:52] showed you
[5:53] the history of "head and
[5:54] shoulders" in
[5:54] structural "bull
[5:55] " markets; Some of the
[5:56] most powerful
[5:58] signals appear
[5:59] after a neckline break
[6:00] and a subsequent
[6:01] upward reversal. So far
[6:03] we have done it.
[6:04] Give it a little more time
[6:05] to see what happens
[6:07] next. There is still a lot of leverage in the system
[6:09] . Therefore, it is quite
[6:11] possible, even if
[6:12] only for one day, that
[6:13] we will drop closer
[6:14] to the 200-day
[6:15] average. As I
[6:16] said, I started
[6:17] buying last
[6:18] week. I did
[6:20] n't buy anything this week.
[6:21] But I bought
[6:22] last week and
[6:23] will continue to look for
[6:25] buying opportunities during
[6:26] consolidations because I
[6:27] believe
[6:28] AI pessimism is growing again
[6:30] . We go through
[6:33] periods like this, and when
[6:34] people become
[6:36] pessimistic again, I
[6:37] hear their
[6:38] arguments, like,
[6:39] "Anthropic is the end,"
[6:40] just because
[6:42] their growth rate is
[6:43] going to slow down, that's when
[6:44] I start to get
[6:46] more interested in it. As for
[6:47] profits, that will be the
[6:48] main topic. You won't
[6:50] hear me
[6:52] talk much about earnings,
[6:54] but 86% of companies, and we're
[6:55] almost a
[6:56] third of the way through,
[6:57] reported above
[6:58] expectations. They
[6:59] exceeded predictions
[7:01] . 3% were at the level
[7:02] of expectations, 11% were lower.
[7:04] So we have an
[7:05] 8 to 1 ratio of
[7:06] companies that
[7:07] beat forecasts
[7:08] to those that didn't. The annual
[7:09] average
[7:10] is 80%,
[7:11] the five-year average is 78%, and the
[7:12] ten-year average is 76%.
[7:14] Again,
[7:14] financial results
[7:15] are appearing across the
[7:16] market. You will hear
[7:17] this from me repeatedly.
[7:19] Profit margins. So,
[7:20] again, this
[7:21] includes Google, which
[7:22] had a one-time
[7:23] incident. Well, not exactly a
[7:24] one-off, because it
[7:25] also affected these
[7:26] numbers here. Without
[7:27] Google, we
[7:29] have 14.4. If
[7:30] you take
[7:32] Google out of here, we're
[7:33] down closer to
[7:34] 14. So we're still
[7:37] showing
[7:38] S&P margin growth. And
[7:41] this is the main news.
[7:43] During the week, I
[7:44] released a video for
[7:45] subscribers, and it
[7:46] was about China. Uh,
[7:49] China
[7:50] sent an important
[7:51] signal last night. Capital
[7:52] markets have become
[7:53] critical to
[7:54] achieving their
[7:55] AI ambitions. That's
[7:56] one of the reasons I
[7:58] want you to understand
[7:59] —this is a race. This is
[8:00] important to realize.
[8:03] Although the goals are different, and I
[8:04] agree that China is
[8:05] more focused
[8:07] on open source and the
[8:08] idea of ​​accessibility for
[8:09] all, while the US is
[8:11] still focused on
[8:12] capitalism, which I
[8:13] think will lead
[8:14] to problems with
[8:15] making money from AI, as
[8:16] many
[8:17] hope. I'm not
[8:19] pessimistic about
[8:20] Anthropic, but I believe
[8:22] their
[8:22] growth rate will never
[8:23] reach the levels they were
[8:24] talking about
[8:25] a month ago, and that's an
[8:26] important shift. So,
[8:27] China is providing
[8:28] massive state
[8:29] support to
[8:30] stop the sell-off in
[8:31] tech stocks. It's
[8:32] important for you
[8:33] to think about this
[8:34] because, as far as the AI ​​base is concerned
[8:36] , China has been very active
[8:37] in supporting it
[8:38] , because they will have their
[8:39] own problems if
[8:41] their stock market
[8:42] crashes. China
[8:44] is in the same
[8:45] situation as the US, where the
[8:46] stock market is becoming
[8:47] increasingly important. Not
[8:49] because it's a huge
[8:50] part of the investment, but
[8:51] because of the
[8:52] capital needs, so they
[8:53] need a stable
[8:54] market. They don't have a
[8:56] housing market because
[8:58] property prices in
[8:59] China continue to
[9:00] fall every month, which
[9:01] leads to
[9:02] instability unless
[9:03] another asset class grows
[9:04] , and stocks are
[9:06] a bit easier to support now.
[9:08] China's sovereign wealth funds
[9:09] are buying nearly 9
[9:10] billion shares to
[9:11] support the market. They
[9:14] are changing
[9:16] listing standards, as they
[9:17] did in the US. So,
[9:19] with regard to IPOs and going
[9:21] public, they are changing
[9:22] the rules to
[9:23] get
[9:24] people into these stocks quickly again. It's
[9:26] much easier
[9:27] to maintain them if
[9:28] they can
[9:29] get into the indexes faster.
[9:31] Chinese stocks
[9:31] continue to rise
[9:32] as state-owned
[9:33] insurers
[9:34] tout
[9:34] long-term
[9:35] investments. The latest
[9:36] sign of government
[9:36] support: five
[9:37] state-owned
[9:38] insurers have said
[9:38] they will continue to
[9:39] increase the share of
[9:40] equities in their
[9:41] portfolios. A lot of this has
[9:42] to do with a
[9:43] very important
[9:44] company in the AI ​​field, the
[9:45] most important AI
[9:46] company in China—
[9:47] Deepseek. Its
[9:48] founders' fund fell by 16
[9:50] %. So, remember,
[9:52] this is a hedge fund.
[9:53] It
[9:55] fell 16% last week. Deepseek and the AI
[9:58] boom in China are
[9:59] increasingly fueled by
[10:00] public money.
[10:01] This was in May. On
[10:03] June 16, China's
[10:05] DeepSeek raised over $7
[10:06] billion in
[10:07] funding in an
[10:08] unusual
[10:08] deal structure. And
[10:10] it is unusual because
[10:12] China's state structures,
[10:13] that is, the government, are the only
[10:15] external party with a
[10:16] vote.
[10:18] That's why you can't
[10:19] separate this race.
[10:21] China is deeply
[10:22] involved, and if the
[10:23] most important AI
[10:24] company in China
[10:25] collapses, and China doesn't
[10:30] support them, and they
[10:31] need to raise
[10:32] more capital,
[10:33] which I will tell, and not
[10:35] just them, but everyone.
[10:36] Chinese AI
[10:37] companies are rushing to
[10:38] raise funds to
[10:39] close
[10:39] the gap with the US.
[10:41] Chinese AI developers are
[10:41] competing to
[10:42] raise money
[10:43] through equity issuance
[10:44] or loans, believing
[10:44] they need a
[10:45] larger pool of
[10:46] resources to keep up with
[10:47] US competitors.
[10:48] Capital is
[10:49] crucial.
[10:50] China's DeepSeek
[10:51] intends to raise new
[10:52] capital. They just
[10:53] made an engagement.
[10:54] The government participated, 7.4
[10:56] billion, and that was only a
[10:57] month later. Moonshot,
[11:03] known for Kimi 3.0.
[11:06] 0 is aiming for a
[11:07] valuation of 50 billion
[11:09] during pre-IPO rounds.
[11:12] So they're going to
[11:13] raise money, and
[11:15] again, I
[11:16] talked about this
[11:17] last week: Anthropic is
[11:18] now valued at 23
[11:19] trillion, this is a valuation
[11:21] of 50 billion—
[11:22] no comparison
[11:23] between backing AI from a
[11:24] solid base, I'll talk about Kimi
[11:26] later. Um,
[11:30] all this capital
[11:31] needs, as I
[11:32] showed last
[11:33] week, and it will
[11:34] continue.
