November 5, 2025 • 27 mins
China's AI rollout diverges sharply from the US, favoring open-source models and broad adoption over high-cost infrastructure. China's top tech firms including Tencent, Baidu and Alibaba will invest less than 10% of the $370 billion that US hyperscalers plan to spend on AI capex this year. This cost-conscious approach aligns with Beijing's AI Plus strategy, aiming to embed AI across all sectors by 2027 to boost productivity.
Who will win the AI race? And are we in a bubble in AI infrastructure spending? Robert Lea, senior tech analyst at Bloomberg Intelligence, joins John to discuss why China's broader economy, rather than its tech companies, may be the main beneficiary of the nation's AI rollout.
See omnystudio.com/listener for privacy information.
Mark as Played
Transcript
Episode Transcript
Available transcripts are automatically generated. Complete accuracy is not guaranteed.
Speaker 1 (00:00):
The world's two largest economies, the United States and China,
are taking a radically different path in their AI rollout,
and the stakes couldn't be higher. US technology companies are
simply on a spending spree, pouring trillions into AI infrastructure
and chasing technological breakthroughs like artificial general intelligence. Meanwhile, China's
(00:22):
playing a different game, prioritizing broad adoption across industries to
boost self sufficiency and leaning into open source models that
come at a fraction of the cost of their Western counterparts.
So who's going to win the AI race? Are we
seeing signs of an AI bubble in spending and valuations?
And what does this all mean for global economic power dynamics.
(00:45):
You're listening to Asia Centric from Bloomberg Intelligence. I'm John
Lee in Hong Kong. Joining us to unpack all these
questions is Robert Lee, senior tech analyst at Bloomberg Intelligence
covering the China AI sector.
Speaker 2 (00:58):
Robert, Welcome, Thanks very much, John, pleasure to be with
you again.
Speaker 1 (01:03):
How is China's rollout of AI different from that of
the US?
Speaker 2 (01:08):
I think, as you said in your intro, China is
much more focused on really a pragmatic application of AI
into readily available, readily addressable real world solutions, and tying
in with that is their focus on free to access,
open source models to help proliferate the technology across every
(01:30):
single sector, across every geography of the economy.
Speaker 1 (01:35):
And why do you think the US companies in particular
at spending so much money on this infrastructure rollout? According
to some estimates, I think Jensen Huang was suggesting that
there could be over three to four trillion dollars in
spending over the rest of this decade. And that's amazing
because that's almost equivalent of ten percent of the US GDP.
Speaker 2 (01:56):
Yes, no, it's a phenomenal number, isn't it. I mean,
if we just pull some numbers from the Bloomberg terminal,
looking at twenty twenty five alone, the big four US
hyperscalers so that's Amazon, Meta, Microsoft, and Alphabet are due
to spend about three hundred and seventy billion US dollars
on CAPEX this year. Those are consensus numbers, so that's
based on the estimates of the top thirty forty to
(02:18):
fifty leading investment banks in comparison the big three technology platforms.
So it's not quite an apples for Apples comparison, they're
spending less than one tenth of that amount, around thirty
billion dollars. And the much lower spend of the Chinese
companies really reflects their focus on lower cost, less computationally
(02:40):
intensive models, which partly results from these earlier export controls
we saw in video chips, because the result of those
was really to force Chinese companies to focus more on
software optimization and clever cost down methods to reduce the
overall capex cost and overall preating costs of their models.
(03:01):
But you know, based on our research, that's arguably delivered
China a significant advantage in low cost model development, which
puts them in a great space going forward to proliferate
their technology both nationwide within their economy and also potentially
tap into more lucrative export markets, particularly in emerging economies.
Speaker 1 (03:24):
Now, would you classify America's approach is more of one
sort of brute strength to spending as much money as
possible to get the best technological performance.
Speaker 2 (03:36):
I think it's true of both to some degree. But
obviously the one key and obvious advantage American AI companies
have over their Chinese counterparts is access to leading edge
AI accelerator technology i e. From the likes of Nvidia
and with the export controls we've seen their Chinese companies
just don't have that, so again they have to work
(03:57):
within the confines of their world currently. But yeah, absolutely,
technology leadership is something that American companies are clearly gunning for,
as well as AGI as it's referred to or artificial
general intelligence plus potentially superintelligence at a later date. Having
said that, leading Chinese companies also have that in their sites,
(04:20):
particularly like Sa Valid Barbar, which will frequently talk about
AGI as being one of its prime objectives for ramping
up x capex this year.
