Episode Transcript
Transcripts are displayed as originally observed. Some content, including advertisements may have changed.
Use Ctrl + F to search
0:00
The Avengers assemble once
0:02
again. Rave of
0:04
the rhodium group JP Kleinens of
0:06
the Think Tank SNV, Jay Goldberg
0:10
of
0:10
digits and dollars and Dylan Patel
0:12
who writes the semi-analysis sub
0:14
stack here to talk about
0:16
chip sack, tech
0:19
cold war, uh, Euro
0:21
chips act, AI, China
0:24
retaliation, autonomous driving,
0:26
and whatever else we get into. Welcome to China
0:28
Talk, everyone. So
0:31
a few weeks ago, the CHIP's Act released
0:33
its first request for
0:36
funding. So all the industry
0:38
who's doing the entire
0:41
global semiconductor industry who's interested in getting
0:43
money to make manufacturing in America
0:46
now has their guidelines
0:48
of what they're going to have submit in
0:50
order to hopefully come
0:52
away millions or even potentially
0:55
billions of dollars in support
0:57
for building in the US? What were everyone's
0:59
impressions of that
1:01
What did the priorities seem to be about
1:04
what the administration is most focused on trying
1:06
to get built in the US? To
1:08
me, it seemed like a lot of the priorities were around
1:11
leading edge and packaging, advanced
1:13
packaging, and maybe so much
1:15
traditional packaging and trailing
1:17
edge process technology, at least. It seemed I
1:19
mean, it certainly are gonna be, as
1:22
efforts in that direction, to me,
1:24
it just felt like they were a lot more concerned about
1:26
the leading edge, which is just an interesting
1:29
to down.
1:31
Yeah, when they say two leading edge
1:33
networks or hubs, that's a lot
1:36
of money and, and getting,
1:38
you know, two places in America up and
1:40
running on leading edge is like committing a
1:43
not insubstantial portion of the $40 billion,
1:46
which is going to be up for My sense
1:48
was that they're still kind of leaving their
1:50
options open and there almost weren't
1:52
a ton of choices that were made in this document.
1:55
it kind of listed like everything
1:57
that you could think of them ever
1:59
wanting to do. So you saw leading edge,
2:01
you saw lagging edge, you saw packaging.
2:04
And the sort of hierarchy
2:06
of needs was just, we
2:09
want you to address national security and economic
2:11
objectives. And you don't really have the rubric of, we're
2:15
giving 50%
2:16
to leading edge and 50% to everything
2:18
else. You didn't quite get
2:20
all the way to that level of thinking.
2:22
Anyways, Jay, did anything strike out to you
2:25
about the broad funding priorities?
2:27
I mean, it looked to me
2:29
a lot like a policy document
2:31
as opposed to a technology document.
2:35
It didn't strike me that it was really clear, strategic
2:37
on which technologies really mattered. But
2:39
I mean, there is an interesting negotiation
2:43
dynamic thing here when you say, 100%, I want
2:47
leading edge and nothing else, then all of a sudden
2:49
the TSMCs and Samsungs of the world, their bargaining
2:51
position all of a sudden gets a lot better, right? Yeah,
2:54
but that again, it speaks
2:56
to almost a political dynamic
2:58
where it's like,
2:59
they clearly negotiated a bunch of deals ahead
3:01
of it for TSMC and some of the others.
3:04
And that's why those are seem prominent. It's
3:06
not from a sort of best practices technology
3:08
perspective. It's just like, hey, we have this deal.
3:11
Let's make sure that there's language in the policy
3:13
that covers deals we've already closed. And
3:16
then everything else is sort of like, here's our priority
3:18
list. I think they didn't
3:21
negotiate because they say two leading edge fabs,
3:23
right? Well, San Antonio's got a huge announcement in
3:25
Texas, right? Intel's got a huge announcement
3:27
in Ohio and TSMC's
3:29
got a huge announcement in Arizona. And
3:31
so all three of those could be significantly
3:33
paired back
3:35
or you know, actually go through depending on
3:37
how these negotiations go through. I don't actually
3:39
think they negotiated these deals yet,
3:41
right? I think they're gonna play each other off
3:43
of them and in the end one of the three will
3:45
maybe not get the money that they want.
3:47
and they will not come up to bat. Because
3:49
I think all three are leveraging for as much hip-sacked
3:52
as possible, right? I think that's
3:54
true. I also think some of them
3:56
might not actually take the deal. I think, you
3:58
know, I have to look at Samsung, who's our.
3:59
committed a lot of this before the CHIPs
4:02
Act and they started looking at this, are
4:04
they really gonna be, I'm not sure they're gonna be excited about
4:06
it or want to take that money, which would be
4:09
an interesting choice. They're leaking
4:11
stuff for Reuters, right? Yesterday I think they
4:13
got Reuters to say that the costs went up
4:15
by X percent. It was like five
4:17
billion dollars or something, right? Versus what they had
4:20
planned a couple years ago. So I think that's just
4:22
more negotiation leverage that they're trying to put
4:24
out there, I imagine.
4:26
So So, Revo, we
4:28
have money, but there are also many, many
4:30
strings attached that these companies
4:33
are going to have to weigh whether or not they want to take the chips
4:35
act money. What were some of the more dramatic or
4:37
potentially more costly asks that the
4:40
chips act is going to make on these companies
4:42
that are considering taking this money?
4:44
Yeah, from a policy perspective, industrial
4:47
policy provides a fantastic
4:49
opportunity for governments to impose
4:52
conditions on industry. And
4:54
you can see that in the language of
4:57
the CHIPS Act and in the provisions
4:59
that we see coming out now in
5:01
this implementation phase. So
5:04
one line that really stood out
5:06
is on the clawback provision,
5:09
which said that commerce can clawback
5:12
funding if recipients, quote,
5:14
knowingly engage in any joint research or technology
5:17
licensing effort with foreign entities of concern
5:20
that
5:20
raise national security concerns. obviously
5:24
very broad and the Secretary
5:26
of Commerce has the remit
5:28
to interpret that, adjust
5:31
that definition over time.
5:33
And that should make companies of course nervous,
5:35
right? Samsung and SK-Hynix are already in
5:38
a really uncomfortable position just
5:40
being beholden to BIS
5:43
licenses to continue operating
5:45
their fabs in China. And
5:48
the signal has already been sent, right? steer
5:50
your leading edge memory chip production out
5:52
of China, you're on a timeline, go. So
5:55
I think part of the reticence you're
5:57
seeing from companies like Samsung is
6:00
Well, how tight are these conditions
6:02
going to be? How will that US
6:04
interpretation evolve over time?
6:07
And then you see TSMC is
6:10
like, yeah, no, this is all very reasonable.