[11:36] Hyperscaler bonds
[11:37] are falling. This
[11:38] scares everyone. This is hitting
[11:39] global
[11:41] bond markets. This
[11:42] scares everyone. I'm not
[11:43] worried about this. Um,
[11:45] SpaceX is preparing to
[11:46] raise $116 billion
[11:48] . This is all just to
[11:49] show that
[11:50] again, when I was
[11:50] listing the reasons for the
[11:52] mid-cycle slowdown. One
[11:53] negative
[11:54] factor was the
[11:55] huge amount of
[11:56] capital raising
[11:57] that was to take place,
[11:58] not just what has happened
[11:59] so far. Google raised 80
[12:00] billion. All
[12:01] Chinese companies are
[12:02] raising money.
[12:03] You still have Open AI,
[12:04] Anthropic, SpaceX
[12:05] has already completed it. It should come as no
[12:07] surprise to anyone that
[12:08] the market is in trouble,
[12:09] as all this
[12:10] liquidity
[12:10] is being absorbed by
[12:11] securities issuance
[12:12] . It takes time
[12:13] to digest this,
[12:14] and it happens
[12:15] in the summer, when
[12:16] trading volume and liquidity are already
[12:17] low and
[12:18] most people are on
[12:19] vacation. Eric
[12:20] Schmidt says
[12:21] cash, not energy,
[12:23] is the real limit
[12:24] to AI growth. I've
[12:26] seen this chart
[12:28] many times before. Phil
[12:29] Rosen, my buddy, uh
[12:30] , again, I
[12:31] think this is
[12:32] going to go on for a long time. I
[12:34] don't think this is anything
[12:35] new. But it
[12:36] fits into the
[12:37] big picture.
[12:39] So, this is from Sundar
[12:40] after the results were declared
[12:41] . And I
[12:42] just want to
[12:43] make sure you
[12:44] can walk through and
[12:45] talk about
[12:45] free cash
[12:46] flow. You can
[12:47] talk about the fact
[12:48] that they won't
[12:49] get it. These are seven
[12:50] slides of simple
[12:51] honesty about what's
[12:53] going on. The first
[12:56] part—you can't
[12:58] ignore that their
[13:00] revenue, not profit, but
[13:01] revenue grew 24%
[13:03] overall,
[13:05] 17% in search, and 82% in
[13:07] cloud. And 13% in
[13:11] YouTube advertising. The reason
[13:14] I mention this
[13:15] when we look at
[13:16] this is because
[13:18] they are definitely
[13:19] growing by spending
[13:20] money. Their problem is
[13:23] not that the business is
[13:24] not growing. The problem is
[13:25] the lag, because
[13:26] they don't have
[13:27] the capacity. We
[13:28] continue to be
[13:29] limited in
[13:30] supply. Proof of
[13:31] rapid
[13:32] implementation. Demand still
[13:33] outpaces
[13:34] investment, even
[13:35] after significant
[13:36] capacity expansion.
[13:37] The company has already spent a lot
[13:39] , but there is
[13:39] still
[13:41] not enough capacity. They are
[13:42] expanding
[13:42] the capabilities of third-party
[13:43] companies to
[13:44] fill the gap.
[13:45] Again,
[13:46] think about why Meta is
[13:47] going through this.
[13:49] Meta can't
[13:50] use it. They
[13:51] don't have the cloud yet, so
[13:52] they're willing to sell
[13:53] it to others because others
[13:55] can't fill it.
[13:56] This is exactly what
[13:57] Meta should do. If they
[13:58] can't
[13:59] use
[13:59] the computing
[14:00] power for the
[14:00] next six
[14:01] months, why not
[14:02] use it? You
[14:02] get a great
[14:03] margin on this. They are
[14:04] willing to rent more
[14:05] expensive external
[14:06] computing
[14:07] power.
[14:08] Short-term
[14:08] expensive capacity deals
[14:11] can still be very
[14:11] attractive if
[14:12] done externally. So,
[14:14] they increase their
[14:15] capital expenditures.
[14:16] This is again good for
[14:17] all
[14:18] memory and chip manufacturers. They
[14:19] will continue
[14:20] to invest as long as
[14:21] they see
[14:22] attractive returns.
[14:23] On the direct question
[14:25] of return on
[14:26] investment (ROIC), Google
[14:27] has become more
[14:28] optimistic about
[14:29] opportunities for the year
[14:30] ahead, and with good
[14:31] reason. First of
[14:33] all, the cloud is growing by
[14:35] 82%. The backlog
[14:37] has increased by another 50
[14:38] billion. They
[14:40] expect to recognize
[14:41] a little more than
[14:42] half of that,
[14:43] say 260 billion,
[14:45] over the next 12
[14:46] months. So if
[14:49] they're going to
[14:50] spend 200
[14:51] billion and expect to
[14:52] receive 260 billion
[14:54] over the next
[14:55] two years, you can
[14:56] start to do the
[14:58] general math.
[15:00] Okay, they have a lot of
[15:01] revenue coming in,
[15:02] and very few people
[15:03] are talking about this
[15:04] lag as if
[15:06] it's not going to happen. I'll
[15:07] get to the risks
[15:08] later, so let's
[15:09] just go through
[15:10] it. So, free
[15:11] cash flow was
[15:12] negative;
[15:12] operating cash
[15:14] flow was still
[15:15] 39 billion, but
[15:15] capital expenditures
[15:16] were 45 billion. This
[15:17] was the first quarter
[15:18] when it was
[15:19] negative. They still
[15:20] have huge
[15:21] cash flow, but as they
[15:22] continue to
[15:23] spend, they
[15:24] rely on
[15:25] revenue to keep
[15:26] those expenses close
[15:27] to zero so as not to touch
[15:28] their cash
[15:28] reserves. Despite everything, this is
[15:30] not some crazy
[15:31] story. They have
[15:32] the ability to raise
[15:33] capital; they are not
[15:34] trying to achieve
[15:35] positive free
[15:36] cash flow in the
[15:37] near
[15:38] term. So
[15:41] here are the returns, and I want
[15:42] you to pay attention
[15:43] to the parabolic movement;
[15:45] this is not a
[15:45] short-term
[15:46] leap. That's why the
[15:47] annual growth rate
[15:49] is 82%. So,
[15:50] Bob Elliott, you
[15:51] have to think about
[15:52] how much they
[15:52] paid to
[15:53] achieve this. As of the first
[15:54] quarter of 2024, Google
[15:55] has generated $30 billion
[15:56] in excess
[15:57] revenue in its
[15:58] cloud division,
[15:59] but spent $140
[16:00] billion in
[16:00] capital expenditures
[16:01] to get there.
[16:03] OK. So I take this,
[16:06] enter it into Grok, and
[16:07] actually ask
[16:08] for Bob
[16:09] Elliott's argument for a direct
[16:10] comparison. Since
[16:13] capital expenditures
[16:14] are investments in
[16:14] the future, shouldn't they
[16:15] take into account the
[16:16] growth of the
[16:17] order book, because that's what
[16:18] they're
[16:19] designed for? And, essentially
[16:20] , the system processes
[16:22] it, and you can't
[16:23] ignore it. The rapid
[16:24] growth of
[16:25] the order book
[16:26] supports the thesis
[16:27] that profits
[16:28] will materialize
[16:29] within the 5-7 year
[16:30] amortization window
[16:31] if conversion rates
[16:32] are maintained. So,
[16:34] again, this is
[16:36] just a case of
[16:37] investors having
[16:38] difficulty. This is an
[16:39] order book that grew
[16:40] from 106 billion to 514
[16:42] billion in one year
[16:43] . I mean, it's
[16:44] incredible.
[16:45] Growth from 106
[16:46] billion to 514 billion in
[16:47] one year is
[16:48] unprecedented. This is
[16:49] hard to believe.
[16:51] Elliott-style fears
[16:51] reflect the
[16:52] difficulty of valuing the
[16:53] unprecedented
[16:53] acceleration of
[16:54] AI compared to
[16:55] traditional
[16:57] ROI models. However, the
[16:58] order book in
[16:59] this regard suggests
[17:00] that investors
[17:00] often lag behind
[17:01] exponential
[17:02] curves, and the steady
[17:03] execution of tasks
[17:04] dispels doubts.