Speaker 1 (04:28):
Asia Centric is produced by Bloomberg Intelligence. We're more than
five hundred experienced analysts and strategists work around the clock
to bring you timely, world class research. Our coverage spans
two hundred market industries, currencies, commodities, and industries, as well
as over two thousand equities and credits. To learn more
about Bloomberg Intelligence, visit BI go on the Bloomberg terminal.
(04:51):
If you like what you're hear, don't forget to subscribe
and share. So American companies are looking or trying to
achieve AGI or art official general intelligence, like for the
lame and what exactly is that?
Speaker 2 (05:04):
I'm not sure there's any dictionary definition of it. But
in crude terms, it's a level of AI that has
a human like or matches human like intelligence. But again,
what exactly is that? So for fans of science fiction,
if you look at something like Blade Runner or back
to two thousand and one AD, we're talking of models
which aren't just intelligent, but maybe have some sentience, have
(05:28):
some degree of consciousness. There are many ways to define it,
but what really defines the models that we see and
understand at the moment, And I think the key technique,
if you like, the key thing that underpins these models
is pattern recognition. Pattern recognition is very powerful and useful technique.
I mean, when we see someone familiar, familiar face in
the street, that's pattern recognition.
Speaker 1 (05:51):
But there are so.
Speaker 2 (05:51):
Many other aspects of human intelligence that the current generation
of models just can't practically replicate. You know, AI is
a very powerful technique that can generate productivity and efficiency
savings and maybe cost savings in certain applications, and even
the potentially high level applications like drug discovery or whatever.
(06:15):
There are potential usage cases there, but again, in no
way does it replicate human intelligence. As we know, it's
very poor at analytical thinking. For example, even though there
are models with supposed reasoning capabilities at the moment, in
no way does that replicate human reasoning in any way.
(06:35):
I'll give you an example. There's been some experimentation done
looking at some of these very high level models which
can solve or allegedly solve high end maths problems. But
when there was looked into in more detail, they're really
just replicating the maths problems that they have been trained
on within their training data. Again, this is an example
(06:57):
of pattern recognition in the way they're doing it blindly,
they're doing at parrot fashion. They're a mimic of what
they have been trained on. They lack any real semantic
understanding of the problem. And I think that's really what
underpins and defines generative AI as we understand it at
the moment. It's very good at understanding the language, the
(07:20):
semantics of language, the relationships between words. If I had
an apple here in my hand in the moment in
English language, there are certain words that will be associated
with it. You know, you could describe an apple as green,
as red, as crispy sweet, whatever it may be. There
are certain words that are associated with the word apple
and other words like the word rectangular or electronic that aren't.
(07:43):
So when these models are trained on potentially billions and
billions of pages of data, they can quite easily work
out the syntax and the relationships between different words, but
they have no real understanding of the semantics the meaning
of those under alliging words. They're not conscious in any way.
They have no understanding of the real world of where
(08:06):
they are, are, feelings, all these other additional aspects that
really define what it is to be a human being.
And don't just take my word for it, look at
the academic research that is there at the moment.
Speaker 1 (08:17):
So, Robert, when do you think that we would reach
this supposed AGI level.
Speaker 2 (08:24):
Well, I just don't think it can be achieved with
the current generation of models. If you look at the models,
and without exaggerating, there is a new model delivered or
developed literally every other day at the moment, not just
in China, but on a global basis. And if you
look at those models, they're very very specialized in one
individual task or one individual application, whether that be coding,
(08:47):
whether it be image generation, whatever it may be, you know,
and maybe they're pretty good at that individual task. But
in order to generate an AGI system, you need to
be good at everything. You need to be capable of
delivering on any question on antitasks put to you. The
current generation of AI models are nowhere near that. So
(09:09):
what we really need is a major technological breakthrough. We
really need a new generation of models, which we don't
have foresight on at the moment. As and when that
will be achieved, nobody really knows. Could it be in
six months, could it be in one month, Could it
be in three years, could it be never. It's incredibly
difficult to predict. So really, what I'm saying is it's
(09:32):
going to take a major technological breakthrough that may or
may not happen in the future. But what we do
know is the current generation of models can't achieve that.