6:13
We're good. We're going to go for it. So
6:16
it's really interesting to see kind of middle
6:18
power balancing manifested
6:20
through these companies' policies as
6:23
they're reacting to what's coming out. Sure,
6:25
sure. So there's a provision of
6:27
the ChIPSCAC that basically says that
6:30
any excess profits would be shared with
6:32
the government,
6:33
which is very interesting
6:35
because I don't know policy, but I don't think
6:37
that's happened a lot with prior sort of subsidies
6:40
to companies in various
6:42
industries that the US government has subsidized. So
6:45
like, what's up with that? Yeah, I don't know.
6:47
Yeah.
6:48
I read that as they
6:50
pasted that on in the last minute because it emerged
6:52
that the Eurochips Act had those
6:55
provisions in it. And I think some of the US policymakers.
6:57
So we thought, oh yeah, we should have that too.
6:59
That sounds like a good idea. Actually,
7:02
depending on your reading, I would
7:04
say some of the provisions
7:06
in Europe are even a little bit
7:09
more drastic. So in the
7:12
one subsidy vehicle,
7:14
it is negotiated that for companies,
7:16
they have to make the financial gap
7:19
analysis and tell the government this is
7:21
kind of the delta that we are looking at
7:23
and this is why we need subsidies.
7:25
And then some governments argue
7:28
that, well, obviously
7:30
you have a lot of knee-way with this
7:32
financial gap analysis because if you
7:34
calculate extremely conservatively
7:37
how profits will develop in your market,
7:40
then your financial gap analysis might
7:42
be extremely big,
7:44
even though the market is
7:47
extremely profitable for you. So
7:49
over the seven year period that the
7:52
subsidies are going out,
7:54
if your profit is
7:57
above a certain percent, You
7:59
have to... pay back the entirety
8:01
of the subsidy.
8:03
So some
8:05
companies have to put back
8:07
money
8:08
or put aside money because
8:11
depending on how the market develops, they
8:13
might have to pay back
8:15
all of the subsidies that
8:17
they have received. And I mean,
8:19
this is definitely
8:20
an interesting situation, I think, for companies
8:23
where you receive subsidies, but at
8:25
the same time, you have to put aside money because if
8:28
your investment is too profitable. And if the
8:30
market develops too nicely, you
8:32
might have to pay back all of those subsidies.
8:35
How have companies responded to that on the
8:37
European side so far? And Peter? I
8:40
mean, interestingly, there's
8:42
not a huge discussion about it in Europe
8:44
on the public level. But
8:47
there's a very
8:49
active conversation among member
8:51
states because as always in Europe, say
8:54
it's different on their perception of how
8:56
subsidies should be used, how stringed
8:59
the rules should be and so on. So
9:01
to use a cliche you can imagine that
9:03
the more frugal
9:04
Scandinavian countries are more opposed
9:07
to the idea to just give away subsidies
9:09
freely
9:11
and want to have strings attached in
9:13
terms of profit margins
9:15
whereas other countries including
9:17
the one where I'm have a
9:20
different idea of how to give
9:22
away subsidies to companies. It's
9:24
a little bit of a mess, but it's interesting to
9:26
see that for this particular regard, apparently,
9:29
as Jay put it, the US government got
9:31
maybe some inspiration from Europe,
9:33
which is definitely interesting. Maybe
9:36
just as a side note to the previous conversation,
9:39
for me, kind of the big underlying
9:41
question in all of this,
9:43
because Jay, you mentioned before that for you
9:45
it read more like a policy document document and
9:47
less like a technology document.
9:49
For me the underlying question is to what
9:51
extent can governments
9:54
be reasonable technology pickers?
9:57
Because ultimately if you look at industrial policy
9:59
and
9:59
the implementation of the US chipsec, the implementation
10:02
of the EU chipsec,
10:03
governments
10:05
want to be the winning
10:07
technology pickers.
10:08
And I'm simply looking
10:11
at the financial sector and how tricky it
10:13
is to pick a winning technology for VC.
10:16
I'm just wondering how the
10:18
government's best place to
10:21
make decisions, for example, should
10:23
we subsidize the silicon carbide fab
10:25
or should should we subsidize the gallium nitride
10:27
fab? Or what's up with all
10:29
the analog processors for AI?
10:32
Should we give subsidies to chip design for analog
10:34
processors? Or is it really about
10:37
wafer level packaging? Or like is government
10:39
in the right place to make those
10:42
rather intricate and long-term decisions
10:45
on the technology level when the
10:47
sector itself already struggles to identify
10:50
winning technologies in the long term? My
10:52
knee-jerk reaction to that is
10:55
no, but if you think historically,
10:57
there
10:57
have certainly been a lot of instances
10:59
in which the US government has had a technologically
11:03
savvy policy towards tech
11:06
and what technology is to prefer. And usually it's
11:08
rooted in some basic
11:10
need of the government,
11:12
usually around defense and military.
11:15
And so I'm just wondering to what extent some
11:17
of these priorities are reflected
11:19
in the in the CHIPS Act. But certainly the
11:22
CHIPS sanctions back in October were
11:24
clearly heavily influenced by the Defense
11:27
Department and adjacent agencies.
11:30
I don't get the sense from
11:32
this latest document around the CHIPS
11:35
Act. You know, this sector of the
11:37
US economy or the US government needs this
11:39
technology and we're going to favor it. It
11:41
feels broader to me.
11:43
And I think that's reflective of it coming out of commerce,
11:45
which itself, the Commerce Department doesn't have
11:48
a big need need for
11:49
advanced chips the way that the Defense Department does.
11:52
I don't get the sense that we have a clear laundry
11:55
list of technologies that we need. But
11:58
it can evolve over time, right? I mean that's. Part
12:00
of the design of this
12:02
is to stay ahead
12:04
in that innovation curve. I
12:06
mean, we'll see how that evolves,
12:09
but the trailing edge
12:11
policy debate still seems to be very much in
12:13
play.
12:14
I mean, yes, the emphasis is on leading
12:17
edge, but you've still got $10 billion
12:19
for
12:20
legacy chip manufacturing.
12:23
There's still big gaps there. Of course,
12:25
policymakers want to steer sourcing
12:28
away from China. That's hard to do,
12:30
just given market dynamics. But
12:33
it's interesting to see where some emerging policy
12:35
areas are cropping up to,
12:37
to try to address that. For
12:40
example, in demanding
12:42
more transparency on subsidy
12:45
disclosure. This is where the
12:47
EU is already a couple of steps ahead of the
12:49
US
12:50
in throwing up procurement
12:53
investment restrictions around subsidy
12:56
driven just market distortions,
12:58
the US is playing around with that now too.