[17:05] Now, to give
[17:08] everyone the benefit of the doubt,
[17:09] let's just
[17:10] jump into Fable 5 and
[17:11] let the
[17:11] smartest language
[17:13] model on the market
[17:14] analyze
[17:15] the situation and explain to
[17:16] me what's going on
[17:17] . And I wanted her to
[17:20] break everything down in detail
[17:21] and provide
[17:22] a probability that
[17:23] it would end
[17:24] successfully. First of all,
[17:25] she says that the
[17:26] order book
[17:27] is the most important thing. I didn't
[17:28] ask her to say that.
[17:29] She just did
[17:30] it herself. They are
[17:31] not spending money
[17:32] on faith.
[17:33] Incremental
[17:33] costs are driven by
[17:34] demand. Which they
[17:35] already see. And
[17:36] the economics of the unit
[17:37] improve as it
[17:38] scales, not
[17:39] worsen. They
[17:40] literally rent
[17:41] Nvidia chips from SpaceX
[17:42] for about 920
[17:43] million a month.
[17:44] That's why if you
[17:45] just type in the query,
[17:47] give me the details, and
[17:48] all I asked for is the
[17:49] Google earnings report
[17:51] . Let's look at
[17:52] the details. Let's
[17:53] look at the
[17:54] order book. Let's
[17:55] look at cash
[17:56] flow. How should we
[17:57] interpret this? And
[17:58] finally, give
[17:59] me, uh, an estimate of the
[18:00] probability that
[18:02] this will work. One
[18:03] honest comment for
[18:04] the corner. Reported
[18:06] earnings per share of
[18:07] $911 included
[18:07] about $99 billion in
[18:10] equity gains.
[18:10] Operating earnings
[18:11] per share were closer
[18:12] to 285, slightly below the
[18:13] consensus estimate. The
[18:15] great thing about this is that
[18:16] it just gives
[18:16] you honest information.
[18:18] And here's what was
[18:18] said there. Should
[18:20] they continue to
[18:20] spend? Yes. I would
[18:22] give about a 70–75%
[18:23] probability that
[18:25] this
[18:25] capital expenditure program
[18:27] will prove valuable or
[18:28] creative over a
[18:30] 3–5 year horizon, that's the
[18:31] logic. Now it
[18:33] also gives you the cons, why it's
[18:35] only 70–75%. So this
[18:37] raises a question, and that's why
[18:38] , guys, don't
[18:39] make everything
[18:40] absolute. It's all about
[18:43] thinking in terms of stakes.
[18:45] The first is
[18:47] chip obsolescence, the second is
[18:48] off-balance sheet
[18:49] liabilities. So
[18:50] they're looking at
[18:51] obsolescence, the
[18:52] fact that they won't
[18:53] last
[18:54] long enough and their
[18:55] lifespan is actually
[18:56] four years, not
[18:57] six. Most
[18:58] people say
[18:59] they are longer than
[19:00] six years. But
[19:01] let's just
[19:02] assume that this is the
[19:03] risk that has been
[19:04] put forward. Regarding
[19:05] off-balance sheet
[19:06] liabilities, if you
[19:07] read this week
[19:08] about 1 trillion
[19:09] off-balance sheet
[19:09] liabilities for all
[19:10] these companies
[19:12] mentioned here. Third,
[19:13] order portfolio concentration and
[19:15] conversion risk. If
[19:16] you consider that this
[19:17] 514 billion order book is
[19:18] mostly OpenAI and
[19:19] Anthropic, then yes, they have
[19:20] concentration risks.
[19:21] If OpenAI goes bankrupt,
[19:22] they will lose this.
[19:23] So, these risks are associated with that
[19:25] . This was
[19:26] put forward. The verdict
[19:27] changes not because of
[19:28] any single
[19:28] free cash
[19:29] flow report, but because of whether
[19:31] contracted
[19:31] demand continues. The thing is, it
[19:32] all depends on the
[19:33] demand side,
[19:34] guys. This is what
[19:35] is considered through
[19:36] Jevons' paradox. This is something
[19:37] I will spend
[19:39] some time on. Now I
[19:40] want to make sure that
[19:41] this visual series is there
[19:42] because I think you
[19:43] need to
[19:44] focus on the
[19:45] long-term
[19:45] flow. So when I
[19:47] spend weeks
[19:48] saying that my feelings are
[19:49] finally starting to
[19:50] conflict with my
[19:51] memory and they say it's
[19:53] unstoppable. When I
[19:54] finally say, okay,
[19:55] Korea is collapsing and
[19:57] we see the
[19:57] market collapsing. the
[19:58] only two companies
[19:59] growing when
[20:00] it came to
[20:01] all
[20:01] the engineering,
[20:02] all the construction,
[20:03] all that stuff from
[20:04] the middle of the day. This is
[20:06] short-term noise
[20:08] . Markets never
[20:09] only go up.
[20:11] Even if you
[20:11] are in a
[20:12] long-term
[20:13] bull market, which
[20:14] we are for AI, there is no
[20:15] way to
[20:16] say that this is not a
[20:17] long-term
[20:18] bull market. The only thing that
[20:19] can be said is that
[20:20] he is running
[20:21] too far ahead.
[20:22] We need a big
[20:23] correction. I don't think
[20:24] we will have a big
[20:25] correction. And that's why,
[20:26] if we do that, I'm
[20:28] going to start seeing it on
[20:29] some of the signals
[20:29] inside. I've seen it
[20:31] from one side, and we
[20:32] really do have a
[20:33] lot of emissions going on. So,
[20:34] let me
[20:35] give you an example. Here's the S
[20:36] &P over the last six,
[20:38] five and a half
[20:39] years. We have been
[20:41] profitable since
[20:42] 2019. This includes COVID—
[20:44] 15% of the annual. It seems to be a
[20:46] mild bull
[20:47] market if
[20:47] you look at the chart.
[20:48] Here is the 63-day moving
[20:50] average. That's 22 trading
[20:51] days per month. This is the
[20:52] rolling return
[20:53] for three months
[20:54] during this
[20:55] period. Look at
[20:56] how many times they don't
[20:57] just become
[20:58] negative, but
[20:58] fall into double
[20:59] digits. One, two, three,
[21:02] four, five, six.
[21:05] Every year we have a
[21:06] three-month figure of
[21:08] -10%. Even during a
[21:11] long-term "
[21:12] bull" market with 15%
[21:14] annual interest. Now I think
[21:16] it will be much
[21:17] higher than 15%. The reason
[21:19] I created this is because
[21:21] for the next
[21:23] 3-5 years I expect it to
[21:24] outperform the S&P by
[21:25] at least 15% per year.
[21:27] So, if that's the case, here's the
[21:29] 63-day moving
[21:30] average for the
[21:31] thematic portfolio
[21:32] . Can we fall
[21:34] another 20% depending on the S&P?