Let me just quote to you the results of a
survey that was put together by the Association for the
Advancement of Artificial Intelligence. They did a survey of just
under five hundred members back in March of this year.
(09:55):
I should point out that the majority of the respondents,
in fact, over two thirds, we're academics. These academics who
have got no skin in the game. You know, they're
not trying to sell your model or a business. They're
not trying to raise money, but they know this technology
inside out. The majority of respondents, or in fact, seventy
six percent, asserted that the current approach of scaling up
(10:15):
AI is not going to yield AGI. It was basically
unlikely to be achieved on a near term timescale. And
this is the approach that the likes of open Ai
and the Chinese firms are taking at the moment. They're
taking existing generation and models, they're trying to make them
ever bigger. They're taking a scale up approach. So based
on the academic literature that's there, based even on the
(10:38):
likes of someone called Roger Penrose, Roger Penrose is a
Nobel Prize winner in physics, you know, he's of this view.
And I think there is a significant critical mass of
academics out there are of the view as I am,
that the current generation of models cannot achieve AGI. So really,
I think at worst people have got a false hope
(10:59):
of what can be achieved in terms of AI realization
on any reasonable near term timeframe. It's going to take
a major technological breakthrough to achieve this. If and when
that happens, we don't know.
Speaker 1 (11:12):
And why do you think that Chinese approach is much
more pragmatic than say the US approach.
Speaker 2 (11:18):
I think again it's a consequence of the confines they
have to work within because of the in video bands.
So whilst you know Chinese developers and engineers are of
equal competence to the US colleagues, if they don't have
access to leading edge and video chips, it's very hard
for them to compete at the leading edge. And I
think this sort of ties in with one of the
(11:39):
major policies of the Chinese government. This is a strategy
called AI Plus. This was launched in twenty twenty four
and essentially the government is trying to proliferate the application
of AI throughout the economy again, every sector, every geography
of the economy by twenty twenty seven, in order to
generate term cost and efficiency savings. So I think ultimately
(12:03):
to help boost economic productivity throughout the economy. So this
ties in with Chinese economic agenda with their economic objectives.
You know, faced with the challenge as many countries do,
with an aging population, trying to rebalance the economy away
from these previous years of being very heavily reliant on
(12:24):
fixed asset investment, turning to the high tech sector as
a driver, as a government would describe it, of new
productive forces. So the application of AI throughout the economy
in a very pragmatic sense very much is part and
parcel of that overriding government strategy.
Speaker 1 (12:43):
Okay, And also if you look at the Chinese tech companies,
they're not monetizing AI as much as the US counterparts.
Can you go through that?
Speaker 2 (12:51):
That's true, but again go back to what I said
on capex, Chinese companies are spending approximately one tenth in
capex of the US PUS, so whilst their level of
monetization may not be as high, nor are they spending
as much. So again not picking on open Ai, but
there was some very clever analysis done based on Microsoft's
latest quarterly reporting. Because Microsoft will equity account they're holding
(13:18):
in open ai, and within the notes of the accounts,
the loss that was attributable to them was disclosed and
the equity accounting and then you can take that and
gross it up and approximately work out what open ai
is losing per quarter at the moment. So based on
that analysis, per quarter, open ai is losing somewhere between
(13:39):
ten and eleven billion US dollars. Now, as you probably know,
and again there's a lot of data out there in
a moment, the consensus view is that open ai is
going to generate around twelve billion of revenue this year.
So that's substantially ahead of their Chinese counterparts. But as
I said, the level of capex investment, the level of
loss that results from that substantially higher. Now the lower
(14:03):
level of monetization within China, even though it is in
terms of population size obviously of a substantially larger place,
I think it reflects two things. One is the open
source approach within China, that most of these tools are
available for free, free to access.
Speaker 1 (14:19):
Like Deepsek's open source.
Speaker 2 (14:20):
Yeah exactly, and I've got deep Seak on my phone.
I pay nothing for it in order to try and
seed the market and to grow user bases. But the
problem is when every company is employing the same strategy,
it's very hard to differentiate yourself from the competition. And
just to throw another number at you, at the last count,
there were more than five hundred large language models in China.