13:00
So there are there are still ways for the US
13:03
to, I think, try to steer
13:05
some of that policy and diversification,
13:08
although it's again, the economics of this is
13:10
really hard for us companies, companies
13:13
to reconcile with.
13:14
But it's this flexibility built
13:16
into these latest policies. And my question is,
13:19
do
13:19
we think that's a feature or a bug? Because
13:22
the glass
13:24
half full version is like what you said, which
13:26
is they're going to adapt with
13:28
time depending on needs. The glass
13:30
half empty view of that is they
13:32
don't really know what they want and it's going to be come down to
13:35
backroom deals and last minute horse
13:37
trading
13:38
as opposed to a coherent policy objectives.
13:41
You got to have a sense that there's been more strategy
13:44
thinking within the CHIPS Act than just
13:46
what we saw in the vision document
13:48
and the notice of funding opportunity.
13:51
Yeah, so I question to this,
13:53
what's your reading to what
13:55
extent, I mean,
13:57
in Europe the conversation about subsidizing
13:59
chips.
14:00
from the beginning was focused
14:02
on front end and very much focused on cutting
14:04
edge front end because that's the stuff we don't have at
14:07
all. But then over the 18 month or so, it
14:12
got a little bit more nuanced that policymakers
14:15
understood, oh, there's this thing called
14:17
packaging and oh, there's backend and packaging
14:19
is advancing. And then you call it advanced packaging
14:21
and maybe we need that. And I still remember
14:24
in one of the requests for comments
14:27
from DOC, a couple
14:29
of US PCB manufacturers
14:32
also replied and said, you know what, if you're worried
14:34
about national security, don't forget
14:36
about us. And with PCBs,
14:40
printed circuit boards, the interesting thing is
14:42
actually that China is not lagging behind.
14:44
Actually, the most cutting edge stuff is in
14:46
China for PCBs.
14:48
And it is more that the US and Europe
14:51
are way behind in PCB
14:53
manufacturing.
14:54
So what's your reading from conversations,
14:57
also looking at UJ, hypothetically,
15:01
if I am a really cool
15:04
PCB startup or a small
15:06
PCB company, do
15:09
I get anything out of the chip set?
15:11
Because out of the EU one, you don't get anything out.
15:13
So I don't know the answer to that,
15:15
but I will say that being
15:18
an advanced PCB manufacturer
15:21
is not as exciting as it sounds. That's
15:23
a pretty miserable business,
15:25
very low margin, fairly
15:27
labor intensive. Right. And
15:30
for sort of standard packaging, I don't,
15:33
I don't know how much that's going to come back to the U S now, advanced
15:36
packaging, that's important. And Dylan
15:38
can talk about this way more than I can,
15:40
but there's, you know, some of the manufacturing stuff
15:42
is super, super hard to do,
15:44
and it's going to be done by the foundries,
15:46
but the bulk of the business in the place where.
15:49
Where we're really talking about PCBs. That's, you
15:52
know, that business is not coming to the US. like
15:54
putting stuff on a board. If it's not
15:56
in China, it's going to go
15:58
to Malaysia.
15:59
But I mean, at the same time, for
16:02
me, the weird thing is, since we
16:04
are all doubling down on national security,
16:07
if you are worried about national security,
16:09
the PCB should be of concern
16:12
to you. Like maybe even more
16:14
so
16:14
than the 15
16:16
billion transistors
16:19
in your cutting edge processor, but
16:22
hiding
16:23
or compromising a chip that then ends
16:25
up on the PCB is much
16:28
more viable in terms of an attack
16:30
vector
16:31
than trying to infiltrate Intel's
16:33
or NVIDIA's next cutting edge processor
16:36
and hiding a backdoor among the
16:38
billions of transistors. I'm just
16:41
I'm just I just find it interesting that
16:44
for
16:44
some reason with PCBs, whether
16:46
US government forgot to think about
16:49
the national security angle. The question
16:51
is, if the policymakers believe
16:54
that there is a certain level of minimal
16:56
minimum production that is in the US, or
16:59
in Europe for various technologies, right?
17:02
Should we be able to make a,
17:04
you know, an advanced Nvidia AI server
17:06
and the networking equipment to connect them all together,
17:09
even if it's a small amount? And to Jay's
17:11
point, right, that's not economically viable, right?
17:13
There's too much labor involved. A front end fab
17:15
requires very few people and labor
17:17
is, you know, less than 20% of the cost of
17:20
a final chip. It's all capital, right? And
17:22
so that's something that maybe makes more sense and
17:24
is more automated, at least from an economic
17:26
standpoint. But then our policymakers going
17:28
to subsidize, hey, actually, we want to be able
17:31
to have, you know, 3% of the world or 5% of
17:33
the world
17:33
or 10% of the world's PCB manufacturing here.
17:36
But remember, there are multiple policy
17:38
tools in play, right? So your industrial
17:41
policy is your incentive, but
17:44
then the other huge whopper
17:47
of a policy that's emerging now
17:49
post October 7th, I think the next big
17:51
thing is now the restrict act,
17:53
which last week was unveiled.
17:56
Senator Mark Warner is
17:58
behind this. of course a very serious
18:01
player in Congress.
18:02
And it's really interesting because this is
18:04
something that focuses on ICT
18:06
supply chains. Remember, commerce
18:09
already has ICTs rules.
18:11
In effect, this was something that was
18:14
introduced on the last day of the Trump
18:16
administration, and then
18:18
went into effect a little bit
18:20
after that by the Biden administration where
18:23
they carried through.
18:24
And we've been waiting to see, okay,
18:27
will this be fully utilized? The
18:29
TikTok debate has just catalyzed
18:31
this. But if you look at the text of
18:33
the restrict act, it is really,
18:35
really deliberately wide
18:38
ranging.
18:39
And it's meant to cover hardware,
18:42
software, services, you know,
18:44
just the provision of, you
18:46
know, those those services, anything
18:49
around ICT supply chains that
18:51
has a nexus with a foreign entity
18:53
of concern.
18:54
So,
18:55
you know, to your point, Peter, on, okay,
18:58
well, what about,
18:59
you know, printed circuit boards and the national security
19:02
risk?
19:02
Well, this is the catchall for
19:05
something like that, potentially. I'm
19:07
struggling to think of a significant
19:10
domestic PCB maker.