[21:36] Yes. I don't think that's
[21:38] going to happen here, because I
[21:39] think a lot of
[21:40] that first
[21:41] wave was something
[21:42] else. Do I think that
[21:43] once we take
[21:44] another step up, it
[21:45] could happen? Yes. Do
[21:46] I think we will go into the
[21:47] red during
[21:48] this period? Yes,
[21:49] I think so. If I
[21:50] go back and
[21:51] look at the "Magnificent
[21:52] Seven," you'll see the same thing
[21:53] . Three podcasts
[21:54] for you to
[21:55] listen to. I'll put
[21:56] them in the subscribers section
[21:58] as a summary so
[21:59] you don't have to
[22:00] search, but there are
[22:01] hints to them there. But I
[22:02] listened to these three: All-In
[22:04] and Moonshots, which I
[22:05] thought were
[22:06] fantastic. I
[22:07] like Emad
[22:09] Mostak
[22:10] not only for his perspective
[22:11] on technology,
[22:12] as the founder of an AI
[22:13] company that he sold
[22:14] , but also from a
[22:15] hedge fund perspective on
[22:16] economics and
[22:17] macroeconomic
[22:18] vision. And also
[22:21] Stephen Ho, who led
[22:23] Forward Guidance—all three of those
[22:24] were interesting. Now,
[22:27] if any of you
[22:28] attend multi-day
[22:29] conferences where you
[22:30] listen to several
[22:31] companies and then
[22:32] several
[22:33] macro strategists and
[22:34] they all talk about the same
[22:35] topic. I
[22:39] wanted to put something
[22:40] together, which is what I did
[22:41] here first. I
[22:44] loaded all these
[22:45] transcripts into one
[22:46] context window and the
[22:47] first thing I
[22:48] asked was: review
[22:49] them all, highlight
[22:50] common themes from these
[22:51] three conversations, and name the
[22:52] most important moments
[22:54] . And here's what I got:
[22:56] we're getting closer to a
[22:57] glut of models, which
[22:58] means they're
[22:59] getting better and
[23:00] better. They
[23:02] are approaching very
[23:04] similar capabilities, and
[23:05] their price is
[23:06] falling rapidly. But we still
[23:08] have a shortage of
[23:08] computing
[23:09] power. So,
[23:10] all these different things are
[23:12] quotes. Intelligence
[23:13] can be copied,
[23:14] strength can be
[23:14] copied.
[23:15] Computing
[23:16] power—no.
[23:17] Jevons paradox.
[23:18] Cheaper intelligence
[23:19] means more
[23:20] intelligence consumption
[23:21] . AI implementation
[23:22] moves from
[23:23] experiments to
[23:23] workflows
[23:24] only when it is
[23:25] cheap, private, and
[23:26] managed. This is a shift in the
[23:27] corporate
[23:27] sector from
[23:28] maximizing tokens
[23:29] to their
[23:30] efficiency. And the
[23:32] inevitable winner
[23:33] in this will be
[23:34] routing, which
[23:35] means including
[23:37] Anthropic, but not only it.
[23:39] We will move towards
[23:40] routing to the
[23:41] best option.
[23:42] Here's what you get when you
[23:43] take three of them.
[23:45] They are all wonderful. It's like a
[23:47] brainstorming session with
[23:48] , say,
[23:49] nine people. And you
[23:51] get this
[23:52] cross-eyed look.
[23:54] Now, what I did
[23:55] next—using
[23:56] my thought process.
[24:01] I wanted to create a
[24:02] prompt using the mosaic
[24:03] method. This is essentially
[24:06] creating a mix of
[24:07] experts for
[24:08] forecasting. I'll
[24:09] post this on the
[24:11] subscribers page. There
[24:12] will be a prompt that
[24:13] will show you what it is
[24:15] . The human brain is a
[24:16] sequential processor
[24:17] . You listen to three
[24:19] podcasts and
[24:19] take them in
[24:22] turn. By the third
[24:23] podcast, the first one has already
[24:24] blurred to
[24:25] a few bright
[24:26] moments, and the connections
[24:27] between conversations and
[24:27] precise details
[24:28] disappear. This brings
[24:29] us back to the issue of
[24:30] the conference.
[24:33] The most powerful thing is to
[24:34] have a prompt that
[24:35] goes through everything and
[24:36] gives you a result
[24:37] based on what you heard in the
[24:39] form of a summary, where
[24:42] you
[24:43] reconnect all these little nuances that keep coming up.
[24:48] So think of it
[24:49] again as
[24:50] using AI as a
[24:51] conductor of a
[24:52] symphony, where each of
[24:53] these podcasts is a
[24:54] musician in its own
[24:55] place. Unlike the
[24:58] first time, when I
[24:59] simply asked what the
[25:00] most important points were
[25:01] coming out of the conversations.
[25:02] This is a prompt that
[25:03] literally goes
[25:04] through everything, catches
[25:06] nuances, even if
[25:07] it's a minor moment,
[25:08] and then puts them together into a
[25:09] mosaic. All I
[25:11] like to do when
[25:12] working with this data is
[25:13] look at the
[25:14] mosaic. If I
[25:15] believe a
[25:17] bear market is coming, then there
[25:18] must be a recession.
[25:19] If there is going to be a recession
[25:21] , there are certain things I
[25:22] want to see, like
[25:23] widening credit
[25:24] spreads, declining
[25:26] profits, declining
[25:27] profit margins, and so on.
[25:28] Not when they are
[25:29] all pointing upwards,
[25:31] because there will be no
[25:32] mass layoffs
[25:33] if profit margins
[25:34] are increasing. These are simple
[25:35] things. So, if you're
[25:38] using this,
[25:40] one thing I would
[25:41] advise, and I'm only saying this
[25:42] because of the time of year,
[25:43] sorry, I'm having a hard time
[25:45] with this, uh. Okay, that's it
[25:56] . It took a
[25:58] while. Um, it's
[25:59] fantasy football time.
[26:00] Using this
[26:01] mix of expert
[26:02] approach, mosaic
[26:03] approach,
[26:04] fantasy draft preparation has the
[26:05] same structure.
[26:06] This method was
[26:06] created for an ocean of
[26:07] expert conversations, a
[26:08] visible market
[26:09] consensus to
[26:10] beat, and a
[26:11] scoreboard to decide
[26:12] who was right. Thinks in
[26:13] mosaics. You will
[26:14] have a bigger picture.
[26:16] This is one way to
[26:17] think about things in a
[26:18] mosaic
[26:18] perspective. So I
[26:21] wanted to give you a
[26:22] visualization of that
[26:23] first context
[26:24] window, that we have a
[26:25] surplus of models, the
[26:26] Jevons paradox is
[26:27] accelerating
[26:29] because we have
[26:30] enterprise implementations, we're
[26:31] figuring out the
[26:33] routing structure. All of this
[26:36] means that demand
[26:37] continues to grow
[26:38] exponentially.
[26:40] The problem is that
[26:41] physical
[26:42] infrastructure
[26:43] cannot scale
[26:44] at the speed of
[26:45] software. We will
[26:47] forever run out of
[26:48] computing
[26:49] power. I don't
[26:52] know when we'll get
[26:53] to that point, because as
[26:54] I
[26:55] go through the
[26:56] rest of this, you
[26:57] have to trust
[26:58] me on one thing. If you
[26:59] hear that there will be
[27:00] excess
[27:00] capacity for a
[27:01] certain period of time, we
[27:02] may have
[27:03] excess
[27:03] capacity. If OpenAI
[27:05] isn't generating revenue and the
[27:05] market just says
[27:07] we're not going to give you any more
[27:08] capital, and nobody
[27:09] wants to go through that
[27:09] , these could
[27:10] be scary times.
[27:11] Again, I don't
[27:12] think that's
[27:13] going to happen because
[27:14] they're growing their
[27:15] revenues and they have a
[27:16] huge
[27:17] cash situation. But when you
[27:18] go through that,
[27:19] we're at a
[27:20] point where demand is
[27:21] outpacing
[27:22] efficiency gains. So I
[27:23] want you to think
[27:24] about the cycle of
[27:25] commoditization of
[27:26] intelligence. Everything is getting
[27:27] cheaper.
[27:29] Jevons' paradox stimulates
[27:30] higher demand. Demand is
[27:31] outpacing
[27:33] efficiency gains. The
[27:33] physical wall
[27:34] is approaching. And that was
[27:36] all before the video appeared. I
[27:38] just want you to
[27:40] always remember
[27:41] how much more
[27:42] video the world sees,
[27:43] humanoids, full
[27:44] self-driving—we're
[27:45] still in the early
[27:47] stages. And since I'm
[27:50] going through the
[27:51] agent side, we're
[27:52] also nowhere on the
[27:52] agent side in terms of
[27:54] implementation. So
[27:56] we have Kimmy K3 and QE for the
[27:58] mind. I wrote this
[28:00] this week; If you
[28:02] haven't had time to read it
[28:03] because of the summer holidays,
[28:04] be sure to
[28:05] read it, especially the
[28:06] part about "
[28:07] quantitative
[28:08] easing for the
[28:08] mind", which I will get to
[28:10] soon. This is
[28:11] very important. This is a
[28:12] key point. Kimi k3
[28:13] requires enormous
[28:15] computing
[28:16] power. This is not a
[28:17] small model. She
[28:18] needs a lot of things.