(14:44):
Given that China was coming from behind. You know, in
approximately March twenty twenty three, when Chat, GPT and open
Ai first started to make the headlines in a meaningful way,
there were no large language models that we were aware
of in China. So they've narrowed the gap rapidly and significantly.
But now the market is flooded with supply and therefore
(15:06):
you're getting a low level of monetization. But I think
there are very clear indications that the large language model
sector in China at least is commoditized already because the
level of product differentiation between the different models that are there,
although some are a little bit better than the other,
the overall level of differentiation is not overly significant and
(15:29):
therefore it's very very hard for companies to gain any
sustainable competitive edge. And if we go back to basic
you know, the ABC of business school and the attributes
you would want to see in any startup company or
any early stage company, having a sustainable competitive edge is
absolutely key to your future growth prospects and your future
(15:51):
profit outlook. So when there are more than five hundred
models that are pretty much similar to each other, it's
very very difficult. So really we need to see a
solidation in the Chinese sector in order to help put
the sector in a better profit footing. But then the
other challenge that the Chinese companies have is the lack
(16:12):
of so called killer apps. There's been a lot of
promises made on AI, particularly from the tech brows in
the US, that AI can potentially cure cancer, that you know,
it's going to be really revolutionary, and whilst it's obviously
a useful productivity and efficiency tool, I think it's a
(16:32):
very very long way away from fulfilling any of those promises,
let alone AGI. And this is the problem. So if
you look at the AI apps that are there at
the moment, there are no pure play AI apps in
the China market. What we're seeing is really AI being
used to augment or improve the existing apps that are
out there, and by and large, it's not having a
(16:56):
significant impact on the revenue generation in the sector. I'll
give you an example. If you look at the total
monetization in the Chinese AI chatbots sector over the last
twelve months, and I'm going to quote you a number
on the iOS system, so it doesn't include Android, but
my point is going to be made anyway. It was
less than one million US dollars. One million US dollar
(17:19):
revenue return in a sector of about twenty or thirty
chatbots is an extremely low level of monetization. So monetization
what it's all about at the moment. I think the
real winners of this are not so much to private
sector companies. It's the Chinese economy as a whole, again
tying into the government strategy of proliferating AI within the economy.
(17:43):
It's the productivity gains. It's the Chinese economy. It's the
Chinese consumers. It's small to medium sized Chinese businesses which
will ultimately be the winners of China's overall investment in AI,
not its private sector companies. The private sector companies are
effectively subsidizing this national rollout.
Speaker 1 (18:03):
Well, that's quite interesting because in the US, customers, you know,
both enterprise and individuals are still willing to pay for
the chet GPT service. They're paying twenty to one hundred
dollars a month depending on which level. And you just
mentioned Open AI is going to make what thirteen billion
dollars this year. I think Anthropic is going to make
a few billion dollars as well. Could you say that
(18:24):
the Chinese models don't perform as well as the American ones.
Speaker 2 (18:27):
Well, you might think that we'll come back to that,
but as I said, sure, the level of monetization that
open ai is substantially higher by an order of magnitude
or several orders of magnitude versus the entire Chinese AI sector.
These are hard numbers that is not up for debate.
But as I said, the level of KAPEX investment in
(18:48):
R and D investment in America is also substantially higher.
And as I quoted to you, open Ai, based on
this calculation, is losing ten or eleven billion US dollars
per quarter. So whilst it may be annualizing a run
rate of around twelve billion per year, it's still losing
a substantial level of investment and that is coming at
(19:11):
a huge financial cost. And also what we don't know
is what open ai is capitalizing on its balance sheet
at the moment. So in cash terms, the US companies
are burning through substantial amounts even if their monetization is
at a more advanced stage. In other words, in pure
financial terms, they're still burning through a ginormous level of
(19:36):
cash and generating a substantial level of loss. So therefore
I think, you know, China spending less money and they're
focused more on a pragmatic near term application of AI
proliferation of AI in terms of real world applications, I
think that on balance is more likely to generate a
payback in the near term than the US approach.
Speaker 1 (19:58):
So could you make an rument that the American companies
are better at monetizing AI, but for the wider economy,
China's approach could actually be better in the long run
because it's cheaper, open source, and it's proliferating across the
economy at much quicker scale and speed.