19:13
And I'm sure I'm going to get some angry comment,
19:15
I forgot somebody, but I don't think there is a constituency
19:17
in the US who's advocating for this, because
19:19
we, we, I don't think we make many of them any
19:22
PCBs anymore. So maybe coming
19:24
up from PCBs for a second, every
19:27
single company at any point
19:29
in the value chain is going
19:31
to do the sort of thing that whichever
19:33
PCB manufacturer
19:36
hired a smart lobbyist
19:38
to help them frame their thing as a national
19:41
security perspective, such as to
19:43
get JP and Jay and us talking about
19:46
this on China Talk. And the, you
19:48
know, having to sort
19:50
of sift through all of
19:53
those hundreds, I mean, potentially
19:55
even like thousands of companies
19:58
that are going to be asking for support saying look,
20:01
maybe you weren't aware of this, but our thing
20:03
is really important to national security.
20:06
So give me a slice of the pie. And
20:08
I'm sure there's actually going to
20:10
be a lot of learning and processing.
20:13
And interestingly, in the NOFO as well, there
20:15
was a lot of
20:16
room for back and forth, where
20:19
they would send an initial 20 pages, and
20:22
then the Commerce Department would say, oh, we want to learn more about
20:24
this, that, and the other thing. So I imagine
20:28
over that iterative process, there
20:30
will be maybe priorities
20:33
that the government will get convinced
20:35
of that aren't necessarily the
20:37
ones that policymakers have already
20:39
internalized around, say,
20:42
leading edge logic, for instance.
20:44
Should we talk about childcare for a second? The
20:48
interesting thing about that is those provisions
20:51
that got thrown in on childcare
20:54
have kind of reopened some of the
20:56
partisan vibes around
20:59
this, you know, where the Republicans
21:01
are saying, wait, you can't just throw that
21:03
in there. I thought there was some consensus around
21:05
this, like how many standards do you want to
21:07
pile on? Then there's, of course, a legitimate
21:10
argument about, you know, what kind
21:12
of labor standards the US wants to lead
21:14
with here. But then you've got Morris Chang over
21:16
here in TSMC saying, Oh, my
21:18
God, the labor costs are so expensive to Bruce
21:21
in China and in the US and globalization
21:24
is dead, but this
21:25
is the new world we live in.
21:27
So, I mean, I don't think
21:30
it stops this in its tracks or anything.
21:32
It's just a revealing
21:34
kind of political element to
21:37
the policy evolution. I
21:39
mean,
21:40
you spend money and you're gonna push as far
21:42
as you can and you're gonna get some sort
21:44
of return on your investment from sort
21:47
of economic security, national
21:49
security, securing the future
21:51
of the industry creating an ecosystem
21:53
or whatever. And like, you know, there's
21:55
a give and take with all of this. and maybe it was
21:58
surprising to me.
21:59
that it made it in there.
22:02
And then it was kind of like the one thing
22:04
that the national media picked up on with
22:06
this because we've just been talking for 30 minutes
22:09
about lots of other things that I think are gonna
22:11
be far more consequential than whether like a
22:13
thousand employees in the US
22:16
end up having
22:17
more subsidized childcare than they would otherwise,
22:19
which is like important. And I
22:22
think everyone should have childcare in America, but
22:24
I'm not sure that like the future of
22:27
America's semiconductor ingos ecosystem
22:29
is going to hinge on whether or not TSMC's
22:33
Arizona FAB has like
22:35
C plus or A minus preschool.
22:39
What is tracking to me, and I'm wondering
22:41
what you make of this,
22:44
the Department of Commerce is on a hiring
22:46
spree for the implementation of the
22:48
US chipset. And some
22:50
of those hires are really impressive.
22:54
Right, like the Department of Commerce
22:56
got, among others, the
22:59
chief economist from one of the
23:01
largest semiconductor companies
23:03
in the world
23:05
to now be the chief economist
23:07
for the implementation of the US chip check.
23:10
These are people who know what they talk about
23:12
when they talk about chips, right?
23:15
So I find this pretty
23:17
impressive in terms of not just
23:19
talking the talk, writing
23:22
regulations and paper, but also
23:24
working with the work and trying exactly
23:27
what you mentioned before, Jordan, how
23:29
do we spend this money most meaningfully?
23:31
And how do we sift through
23:33
probably hundreds of applications
23:36
where some of them might be sketchy? I
23:38
don't see the same happening
23:40
in Europe where we even though
23:42
the EU chip tech talks almost
23:45
about the same amount of money for theoretically 43
23:48
billion euros. But
23:51
I don't see a hiring spree in the European
23:53
Commission. I don't see a hiring spree in member
23:55
states to kind of step up the game
23:57
of industry knowledge. So to me.
24:00
it is, if we compare the two regions, to
24:02
me it is really increasing me about who
24:04
puts the money behind government
24:07
resources to actually implement
24:09
all the stuff.
24:10
So during the Great
24:12
Depression and the New Deal, there was this
24:15
sort of new class of dollar a year
24:17
men, and they were all men back then, who
24:19
basically came in from industry and
24:22
were there to spend all the money that FDR
24:25
got appropriated to rebuild America.
24:27
And I think it's
24:28
worth acknowledging on this podcast
24:31
that the pay cuts that all
24:33
these people from industry are taking by
24:35
doing these jobs are factors of 5 or 10.
24:40
I mean, the CIO spent 30
24:43
years or something at KKR, and
24:45
he probably isn't making more than $200K
24:47
to do this job.
24:48
And this is public service. And
24:51
this is kind of like because
24:53
our government hasn't figured out how to pay talent
24:56
appropriately, this is sort of what we need
24:58
to make a real national
25:00
push in something where the sort of like,
25:03
the alternate employment opportunities are just so
25:05
much more lucrative. So it has been
25:07
really encouraging for me to see the type of folks
25:09
that the administration has been able to persuade to take
25:12
some of these opportunities because they're
25:13
not doing it to get rich. We've talked for 30
25:16
minutes, like these are incredibly difficult
25:19
things to weigh and trade offs to make that
25:21
you can't just figure out
25:24
by having looked at these things for like a few
25:26
months making a decision. Because if you do that,
25:28
then like the lobbyists will beat you and
25:31
you won't be able to get the best bang for your
25:33
buck. So we're recording this on Friday,
25:35
March 17th.
25:36
Yesterday, I
25:38
watched Microsoft's Windows 365
25:40
demo like three times
25:43
and was
25:44
also just screaming in delight
25:46
at what these potential applications
25:49
for Word, Microsoft, Excel, and Outlook
25:51
could do.
25:52
Dylan, what's your take on
25:54
how these sorts of applications of generative
25:56
AI can change the world.
25:58
Car Care
26:00
Routine with WipeNew Auto Appearance Products.
26:03
WipeNew includes everything you need to transform
26:05
your car from dirty and dull to dazzling,
26:08
from waterless car wash to spray wax
26:11
to tire shine. WipeNew Car Care Products
26:13
are affordable, easy to use, and deliver
26:15
outstanding results. Get free shipping
26:18
on orders of $49 or more when you
26:20
shop the full line of WipeNew products
26:22
at WipeNew.com.