[28:19] I can't
[28:20] use it.
[28:21] No one at home will be able
[28:22] to use it
[28:23] unless they have powerful
[28:24] equipment. So it doesn't
[28:25] destroy the business. I
[28:26] use
[28:28] Anthropic, OpenAI, Perplexity, Grok every day. To
[28:31] each of them. Gemini—
[28:33] every day. But never
[28:35] . Kimi k3. Kimi k 2.5. Yes,
[28:38] because it
[28:38] fits my
[28:39] equipment. You
[28:41] still need
[28:41] computing
[28:42] power. Kimi k3 could
[28:43] be a major
[28:44] turning point for
[28:45] AI. Potentially
[28:46] negative for Openthropic,
[28:47] but overall
[28:48] positive for almost
[28:49] every other
[28:50] company in the world. And I
[28:51] mean that literally.
[28:52] Yes, this is the same
[28:53] part about "
[28:54] quantitative
[28:54] easing for the
[28:55] mind" that I
[28:56] wrote about, it got
[28:57] a lot more
[28:58] attention than we
[28:58] expected. Oh, in the
[28:59] last 48 hours, demand
[29:00] has approached the limits of
[29:01] our current
[29:02] capacity. We are
[29:03] ramping up capacity
[29:04] as quickly as we can
[29:05] to reopen
[29:06] new subscriptions. A
[29:07] company cannot
[29:08] function if it
[29:09] runs out of
[29:10] computing
[29:11] resources. Sorry, that's
[29:12] not how it works.
[29:14] Chinese project
[29:15] moonshot suspends
[29:16] Kimi subscriptions amid
[29:17] overwhelming demand. The IPO is
[29:19] near. Aaron
[29:21] Levy, who I have
[29:22] often referenced here,
[29:23] writes a lot, but I
[29:24] think it is important.
[29:25] He cited
[29:26] Daniel Cornum, who
[29:28] says that the biggest
[29:29] winners in AI
[29:30] have the lowest margins.
[29:34] Low-
[29:35] margin businesses could be
[29:36] one of the biggest
[29:37] winners in AI
[29:38] . And all because even
[29:40] with a reduction in costs of
[29:41] less than 1% of
[29:42] revenue, profits can
[29:43] increase by more than 25%
[29:45] , because existing margins are
[29:46] too small. The main
[29:47] problem is
[29:47] implementation among
[29:48] employees, not
[29:49] training them
[29:50] to use the
[29:50] tool. It is worth
[29:52] implementing
[29:52] agents. The agents are already
[29:54] here, and this is the stage
[29:55] we
[29:55] are at now. The
[29:56] main conclusion is
[29:57] that AI should be
[29:58] deployed as
[29:58] infrastructure, not
[29:59] just as
[30:00] software. I want
[30:01] you to think about
[30:02] this, because it's already happening.
[30:04] This means that we have yet to
[30:05] see a
[30:06] profitability advantage for
[30:07] most companies, and
[30:08] some of the
[30:10] world's largest companies by
[30:11] revenue do not have this
[30:14] advantage. So
[30:15] last November,
[30:15] I wrote an article
[30:17] called "Quantitative
[30:18] Easing for the
[30:18] Mind," and I talked about
[30:19] how artificial
[30:20] intelligence is infusing
[30:21] the economy with
[30:22] intellectual
[30:22] liquidity. This is exactly what
[30:24] I was talking about.
[30:25] Under conditions of quantitative
[30:25] easing and
[30:26] zero interest
[30:26] rates, companies
[30:27] use
[30:27] cheap money to
[30:28] manipulate
[30:28] earnings. The script
[30:29] was simple. You
[30:30] take out a loan at 2%,
[30:31] buy back stocks with a
[30:32] yield of 6%, and
[30:32] watch your
[30:33] earnings per
[30:34] share increase even though your
[30:34] overall profit
[30:35] doesn't change. It was
[30:36] financial leverage.
[30:38] But now even a
[30:39] simple
[30:40] 5% cost reduction,
[30:41] going back to what
[30:42] I just showed,
[30:43] gives a
[30:45] 20% jump in earnings per share. You
[30:47] get a much
[30:48] greater impact from
[30:49] cost reductions
[30:50] thanks to AI and how
[30:51] it impacts the
[30:52] bottom
[30:53] line
[30:54] . So I just went
[30:57] to Fable and said, give
[30:58] me five companies
[30:59] from the S&P 500 that
[31:00] fit
[31:01] the criteria for this
[31:02] study. This is an
[31:03] article by Daniel
[31:05] Cornum that
[31:06] mentions UPS, Cisco, Jabil,
[31:08] Centene, and Walmart. You can
[31:10] read whatever you want, but
[31:11] you see that these
[31:12] companies have
[31:13] huge revenues,
[31:14] but also many expenses
[31:16] and a large
[31:17] staff. They have
[31:18] a lot of "friction" and
[31:19] a lot of internal
[31:20] problems. So if you
[31:21] start
[31:22] using
[31:23] agents for
[31:24] workflows, it will make a
[31:25] big difference. But
[31:26] here is the profitability of
[31:28] profit, measured in
[31:29] one way that
[31:30] uses
[31:31] GDP statistics. These are
[31:33] government...these are
[31:34] corporate
[31:35] profits from
[31:35] GDP figures
[31:36] relative to
[31:37] nominal GDP.
[31:38] So, there's another way
[31:40] to look at
[31:41] profit margins, and we're at an
[31:41] all-time high
[31:42] . And again,
[31:43] the reason I
[31:44] brought this up
[31:45] about recessions:
[31:46] profit margins weaken
[31:47] before they start.
[31:48] So, during the
[31:49] dot-com bubble,
[31:50] we already saw this
[31:51] happening.
[31:52] Profits have already started
[31:53] to fall. So could this
[31:55] happen? Again,
[31:56] maybe, but we're not
[31:57] even close to that right now
[31:58] . And
[31:59] you see the time lag
[32:00] in this. So we will
[32:01] monitor
[32:02] profitability. This
[32:03] will be a good
[32:03] indicator.
[32:04] Profits
[32:05] just keep
[32:06] growing year after year. And again
[32:08] , these levels are
[32:09] typical mostly of
[32:10] times when we were
[32:11] emerging from recession,
[32:13] with the exception of the Internet
[32:14] boom. In everything else,
[32:16] you're coming out of a recession
[32:17] , and you have this boom. It
[32:18] was the
[32:18] housing boom and the Chinese
[32:19] factor. But
[32:20] look at what
[32:20] happens before
[32:21] you enter a
[32:22] recession. The same.
[32:23] Not only is
[32:24] profitability falling, but
[32:25] profits themselves are falling as well. Well, now
[32:26] profits are
[32:26] accelerating. That's
[32:27] why this is a critically
[32:28] important part for
[32:29] you. If you
[32:30] are going to bet on
[32:31] the decline, you are making a
[32:31] bet with a low
[32:32] probability of success.
[32:33] Is this possible? Yes. Is
[32:35] this likely? No. And
[32:37] if you're betting against
[32:38] bonds, because
[32:40] hyperscaler bond spreads are
[32:41] widening—they have the
[32:43] best balance sheets in
[32:44] the world, bar none. And
[32:47] you just saw the
[32:48] Google numbers. Don't look at these
[32:50] things in that way. And
[32:51] don't let this
[32:52] hype
[32:54] surrounding the perpetual
[32:54] bearish sentiment
[32:55] drive you into this
[32:56] position. We
[32:57] are in a state of
[32:58] slowdown. Everyone
[32:59] became too
[33:00] optimistic.