Speaker 2 (20:17):
Yes, but again the higher monetization in the US reflects
their greater focus on close source, the fact that most
of them are being charged for whereas out of choice,
the Chinese have not gone for that. So I suppose
if the Chinese gone down the same business model and
adopted closed source, undoubtedly we would see a higher level
of monetization in the moment. But I think, you know,
(20:39):
relative to the much lower levels of R and D
and CAPEX expenditure in China. And also we shouldn't forget
with the substantial level of state support that the Chinese
companies get at the moment. I mean there's a national
data center network as well, so there is you know,
a substantial level of subsidization in there as well, which.
Speaker 1 (20:56):
Is positive for the wider economy absolutely.
Speaker 2 (20:59):
So again I come back to the central point. I
think it's the Chinese economy as a whole which will
be the main beneficiary of its national AI strategy at
the moment. And come back to the performance of the models,
if I may, I'm just going to quote to you
some numbers from a US benchmarking service. This is a
company called live bench so it's actually based out of
the US where they look at the performance metrics across
(21:23):
a range of measures, whether it's coding, language, mathematics, across
all the world's leading large language models. Now, unsurprisingly, US
models are leading the pack. The top ranked Chinese model
is deep seek, which based on the third of November,
was ranked at number thirteen globally. So you know, point
number one, the US models are leading. But if you
(21:45):
look at their overall scores, Open AI or chut GPT
five's high model is scoring on its overall score about
ten percent higher, only ten percent higher than deep Seek.
So it's a better model. It's doing better, but it's
not an order of magnitude better. So I think if
you take everything into consideration, looking at the much lower spender,
(22:06):
the Chinese companies are incurring the much lower level of
loss capital investment and then more pragmatic focus on trying
to drive the application, the real word application of AI
in their economy. The net net I would say they
stand a higher chance of generating economic returns in the
near term than their uspers do.
Speaker 1 (22:29):
Okay, And when do you think these Chinese companies like
deep Seek and you know some of the other ones
from Tensen or by Doo, when do you think they'll
start making money?
Speaker 2 (22:40):
So, based on a thembers in a moment, we don't
foresee any of them making sustainable profits on at least
the next three years. And I think for many of them,
by doing particular, it's very hard to see that they'll
ever achieve that. In all honesty, so state the obvious.
The companies are best place to monetize and to at
least try and make some return. In terms of the
private sector companies in the moment on the infrastructure side,
(23:01):
which would be the huaweis and the s MI cs.
Now Huawei is a private company, so we don't have
great visibility on the numbers, similar to what's going on
in the US, the server companies in video, et cetera.
But if you take a company like deep Seek, I
mean it's really i would say, set up on altruistic motives,
charitable motives. It's there to support that again, the proliferation
(23:26):
of AI across the economy. It's not a profit seeking company.
There's obviously a substantial level of investment going into them.
They're sitting on a big cashhile profit isn't their primary motive.
They're set up effectively on a charitable basis. They're there
to do it for the good of the country, and
therefore I think it's probably at this point highly likely
that a company like deep seek would ever come to
(23:47):
the market. But then if you look at the peers
ten to out and Ali Barbar, I think Ali Baba
its main business is really on the cloud computing side.
They are monetizing it. There is some additional revenue coming in,
but again factoring in the cutthroat competitive dynamics in China's
cloud sector, which are very very different to the US.
(24:08):
It's a far more fragmented market. The level of margin
and cash generation coming through in that business is not
big enough to move the needle on an e commerce
goodliath like Ali Barber. And in fact, the very very
high CAPEX requirements, even though they're substantially lower than the uspers,
are still putting a lot of pressure on Ali Barbar's
(24:30):
free cash flow. So in pure cash terms, Ali Barbar's
cloud business is loss making at this point in time.
And I think to end on a positive on this,
the company that is really best place to at least
monetize and to gain some economic return on its AA
investments is really ten Cent because rather than focus on
(24:51):
subsidizing the state role out of AI, which is really
I would argue what Ali Barbara is doing, they're more
focused on the internal application applying AI to to for example,
within the games business, in terms of accelerating the games development,
in terms of improving the ad placement. I think the
results are already there if you look at the last
few quarters numbers, the margin enhancement that's bringing, I think
(25:15):
that's more likely to generate a return great.
Speaker 1 (25:19):
And I wanted to also ask a bigger, broader question.