26:24
I mean, the consequences
26:27
of the world are huge. All I was thinking
26:29
about was, hey, all
26:31
these applications of AI being pushed forward
26:34
are huge. There's already a shortage
26:37
of AI chips from Nvidia and such
26:40
in the market today, even after
26:42
the China ban. It's like, there's
26:44
a serious amount of hardware that's
26:46
needed to run all these applications. For
26:48
example, if you put large language models in search, you're
26:51
going to need tens of billions of dollars of GPUs.
26:53
If you put these large language models into every
26:55
word, I'm typing into Microsoft Word or Excel,
26:58
like, what? This is going to be ridiculous
27:00
amounts of compute
27:01
needed. And the question
27:03
is, what can you run on
27:05
those? Well, generally, the systems are
27:07
either banned or they're just about at the banning
27:10
threshold, right? NVIDIA modified some of the specs.
27:12
But as we go forward, there's a real question
27:15
if China will even be able to deploy these at
27:17
anywhere close to the same scale that the US can.
27:20
And then that's the question
27:21
of like, hey, does Microsoft
27:23
Word and Excel and Outlook and teams
27:25
with this AI meeting recognition
27:28
and all. I mean, you guys just need to watch this. Everyone needs
27:30
to watch the Microsoft event. But does that
27:32
make me 40% more productive? Yeah, it
27:35
absolutely does. So if I'm 40% more
27:37
productive, does that close a lot of the gap of
27:40
pure population advantage that China has? And
27:43
then when you add in allies and such, it's
27:45
a tremendous question of
27:47
does large language models and AI make
27:49
the whole like China is going
27:51
to take over the world because they of a bigger country with
27:53
more people, with more STEM graduates, break
27:56
that down a little bit.
27:58
So I took my five-year-old daughter.
27:59
to this event at the University
28:02
of Texas in Austin. It was a big science event.
28:04
And they did one of those demos. Have you
28:06
ever seen where they do the mouse trap demo
28:08
for an uncontrolled nuclear
28:10
chain fission reaction?
28:12
It's really cool. Like they drop a ping pong
28:14
ball into this box and then it like sets
28:17
up all the mouse traps and it's
28:19
an uncontrolled reaction. Like, right, one goes off
28:21
and then they all go off. And in my
28:23
head immediately I was thinking, holy
28:26
crap, like open AI. Like this
28:28
is what China sees, right? When
28:31
generative AI is out into the world,
28:33
right?
28:34
I'm sure it's like all around our dinner tables,
28:37
all of the discussions going
28:39
on around, well,
28:40
what is this gonna lead to in this innovation?
28:43
What business model can emerge out of this? How
28:46
do, you know, our own companies need
28:48
to stay ahead of the curve and how can they utilize
28:50
it?
28:51
Think about that conversation on
28:53
the Chinese side, right? Because this is,
28:55
it's out there. It's out into the world. You
28:58
need to innovate fast to
29:01
take advantage of it. And as you said, Dylan,
29:03
you need vast amounts of compute
29:05
power here. And China has
29:08
just been cut off from that in
29:10
a very big way. So all
29:12
of a sudden, we're just looking at this escalation
29:15
ladder when the tech
29:17
controls theme
29:18
catalyzed to a huge degree,
29:21
which really makes you wonder at this point
29:24
what the political decisions are going to be in
29:26
Beijing, like, yes, you can come out of the two sessions
29:28
and say, we got to focus on
29:30
basic R&D. And
29:32
we just need to kind of double down
29:35
and put our resources into this.
29:37
But these are
29:38
really hard constraints that
29:40
you're up against. And if you want your companies
29:43
to innovate, you can't be all over them
29:45
to censor these large language
29:47
models, right? And what they're going to churn out to your
29:49
populace, it's only going to slow you
29:51
down even more. So some really hard
29:54
choices I think going to need to be made here? And
29:57
geopolitically, that's a fascinating
29:59
question to me.
29:59
over the next three years or so. Yeah,
30:02
I mean, it's interesting because
30:04
when the October 7th restrictions were laid out,
30:06
it was, look, we're stopping supercomputers
30:09
to model nuclear bombs or whatever. It was very
30:11
much like, look, this is a national security thing.
30:13
And the messaging out of the administration was,
30:15
look, we're not trying to
30:17
slow down China's economy or whatever.
30:20
But
30:21
we now see the technology
30:24
that is going to be built on the exact same
30:26
chips that are being controlled, which
30:28
is going to produce significant
30:31
productivity gains for years
30:33
and years to come, which China might
30:35
be able to access through the cloud
30:37
in Singapore.
30:39
But it's, I think, very different.
30:41
And one of the really interesting things that has
30:44
seemed really clear in the way Microsoft
30:46
and OpenAI are working together is that you
30:48
really need to be deeply integrated
30:51
with your cloud provider. And
30:54
there's a lot of work that is going together
30:56
or between the model development and
30:58
scaling inference, which is not just
31:01
something you could hook in halfway
31:03
across the world with a cloud provider that already
31:05
has more than enough business than they can handle
31:08
dealing with Western clients that they don't have to be stressed
31:10
out about are potentially violating some
31:12
export controls which could really get you in trouble. So even
31:15
if the pain initially
31:18
with those export controls was only on the
31:20
margin, and yeah, some companies
31:21
were trying to rejigger their chips
31:24
to just get under these thresholds, now
31:26
that generative AI is really going to hit
31:28
in the next coming months and years, it
31:30
ends up likely having a much larger,
31:33
sort of broader public commercial impact
31:35
than just the dual use things that
31:38
the original regulations were ostensibly
31:40
focused on.
31:42
Coming back to you, Reva, with the sort of like
31:45
cold war stuff,
31:47
one of the most concerning
31:49
upsetting lines I saw thinking about whether
31:51
or two particularly concerning things
31:54
I've seen over the past few
31:56
weeks, which raises my expectations that
31:58
we're going to start to see.
31:59
retaliation from China is first,
32:02
you know, all of this reporting, which we've covered
32:04
on in China talk about the Biden administration,
32:07
again, telling China not to
32:09
give arms to Russia. And
32:11
then this one line from some
32:13
anonymous Chinese government official in the
32:16
Straits Times basically saying the
32:19
exact same thing that the Chinese government
32:21
started saying right before it started putting
32:23
tariffs on in retaliation
32:26
to the US tariffs back in what it was a 2017 or 2018, saying,
32:28
look,
32:31
we don't want a Cold War, but
32:33
we're ready to have one.