[33:01] Follow the technical
[33:02] charts. Keep an eye on the
[33:03] weekly data.
[33:04] There's no reason to chase it
[33:05] now. We
[33:07] are currently in what
[33:07] is called a "mini-
[33:08] bear" market.
[33:09] We have 30% of stocks below the 20-
[33:11] day moving
[33:12] average.
[33:12] Let's see how
[33:13] it gradually starts to
[33:14] grow. Just
[33:15] another look at where
[33:16] we were during
[33:17] last
[33:18] week's reports. And again,
[33:19] I won't
[33:20] read it out, but it's just to
[33:21] emphasize that the
[33:22] profits were
[33:23] explosive.
[33:24] Grade review factor. So
[33:26] , to show you
[33:28] how important
[33:29] earnings are even on a
[33:30] different basis:
[33:31] looking at valuations
[33:32] will show you weakness
[33:33] in advance, i.e.
[33:34] when stocks are falling.
[33:37] So I believe the
[33:38] stock market
[33:39] is taking into account, if we
[33:40] see the
[33:41] three-month
[33:42] valuation revision factor
[33:44] coming down, meaning
[33:45] they're falling, then it's
[33:46] fine, even with the S&P,
[33:47] we've seen momentum, we have
[33:49] n't seen change. So,
[33:52] this is imposed on
[33:53] earnings per share.
[33:55] So forward
[33:56] profits are much higher
[33:57] than they were before. This
[33:58] shows you the status for
[34:00] this year. This is a
[34:01] quarterly figure of
[34:02] 305. We
[34:04] have increased this
[34:04] figure significantly. So,
[34:05] the revision of ratings will be
[34:07] something to
[34:08] watch as a
[34:09] factor, and that's where
[34:10] we are,
[34:11] starting in 2010.
[34:12] This is almost a direct
[34:13] overlay. We've
[34:15] been in this
[34:16] part since the
[34:17] iPhone came out that
[34:18] viewing ratings has been the
[34:19] dominant factor
[34:20] for tracking. And
[34:22] now they just
[34:23] say that the stock
[34:25] market is fine. Hmm,
[34:28] this is another version of
[34:29] exceeding
[34:30] consensus forecasts.
[34:31] And the surprise
[34:33] is again 93% as of
[34:34] Wednesday. The numbers I
[34:36] got from FactSet for
[34:37] Friday were
[34:38] lower. Um, again
[34:39] , we don't have
[34:40] enough
[34:40] computing
[34:41] power. So if
[34:42] you want to find a
[34:43] place, again,
[34:44] hyperscalers are
[34:45] spending money.
[34:46] People don't like
[34:46] free cash
[34:47] flow. I don't
[34:48] blame them. There is a 25%
[34:49] chance that it
[34:50] won't work for them. They
[34:52] need to see how
[34:53] their
[34:54] multipliers for
[34:54] memory and everything else are compressed
[34:55] . These guys
[34:56] will get the money. It's
[34:57] just a matter of
[34:58] when we actually
[34:59] reach peak
[35:00] profits and when the
[35:02] capacity will come on stream. I have yet to
[35:03] see any evidence that
[35:04] anything can
[35:05] keep up with demand.
[35:06] So let's
[35:07] take a look at what's
[35:08] happening to
[35:08] see if there are any signs
[35:09] that
[35:10] computing
[35:10] power is becoming
[35:11] more affordable.
[35:12] GPU availability
[35:13] is shrinking again. So we're
[35:15] back to where we
[35:16] were after a
[35:17] brief period
[35:18] when there were few
[35:18] computing
[35:19] resources. Mind you
[35:22] , I don't know if
[35:24] anyone has proven this as a
[35:25] fact, but Nvidia is
[35:26] reportedly able to
[35:28] produce a thousand
[35:29] Vera Rubin racks per day,
[35:30] which is 630
[35:31] billion per quarter.
[35:35] I mention this because
[35:36] Vera Rubin provides
[35:37] the highest
[35:38] performance per
[35:39] watt and lowest
[35:40] token cost in
[35:41] the world. We are
[35:43] at the Vera Rubin stage where the
[35:45] next generation is accelerating.
[35:46] We
[35:47] are definitely in
[35:48] full production
[35:49] cycle. There were many
[35:51] rumors that they would
[35:52] n't do it. So, I
[35:53] finally released a
[35:54] report on Vera Rubin,
[35:56] which I spent a total of,
[35:58] I would say, three or
[35:59] four months
[36:00] really digging into
[36:02] , trying to
[36:03] find companies that were
[36:04] n't in the
[36:05] 100-
[36:06] name portfolio. I think there are
[36:08] 25 that are not
[36:09] related to this. But
[36:11] they will still
[36:13] correlate. So I don't
[36:14] expect you to go into
[36:15] depth on this,
[36:16] but it will give you,
[36:17] especially hedge funds
[36:19] and investment
[36:20] funds, a few
[36:21] additional names to
[36:22] explore within the 800 VDC.
[36:24] To do this, I spent
[36:25] time looking for those who would
[36:26] have significant
[36:27] growth. This is the
[36:28] most important question
[36:29] for the rest of this year.
[36:31] This is the reason why Marvell is
[36:32] still
[36:33] my biggest
[36:34] position. I believe
[36:35] we will see a leap in the
[36:36] field of optics and
[36:37] photonics. I believe
[36:39] we have yet to see how
[36:40] the profits from this
[36:41] have impacted the
[36:41] performance. I expect
[36:42] results to
[36:43] be higher than
[36:44] projected, and we are now at the
[36:45] point where they
[36:46] will be forced to
[36:47] revise
[36:48] their numbers significantly
[36:49] upwards. Whether
[36:49] this will happen with Marvell remains
[36:51] to be seen. But at
[36:52] this point, I also
[36:53] purchased another
[36:54] optical asset this
[36:55] week—AOI. And again
[36:58] , for various reasons,
[36:59] it's part of the portfolio,
[37:00] but it's really my
[37:01] scanning for what
[37:03] technically
[37:04] looks like a big
[37:04] enough
[37:05] correction, and paying attention to the
[37:06] data. I would recommend
[37:08] taking the time to
[37:08] read
[37:09] and familiarize yourself with this report
[37:11] . For those of you who have
[37:12] been waiting, I have posted
[37:15] parts three and four of the
[37:16] Research
[37:17] Skills Architecture. This is the way
[37:19] I navigate the path
[37:20] and do the work for the
[37:21] Vera Rubin Report. Much
[37:23] of the idea for the
[37:24] mosaic model
[37:25] came from the "
[37:26] expert mix" approach
[37:27] that I use in the
[37:28] deep
[37:29] research that you've
[37:30] already seen. These
[37:31] tips will also
[37:32] be available this
[37:33] weekend. Um, I
[37:35] just wanted to
[37:36] mention Steve Ho again because I
[37:37] think it's important
[37:38] for those who really
[37:40] want to listen to something
[37:41] relevant. If you
[37:42] looked at the
[37:43] token index, it is currently
[37:44] run by
[37:46] Silicon Data, which
[37:47] maintains this index. He's
[37:48] sorting it out. I already
[37:49] said that this is a
[37:50] silly index
[37:51] to become a bear,
[37:52] but it really
[37:53] goes through everything.
[37:54] AI computing is becoming a
[37:55] financial market.
[37:56] Demand for computing
[37:57] should fragment over time
[37:57] .
[37:58] Enterprise AI will
[37:59] likely
[38:00] focus around
[38:00] orchestration and
[38:01] routing. We've already
[38:02] talked about this. The
[38:03] token index does not
[38:04] show a collapse in
[38:05] demand. The market
[38:05] is moving from
[38:06] maximizing tokens
[38:07] to their
[38:08] efficiency. I already
[38:09] talked about this. Um,
[38:10] cheaper models
[38:11] put pressure on advanced
[38:12] developments, but they
[38:13] expand the entire
[38:14] market in general. So
[38:15] you can see
[38:16] that when you listen to
[38:17] Gavin Baker, the
[38:19] "All-In" podcast, or this -- it's the same
[38:20] story
[38:21] . So don't sit back and
[38:22] say that cheaper
[38:24] models will destroy Anthropic.