32:35
And whatever that ends up meaning
32:37
is going to have a real impact
32:40
and put back some of the things on the table that
32:43
we've been talking about and try to talk over the past year, that
32:46
China, for whatever reason, hasn't decided to
32:48
deploy to strike back in semiconductors or
32:51
any other industry. So
32:54
any thoughts, reflections on the risk of
32:57
the tit for tat kicking off in
32:59
an even more dramatic way than
33:02
it has so far?
33:03
Things are getting existential really
33:06
fast, right? I mean, you're already
33:08
seeing this dynamic
33:10
in play
33:11
where the US is
33:13
very consciously escalating
33:16
in tech and investment controls, boosting
33:19
defense support to Taiwan, all these different
33:21
fronts,
33:22
knowingly doing that but saying we need
33:24
guardrails, right, to prevent this from descending
33:27
into hot conflict.
33:29
Beijing's response, not having
33:31
much leverage here, just given the
33:34
state of its economy and needing to preserve foreign
33:36
investment, is, oh, you want guardrails?
33:39
Well, come and get it, right? It's
33:41
basically a balloon affair. You want
33:43
your military to military dialogues? Well,
33:46
you're going to need to de-escalate
33:48
if you want to get to a more stable state.
33:50
So we are
33:52
in a really dangerous phase
33:54
of the competition.
33:55
And I don't know if you saw, this
33:58
was also flagged by Trivium but but
34:00
there's ICY is
34:02
a semiconductor research and advisory firm based out of
34:04
Shanghai.
34:05
There was a really interesting opinion
34:08
piece where they did this classification
34:10
of
34:11
American chip offenders. There
34:13
were instigators,
34:15
which seemed to be alluding to the
34:17
big USIC companies
34:19
that they
34:20
talk about preventing the loss
34:22
of talent and technology, but they actually use
34:25
their judicial power to suppress competitors,
34:27
seek their own interests. That was one class. Then
34:29
there were
34:30
what they called white-eyed wolves, the
34:33
memory chip makers that can take advantage of controls,
34:35
and then pawns. There was a specific reference
34:38
to Dell,
34:39
which recently announced that they're going to diversify
34:41
their sourcing of chips away from
34:44
China that are basically trying
34:47
to steer out of China but are
34:49
being forced to from these political controls.
34:52
The conclusion was you need to create a
34:54
blacklist for the good companies
34:57
that are helping China China's tech
34:59
indigenization goals and
35:01
punish those that are not.
35:04
Now in practice, we know that's really
35:06
hard to do in this economic, in
35:08
this geopolitical climate where
35:10
China needs all the help it can get. It
35:13
can't afford to alienate foreign
35:15
investors in a big way, but there may be some easy
35:17
targets along the way as it's
35:19
tempted to retaliate.
35:21
Maybe on that note, because if
35:23
we look at the recent, not so
35:25
recent any water and also for Volkswagen
35:27
to have a joint venture with Horizon Robotics
35:30
for
35:31
I think two and a half billion US
35:33
dollars.
35:34
I mean that
35:35
certainly should put a smile on China's
35:38
face because it's in a critical
35:40
sector, it's by one of the largest
35:43
foreign companies. It's I think
35:45
also for Volkswagen one of the largest investments.
35:47
But if we combine that what you
35:50
just said before, what does it mean
35:52
for Volkswagen's competitiveness
35:54
in autonomous
35:56
driving down the road when
35:59
the China market, Volkswagen most
36:02
likely does not have the luxury anymore
36:04
to access generative AI. How
36:07
risky is it for international companies
36:10
to
36:11
agree to join ventures with Chinese
36:13
companies
36:14
in the area of semiconductors? Because
36:17
it seems like there are so many headaches
36:19
involved in this,
36:20
not just on the side of
36:22
export controls, but also especially if we
36:25
take into account,
36:26
can I still be competitive if
36:29
access to the Chinese market means
36:32
I have to separate my business
36:35
in the stuff
36:36
that's a Western stuff that still has access
36:38
to competitive generative
36:40
AI
36:41
and the China stuff where
36:44
I am out of the generative AI
36:46
ecosystem because I simply cannot I
36:48
cannot access it anymore so I feel
36:50
like for yeah
36:51
I'm just wondering what's the future both
36:54
for this particular sectors or autonomous
36:56
driving in China, does it actually stand a chance?
37:00
Even on ignoring the generative
37:02
AI part, non-generative
37:04
AI, China is ahead and envisioned in audio
37:07
networks by a large amount. And
37:09
if we talk about self-driving, it's most likely
37:12
not a generative AI model that's running it.
37:14
There is talk about certain types of transformers,
37:16
but for the most part, those types of models
37:18
are not that one. So maybe we're
37:20
going down this this
37:21
one path, path A, where we're way
37:24
better at, we as in the Western world
37:26
is better at generative AI, but the other path, self-driving,
37:28
does that also get impacted? Well, as
37:31
you mentioned on the Verizon robotics
37:33
chips, the current generation Nvidia
37:36
chip, shipping in Mercedes soon, all the 24 model
37:38
year, Mercedes
37:40
are gonna have this current generation Nvidia chip
37:42
for self-driving, the next generation one
37:44
is actually, I believe it's past
37:46
the threshold
37:47
of the export restrictions,
37:49
right? So we're talking about 2020, 2022-ish type timeframe, training
37:54
chips, but this restriction, if it doesn't
37:56
get moved forward, actually starts impacting
37:58
their self-driving right as soon as I'm assuming Nvidia's right
38:01
on the amount of compute and all that needed
38:03
to actually run self-driving.
38:07
I presume others are going to be down the same path.
38:09
And if that's the case, then 2025 year
38:11
technology is actually going to be, maybe
38:13
call it 2026 to get implemented in a car,
38:16
is restricted. And so that's another, I
38:18
think there's what, 2 million people who drive for a living in
38:21
the US. We're going to have those people
38:23
unlocked, right? Is China going to be able to unlock
38:25
those people for other productive tasks?
38:28
Yeah, this is the heart of it. This is the heart
38:30
of the October sanctions, unquestionably.
38:33
Were those sanctions put in place to limit China's
38:35
military capabilities,
38:37
as the document says, or is the intent
38:39
to cripple China's manufacturing capacity
38:42
and their whole broader economy? Because
38:47
autonomous driving and automotive is just
38:49
one, is the easiest of
38:51
those to grasp. But if you look at just
38:53
broadly, deploying AI in industrial
38:56
systems and cars and planes
38:58
and everything, that's a big, big
39:01
part of any modern economy. Right. Autos
39:03
are clearly very important to China. As
39:05
you take away the ability for
39:07
them to access leading semiconductor
39:09
manufacturing processes, that's
39:11
the end of autonomous driving in China,
39:14
right? Which is the end of China's automotive
39:17
industry. And so then it's you're talking about
39:19
a much, much broader scope of impact
39:21
for what went down in October. And
39:23
I
39:24
certainly think everybody in China assumes
39:26
it's if that's the case. It's not just about
39:29
holding back China's military. It's about holding
39:31
back China full stop.
39:33
Think about how insanely frustrating
39:36
that is for Beijing too, right? Like China
39:38
is the first mover and NEVs,
39:41
right? And battery technologies
39:43
and it's intent was to go and
39:46
deploy this to the world. Right. But
39:48
look at the IRA in the
39:50
U S and the extremely stringent provisions
39:53
that that imposes, arguably unrealistic,
39:55
but treasury will get creative with it. and
39:57
undergoing this transition. Now you've
39:59
got...
39:59
got these ICT rules getting
40:02
way harsher on top of the October 7
40:05
controls. We haven't even talked about biotech
40:07
too. That's another area where
40:09
we're going to see a lot more restrictions. So
40:12
yeah, this is an active
40:14
degradation in many ways of
40:16
China's development, its tech innovation, and
40:19
there is no easy way to transcend
40:21
those constraints. Something's got to give here.
40:24
We're two years away from election year. And
40:26
regardless of the ambiguity that
40:29
maybe in these potentially unintended second
40:31
order effects that were in the October 7th things, I think
40:34
a Republican president come 2025 is going
40:37
to look at these types of issues
40:39
in a much more aggressive way,
40:41
which is going to be something that Beijing and
40:44
Chinese companies are really going to have to grapple
40:46
with.
40:47
Can I be the European here for one second?
40:50
Because, I mean,
40:51
yes, it's nice to look
40:53
at everything from the lens of long
40:55
term competition with an adversary.
40:58
But escalating that would also make
41:00
retaliation from China much more likely
41:03
and much more severe.
41:04
Plus, forget all hopes about
41:06
engaging with China where you actually
41:08
need it.
41:09
You as in we need it for climate
41:12
change.
41:13
Because it's not just a loss, like
41:15
having no access to,
41:17
let's say, the most competitive AI
41:20
or ICT
41:21
doesn't just mean you lose productivity,
41:24
it also means you lose
41:26
a chance to have a
41:28
greener economy. So I
41:30
think in the, not in the next five
41:32
years, not in the next two elections, but
41:35
in the long term down the road, I'm
41:37
wondering if with this focus
41:40
on competition and looking
41:42
at every piece of technology through
41:44
an economic security lens, we
41:47
are sacrificing making a real
41:49
way in something that actually matters,
41:51
which is climate change and global warming.
41:53
It's interesting, right? is
41:56
dealing still with a war
41:59
on the continent.
41:59
with Russia, it sees some
42:02
level of engagement with China as
42:04
a way to prevent
42:07
a worst case scenario. But
42:09
you could argue those same arguments were
42:12
made around Nord Stream,
42:13
right? And that didn't work out so well.
42:15
But Europe is seeing the US
42:18
escalating at a very rapid clip.
42:21
So naturally, it's going to want to tap the brakes
42:23
on it and use economic engagement as a
42:25
means of doing so. But But then that in
42:27
turn compels the US to reach for
42:29
longer armed provisions, to try to steer
42:32
its partners in place. We've already seen
42:34
that on logic chip controls, but
42:36
really interesting to see where the partners are drawing the line,
42:39
to say, no, no, no, memory chips, you're going
42:41
too far there. And
42:43
Jan-Peter
42:43
and I are working on something
42:45
now, which will be out
42:47
soon, but we are looking at areas
42:51
in more mature tech nodes and
42:53
process nodes for where China can
42:55
compete, where European companies will
42:57
still be engaged with
42:59
China in the IC space. So that's going
43:01
to be an interesting point of divergence
43:04
potentially with the US.
43:06
That's a good point. And I want to go back to something
43:08
JP was talking about before, which was Volkswagen.
43:12
I think it's an interesting case study, because I don't
43:14
know if you all remember this, but there was a report
43:16
for a while that Volkswagen
43:19
was looking to actually buy the automotive
43:22
business out of Huawei. There was a rumor going
43:24
around that they wanted to do that.
43:25
And then it just it quietly went away. And
43:28
I always wondered like, what happened
43:30
there? Because in the end, I think Volkswagen
43:32
just hired a bunch of people out of Huawei. And now they're
43:34
looking to do this deal with Horizon Robotics. Did
43:37
somebody come to approach Volkswagen
43:39
and say, hey, that's a really bad idea. Let's stay away
43:42
from Huawei. Because I think
43:43
it speaks to the larger question of, you
43:45
know, we're reaching a point where there will be signs of of
43:47
light in between divergence between
43:50
what the US and the EU want. Yeah,
43:53
the action reaction here is really interesting
43:55
because from Beijing's perspective and
43:57
needs to steer its tech champions toward.
43:59
its big priorities, which means
44:02
they're involved in a lot of stuff from
44:05
very benign consumer applications
44:07
to legitimate military
44:10
civil fusion dual use type things
44:13
where the US is going to want to target.
44:15
And you see in the US how Huawei
44:17
has been cast as a bad actor,
44:20
and you're seeing now
44:22
even tightening restrictions on existing
44:25
controls with a
44:27
policy debate underway on whether to revoke
44:29
existing licenses to Huawei for,
44:32
for example, sub 5G components. So yeah,
44:35
that makes it hard to
44:38
engage with entities like that when
44:40
the US is going this
44:41
hard and against many
44:43
of them. One assumption
44:46
I think we've had in our conversation,
44:48
which isn't necessarily in the law yet,
44:50
is
44:50
the idea that Western
44:53
companies that have really fancy models
44:56
can't just sell access in China. So
44:58
right now, open AI, it's just
45:00
like their company policy. But
45:03
to my understanding, there's no law
45:05
that says Facebook or Google
45:08
or Microsoft can't just turn this
45:10
stuff on within China. So the
45:12
question of export controls
45:14
around these really powerful
45:17
models is something that I'm
45:19
really interested in. And if and
45:21
when they come on, to
45:23
what extent you can industrial
45:26
espionage your way past those
45:28
sorts of controls. All these are are
45:31
weights. And that's like a file.
45:33
That's like a few dozen gigs,
45:35
it seems to me to be. And deploying
45:38
it across an economy of a billion
45:40
and a half people is a whole other
45:43
ballgame, which is when you start getting into the chips,
45:45
the sort of like chip control angle
45:47
that we've been talking about.