[38:25] This is not true. This
[38:26] will expand the overall
[38:27] market. The pie
[38:28] will grow faster, and the
[38:30] orchestration level
[38:31] will be the smartest
[38:32] intellectual
[38:33] component. So if
[38:35] businesses start
[38:36] doing more,
[38:38] everyone will simply get a
[38:39] bigger piece of the
[38:40] pie. It's just that Anthropic
[38:41] won't grow as
[38:43] fast as it used to.
[38:44] Competition has emerged. Um, and I
[38:46] discussed this with
[38:47] Pomp. And how by
[38:50] 2030 I'm really
[38:51] worried about the fate of all
[38:53] public companies,
[38:54] including
[38:55] infrastructure ones. So
[38:58] it's not forever.
[38:59] In fact, I think
[39:00] I'm one of the most
[39:01] pessimistic
[39:02] about
[39:03] 2030, but that's still a long way
[39:04] off. And that could
[39:05] change if AGI does
[39:06] n't come along so quickly,
[39:07] and if we get to the
[39:08] point where governments
[39:09] simply
[39:09] keep
[39:10] public
[39:11] markets running. Remember,
[39:12] demand remains
[39:13] structurally strong. The
[39:14] next phase is
[39:15] corporate
[39:15] return on
[39:16] investment (ROC), and that
[39:16] brings us to what
[39:17] I was writing about, which is the
[39:18] quantitative
[39:18] easing side. I
[39:19] really think
[39:20] people should
[39:20] focus on
[39:21] that. I just wanted to
[39:22] highlight this. Michael
[39:23] Dell published this a
[39:24] year ago, and it's becoming more and more
[39:25] important as
[39:26] I talk to people,
[39:28] because they really
[39:28] believe that we're
[39:29] running out of
[39:30] computing
[39:31] power. We are here,
[39:33] mostly, because of the
[39:34] agent side. to the side of
[39:37] reasoning.
[39:39] Video generation, I
[39:40] just want to show
[39:41] you how much video it
[39:42] takes, because every time
[39:43] I hear people—I
[39:44] wrote an article about VLM,
[39:45] which is video language
[39:46] models.
[39:48] Video generation—that's how much
[39:49] more it is compared to
[39:50] basic chat. So,
[39:52] chatbot versus this, 10
[39:54] seconds, guys. So
[39:57] what could a movie be like?
[39:58] How much more
[39:58] computing
[39:59] power would it
[40:00] take to
[40:01] create a movie, a
[40:01] full 2-
[40:02] hour movie,
[40:03] which we know is
[40:04] coming? The demand for
[40:05] computing will be
[40:06] eternal. Obstacles to
[40:10] AI. Again, I've
[40:11] shown this before.
[40:13] I'm only showing this
[40:14] because all of this, this
[40:15] year will be a
[40:15] cyber issue. We still
[40:17] have a negative
[40:17] reaction to data
[40:18] centers. We have no
[40:19] electricity. We have
[40:21] cooling problems. We have
[40:22] many bottlenecks.
[40:23] We have government
[40:24] competition with
[40:25] open source. All of
[40:26] this will be present,
[40:27] guys. That's why it's
[40:28] short-term noise
[40:29] versus a
[40:29] long-term
[40:30] secular boom. As long as
[40:31] profits are growing,
[40:32] all this only prevents
[40:34] it from turning
[40:35] into a "bubble." That's
[40:36] what these things do.
[40:38] Friction keeps it from
[40:39] going too far
[40:40] too fast. I
[40:41] showed it. This is the world of
[40:43] software
[40:43] . We
[40:44] are in this
[40:45] world. There will be a lot of things
[40:47] . And if
[40:47] every time these
[40:49] stocks fall, you
[40:50] start to get out of
[40:51] them, worry, and
[40:52] focus on the fact
[40:53] that Iran and the US are at
[40:54] war again. Oh my god, oil
[40:56] will be more expensive.
[40:57] How many times
[40:57] will people give in
[40:58] to the fact that they are
[40:59] growing, and then,
[41:00] suddenly, when we
[41:00] need them to
[41:01] fall, they fall
[41:02] back. We have
[41:03] volatility. We will
[41:04] always have
[41:04] volatility.
[41:05] Credit spreads.
[41:06] We will have periods
[41:07] when credit spreads
[41:07] widen.
[41:08] Perhaps right now is
[41:09] one of those
[41:09] moments. We see
[41:10] this in the growth of
[41:11] hyperscaler spreads. The world does
[41:12] not end during these
[41:14] periods. This creates
[41:15] a correction, but stocks
[41:16] usually
[41:16] outperform, and by the
[41:17] time
[41:18] things
[41:18] improve, the
[41:19] stock market has already
[41:20] risen. So, here is the price of
[41:21] oil. This is a
[41:22] December contract
[41:23] for this year. So, we are
[41:24] almost back to the
[41:25] highs. And again
[41:26] , I want you to
[41:27] look at this. We
[41:29] spent 2023-2024
[41:31] around the average
[41:32] mark, say 68.
[41:35] Now we're at 78 and
[41:37] people are panicking, even though it
[41:39] could have been 200. It's
[41:40] worth it. It's not
[41:42] worth it. Gasoline prices,
[41:43] yes, futures are
[41:44] high here. The price of
[41:45] gasoline continues
[41:46] to rise. So, will
[41:48] we see a jump in the
[41:49] overall
[41:50] Consumer Price Index (CPI)? Yes
[41:52] , I don't think it
[41:53] will be dramatic, but
[41:54] could we see a
[41:56] rebound? Of course. Here is the
[41:57] 100-day moving
[41:58] average or 100-day
[41:59] rate of change of the
[42:01] oil price. So, you see
[42:02] what's happening
[42:03] now. We have already
[42:04] increased the CPI on a
[42:05] monthly
[42:06] basis. We won't
[42:09] get very far,
[42:10] guys.
[42:11] Fuel prices at gas stations have
[42:12] not
[42:13] completely decreased. That's what we're
[42:15] looking at right now
[42:16] in terms of the
[42:17] July inflation rate,
[42:19] which is due out,
[42:20] I think, in the second
[42:21] week of August. So, for
[42:23] the first time, we will see a
[42:24] non-zero
[42:26] consumer price index in
[42:26] mid-September. Do you
[42:28] really want to bet on that
[42:29] ? By then,
[42:30] perhaps the price of
[42:31] oil will fall again.
[42:33] That's why this
[42:34] happens. When
[42:36] oil first
[42:36] rose in price, we
[42:37] saw
[42:38] one-year inflation
[42:40] swaps go up
[42:41] with it. Then
[42:43] inflation swaps
[42:44] started to decline.
[42:46] They didn't move
[42:47] , guys.
[42:49] Smart people ignore the
[42:51] actions of oil, saying that
[42:52] inflation will rise.
[42:54] Here again are two-year
[42:55] TIPS investments,
[42:56] protected against
[42:57] inflation. Same
[42:58] story. The price of
[42:59] oil has fallen. Okay,
[43:01] we won't buy into this anymore
[43:02] . Well, oil would have to
[43:04] go
[43:05] much higher. Here are these
[43:07] two swaps. Here is the
[43:09] inflation rate (CPI)
[43:10] year-over-year for the
[43:11] last five years. I
[43:13] don't know how many more
[43:13] times I have to
[43:14] show this. So if
[43:15] you're worried about
[43:16] oil and its impact on the
[43:17] Federal
[43:18] Reserve and everything else, I
[43:21] just want to show
[43:22] you this, because more and
[43:23] more people now believe
[43:24] that the Fed is going to have to
[43:25] raise rates. And
[43:26] , as I said, they
[43:27] can raise the
[43:28] stakes. We now
[43:29] have a 35% probability
[43:30] for July. This won't kill the
[43:32] market. Uh, here's
[43:34] the salary.
[43:35] Sorry, here are the two-year
[43:37] rates. So, this is
[43:38] a rise in two-year
[43:39] rates. Here is the median
[43:43] salary. Here is
[43:45] their relationship over
[43:46] time. You can't
[43:48] avoid it. The
[43:49] labor market is not strong. Not
[43:50] when it comes
[43:51] to wages.
[43:52] Not when it comes to
[43:54] labor force participation rates.
[43:55] Labor force participation rate.
[43:56] So, the Fed
[43:57] is approaching
[43:57] one of the most
[43:58] unpredictable
[43:58] meetings in recent
[43:59] years. And that's because, well, there are
[44:00] no more
[44:00] forecasting guidelines.
[44:01] Everyone is just in a panic.
[44:03] That means, oh, he
[44:04] can leave. He can
[44:05] leave. Maybe 36%
[44:07] chance. Now
[44:10] we have a forecast again for
[44:11] one and three
[44:12] quarters until the end of the year
[44:14] . Here are the rates
[44:16] rising around the world. This is
[44:17] not just a US problem.
[44:18] Economic growth is
[44:20] good. Credit
[44:23] default swaps on Oracle
[44:25] are rising sharply. This
[44:26] makes people afraid.
[44:29] A guarantee of 7
[44:29] billion dollars. I
[44:31] can't even begin
[44:32] to convey how
[44:33] insignificant this number is in a
[44:34] world of hundreds of
[44:34] billions of dollars.
[44:36] So, let's move on to the
[44:37] Clarity Act.
[44:38] Besson says, "I think
[44:40] we're already at the finish
[44:40] line with the
[44:41] Senate Transparency Act."
[44:43] Still below 50%. So
[44:45] they don't listen to
[44:46] Besson. David
[44:48] Solomon says it's
[44:49] time to advance
[44:50] crypto legislation.
[44:53] My buddy Chris
[44:55] Perkins, um, is bringing in
[44:59] new leaders to
[45:00] combine crypto, AI,
[45:01] and traditional
[45:02] finance. Bank of America and the
[45:08] guy he used to
[45:09] work with at Citi, Sandy
[45:10] Call, have written some really
[45:12] important, uh,
[45:12] crypto material—
[45:14] a lot of articles about the
[45:15] importance of agent-based
[45:16] AI for crypto and
[45:17] blockchain. And I
[45:20] will continue to tell
[45:21] everyone who is not spending
[45:22] time on crypto: in a
[45:25] year you will spend
[45:26] too much time
[45:27] catching up with others.
[45:29] Take the time to
[45:30] understand why it is
[45:31] impossible to ignore
[45:33] the fact that blockchain
[45:34] and crypto are being embraced by
[45:35] Japan, Korea, and
[45:36] every other country. And
[45:40] now, as
[45:41] they try to pass the
[45:42] Clarity Act,
[45:44] everyone is competing, just like in
[45:45] AI, to see who can
[45:46] take over something as
[45:47] important as
[45:49] AI—the global
[45:50] financial system.
[45:54] Pay attention to
[45:56] crypto and blockchain.
[45:57] Remember when Mark
[45:58] Zuckerberg said that
[46:00] AI agent technology was developing more
[46:01] slowly than
[46:01] expected?
[46:02] Going back to
[46:03] Sandy Call's article, I
[46:05] want you to think
[46:06] about this in the context of
[46:07] the movie "50 First
[46:08] Dates" with Adam
[46:09] Sandler and Drew
[46:10] Barrymore. And the reason
[46:12] I mention this,
[46:14] if you remember,
[46:15] is memory. This is
[46:20] an analogy, if you want
[46:22] to understand why the
[46:23] agent world is
[46:24] like this now: it's a chat window
[46:25] that remembers things only
[46:27] within the confines of a single
[46:28] context window.
[46:32] What needs to happen on the
[46:33] agent side is the
[46:35] amount of memory
[46:35] needed to
[46:36] store all of
[46:38] this and know everything
[46:39] that he knows: every
[46:40] experience, everything he's
[46:41] seen, everything he's
[46:42] connected to, the songs
[46:44] he learned in 2
[46:45] years. The amount of memory
[46:47] required for this
[46:48] on the agents' part is
[46:49] enormous. And that's why
[46:51] consumer agents take
[46:53] so long to appear
[46:54] , because you need to
[46:55] understand the person themselves.
[46:56] If Alexa
[46:57] talks to me,
[46:58] I don't want to go back
[47:00] . I want to be
[47:01] able to say, "
[47:02] What's the recipe for this?" I
[47:03] want to ask: when
[47:04] we cooked this together
[47:06] last time, what was I
[47:07] doing? Think about
[47:10] all the things in life that
[47:11] you would simply
[47:12] outsource to AI
[47:13] so you never have to
[47:14] remember anything
[47:15] yourself again. S&P just
[47:18] launched its first
[47:20] crypto index on
[47:21] Bitcoin. I told this
[47:22] whole story about "50
[47:23] First Kisses"
[47:25] because I want you to
[47:25] understand: when we
[47:27] get to the
[47:28] consumer agent stage, the
[47:29] cryptosphere is going to explode
[47:30] in volume. Once we
[47:31] get to that point and
[47:32] things start to grow,
[47:33] it will be massive.
[47:34] So despite everything that was
[47:36] happening, the
[47:37] Clarity Act dropping to the 30s and
[47:38] all that, Bitcoin
[47:39] had another quiet
[47:41] week. So, he
[47:42] kept his gains.
[47:44] It's still stable
[47:45] around 64,000. If
[47:49] you look at it, it's
[47:50] Bitcoin. This is Ethereum.
[47:52] They are both holding on.
[47:53] They bounced off
[47:54] the lows. They have not yet
[47:56] come out of their "
[47:56] bear" market. This is still a
[47:57] bear
[47:58] market until we
[47:59] can break through the 200-
[48:00] day moving average
[48:01] on them. The most important thing is that
[48:02] I show it
[48:03] more and more often. This is Dogecoin.
[48:05] Yes, it's a memecoin, but the
[48:06] most important
[48:07] memecoin. And to me,
[48:09] when people want to beta
[48:10] test the market
[48:12] and jump into
[48:13] things like GameStop, it's a
[48:14] signal that retail is
[48:15] active. The energy in
[48:16] retail is still
[48:17] lacking until we
[48:18] start to see a
[48:19] positive
[48:20] Dogecoin trend. And here's one
[48:21] way to measure it.
[48:23] We were below this 20-
[48:24] day moving
[48:25] average on Dogecoin. We have been
[48:27] below it since mid-
[48:28] May. It's been 67 days.
[48:33] This was as of
[48:34] Wednesday. 67 days.
[48:38] The longest in its
[48:39] history. This
[48:40] matters. We
[48:41] closed above it. I
[48:43] think this is important.
[48:44] Finally, one more thing.
[48:46] Add Russia to
[48:47] the list. Parliament
[48:48] has passed a law
[48:49] regulating
[48:50] the crypto market. Well, that's it for the
[48:51] week
[48:52] , guys.
[48:53] Remember, if you
[48:54] would like
[48:55] more details as an
[48:55] investment
[48:56] advisor (RA), please
[48:57] contact Mark
[48:58] Weiling. This is M Whaling.
[48:59] Mwh22vresearch.com. We do
[49:05] much more for
[49:07] RAAS. But most importantly,
[49:08] everyone is starting to
[49:09] turn to the
[49:10] cryptosphere. I will launch the
[49:11] new crypto projects I
[49:12] talked about in
[49:13] September, late
[49:14] September or
[49:15] early October. I look forward to
[49:17] dedicating
[49:18] one YouTube video per
[49:19] week to this for those
[49:20] who are interested. But more
[49:21] importantly, I
[49:22] think everyone should
[49:23] pay attention to what's
[49:24] happening in the
[49:25] crypto space. Okay,
[49:25] guys. See you
[49:26] next week.