45:49
But as these sorts of technologies
45:51
become even more critical
45:53
and strategic and sort of key your economic future
45:56
is the industrial espionage strategy
45:58
likely going to be.
45:59
and how effective could it be
46:02
to get some of the gains that you'd
46:05
want to have, even if the Western firms
46:07
just won't sell to you?
46:09
But I mean,
46:10
the path dependency kicks in, right?
46:12
Because if you want to deploy
46:15
with a cutting edge large
46:17
language model, you still need
46:19
high performance
46:21
AI accelerators to do anything
46:23
with it. So
46:25
in that regard, the current
46:27
controls that are in place would still
46:29
mean that whatever Google
46:32
or Facebook or any other hyperscaler
46:35
want to do in China, it would still
46:37
be increasingly
46:39
far behind what they can do outside
46:42
of China simply because they have only
46:44
access to 2020 AI accelerator technology.
46:49
But it doesn't have to be done in China. Baidu has, I
46:52
think, 2000 employees in Silicon Valley. So
46:56
they could easily run all of that
46:58
on AWS San Jose.
47:01
They don't have to have a deployment in
47:03
Beijing.
47:04
But this is where those ICTS
47:07
rules are going to be so interesting to
47:09
watch, because if there is any nexus,
47:12
you can be on a US cloud platform.
47:15
But if there's a nexus to to
47:18
Chinese entity of concern that the US
47:20
wants to target. And there's more
47:22
protectionism around data sets.
47:25
For example, as we start to see this
47:27
evolve over time, that's gonna be really
47:29
interesting to watch.
47:30
I wanna pick up something
47:33
that you mentioned before, Jordan, I think, because
47:35
from your take,
47:37
would it be viable for, if we think,
47:39
for example, of more neutral players in
47:41
Asia? So let's pick Singapore,
47:44
right? more
47:46
compared to Japan, a more
47:48
ambivalent and kind of serving
47:50
both sides attitude between
47:53
US and China.
47:54
There were rumors that
47:56
some of the
47:58
semiconductor supplier markets...
47:59
specifically EDA software
48:02
and equipment suppliers, move some
48:04
of their production to Singapore, specifically
48:07
for the reason to serve the Chinese
48:09
market. Could that be also viable
48:11
for cloud and AI where
48:14
we have suddenly huge service centers
48:17
in Singapore for Western hyperscalers
48:19
to serve the
48:21
Chinese market?
48:22
Or is that too much of a stretch?
48:25
Well, I mean, the Chinese government,
48:28
I think would have to get over their data
48:30
security concerns.
48:32
There's been
48:35
enormous blowups, which have taken
48:37
huge chunks out of the market
48:40
cap of some of China's leading players
48:42
because they've been accused of playing
48:44
fast and loose with sending Chinese
48:46
citizens data abroad and sort of exposing
48:49
the PRC to national
48:51
security risk cover, much that like, you know, makes
48:53
sense in a DD context. This is clearly something
48:56
that people with power and influence
48:59
in the system have been able to get their way
49:01
on in policy debates in
49:03
Beijing over the past few years. So
49:05
to be comfortable with the Singapore
49:07
solution, that is a
49:10
big hurdle to cross
49:13
if you're betting your
49:16
path to future productivity
49:18
growth over the next 10 years on
49:20
sending everything out to
49:22
Singapore and working with American
49:26
companies on their cloud servers.
49:28
I mean, that's like a pretty aggressive trust fall,
49:30
which, you know, maybe come GPT-5
49:33
once America's growing at 7% because
49:36
of all this crazy AI, the calculus
49:38
starts to change. But I think you
49:40
probably need to get a little further
49:43
in the,
49:44
not just perceived, but like actual gap
49:47
for those sorts of calculations to
49:49
really start shifting, to allow Chinese firms to
49:51
access the sort of, you know, to put
49:53
themselves in a situation which like absolutely
49:55
would not be kosher today.
49:58
Well, there's a lot of ground to cover. but it's
50:01
a good round of clever. Okay.
50:03
Thanks so much for being a part of chat talk, everyone. Thank
50:06
you. Thanks for having me. Always a pleasure.
50:09
Thank you. Same
50:40
sugar that I'm food
50:44
Same sugar that I'm
50:46
food Only
50:49
the new sugar that
50:51
I'm food High
50:55
is high, low is low
50:58
Everybody wanna go to heaven
51:01
But nobody wants
51:03
to die Nobody
51:07
wants to die
51:10
Nobody wants to
51:13
do the dance Don't
51:16
the dance of color outside the
51:19
lines And nobody
51:21
wants to cry Nobody
51:25
wants to cry Nobody
51:31
wants to push you forward Don't
51:35
the same truth as I'm moving Don't
51:40
the same truth as I'm
51:42
moving Don't
51:44
the same truth as I'm
51:48
moving I
51:51
can feel it all over me right now
52:00
Girl, it felt so good Got
52:03
me on screen locked down
52:06
Yo, I can't shut
52:09
up It's taking me and
52:11
spinning me I can't shut
52:13
up It's making me wanna
52:15
be With someone
52:19
Tonight Make
52:22
her feel your
52:26
eyes Shoo-Bah
52:29
boy you fool, yeah I'm gonna
52:31
send you back Said I'm fool
52:35
I'm gonna send you back Said I'm fool
52:39
Yeah, I'm gonna send you
52:41
back Shoo-Bah boy Up
52:46
the street, round the corner, down the block Round
52:48
the clock, it's time to stop Our
52:50
easy lease, I completed this My brain release,
52:53
pain relief Paraleaf is
52:55
the same belief that make it seem the chief And
52:57
insane as schliff, so obtain a spliff Cellophane
53:00
of reef, trade your beef for these
53:02
magic beans Crying to stalk, ride
53:05
the clouds, surf the waves Unearth
53:07
the caves, expose your soul, never
53:09
close your soul Snow your wall, never
53:11
overload your soul Hold your goals, never
53:13
lose sight of the host I suppose, we can
53:16
all float away as it goes We
53:18
can all know the way, if we pose We
53:20
can all shoot away, no one knows We
53:22
can all go today, if we chose, we
53:25
can all roll away Until
53:28
you're truly ready to say fuck
53:30
your fear You are
53:33
not a ally You
53:38
have to settle and be stuck right
53:40
here You cannot
53:43
be glad
53:48
I'm gonna ride the roller
53:50
coaster Give
53:52
all the troubles of the wild music
53:56
at least come see ya Come see
53:58
the artist I
54:00
say, this may be the way you want your
54:03
ride a bit, oh Shoo
54:05
bop, shoo bop, boo Shoo
54:10
bop, shoo bop, boo You
Podchaser is the ultimate destination for podcast data, search, and discovery. Learn More