Now I know this is this is a bit of
a loaded question, but who's going to win the AI race.
And what does this even mean?
Speaker 2 (25:30):
Well, exactly what does it mean? The world? We're in
a bifurcated system. You know, China and America are already
in many respects across their economies operate in separate and
different worlds, whether that's cultural, economically. The words of globalization
at this point are behind us, aren't they And that
will only continue going forward. And at the end of
(25:51):
the day, just because someone's model is bigger or better,
what does that really mean? I mean again, I've quoted
you the numbers based on a US benchmarking service. American
models are roughly ten percent better. But it comes at
a substantial cost. But so what. At the end of
the day, any investment needs to generate a return. You
(26:11):
need a return on investment unless it's done on purely
charitable basis. And as I say, I think China's lower cost,
more pragmatic strategy is far more likely to generate a
near term return or drive productivity savings within its economy.
Given you know, ultimately China is you know, the government
(26:34):
has a hand in everything. There is a very high
level of state direction here and state oversight as enshrined
by the AI plus strategy. So I think Chinese investors,
maybe not the private sector ones, but I think based
on that strategy, they're more likely to generate a return
than the US pers.
Speaker 1 (26:55):
Robert's been an intriguing conversation. Thanks for joining.
Speaker 2 (26:58):
That's great, John, thanks very much, pleasure to be on.
Speaker 1 (27:01):
You've been listening to Age Eccentric from Bloomberg Intelligence. I'm
John Lead, Hong Kong. You can listen to all our
episodes on Apple Podcasts, Spotify or review Listen and this
podcast was produced and edited by Clara Chen. Thanks for listening.
The world's two largest economies, the United States and China,
are taking a radically different path in their AI rollout,
and the stakes couldn't be higher. US technology companies are
simply on a spending spree, pouring trillions into AI infrastructure
and chasing technological breakthroughs like artificial general intelligence. Meanwhile, China's
(00:22):
playing a different game, prioritizing broad adoption across industries to
boost self sufficiency and leaning into open source models that
come at a fraction of the cost of their Western counterparts.
So who's going to win the AI race? Are we
seeing signs of an AI bubble in spending and valuations?
And what does this all mean for global economic power dynamics.
(00:45):
You're listening to Asia Centric from Bloomberg Intelligence. I'm John
Lee in Hong Kong. Joining us to unpack all these
questions is Robert Lee, senior tech analyst at Bloomberg Intelligence
covering the China AI sector.
Speaker 2 (00:58):
Robert, Welcome, Thanks very much, John, pleasure to be with
you again.
Speaker 1 (01:03):
How is China's rollout of AI different from that of
the US?
Speaker 2 (01:08):
I think, as you said in your intro, China is
much more focused on really a pragmatic application of AI
into readily available, readily addressable real world solutions, and tying
in with that is their focus on free to access,
open source models to help proliferate the technology across every
(01:30):
single sector, across every geography of the economy.
Speaker 1 (01:35):
And why do you think the US companies in particular
at spending so much money on this infrastructure rollout? According
to some estimates, I think Jensen Huang was suggesting that
there could be over three to four trillion dollars in
spending over the rest of this decade. And that's amazing
because that's almost equivalent of ten percent of the US GDP.
Speaker 2 (01:56):
Yes, no, it's a phenomenal number, isn't it. I mean,
if we just pull some numbers from the Bloomberg terminal,
looking at twenty twenty five alone, the big four US
hyperscalers so that's Amazon, Meta, Microsoft, and Alphabet are due
to spend about three hundred and seventy billion US dollars
on CAPEX this year. Those are consensus numbers, so that's
based on the estimates of the top thirty forty to
(02:18):
fifty leading investment banks in comparison the big three technology platforms.
So it's not quite an apples for Apples comparison, they're
spending less than one tenth of that amount, around thirty
billion dollars. And the much lower spend of the Chinese
companies really reflects their focus on lower cost, less computationally
(02:40):
intensive models, which partly results from these earlier export controls
we saw in video chips, because the result of those
was really to force Chinese companies to focus more on
software optimization and clever cost down methods to reduce the
overall capex cost and overall preating costs of their models.
(03:01):
But you know, based on our research, that's arguably delivered
China a significant advantage in low cost model development, which
puts them in a great space going forward to proliferate
their technology both nationwide within their economy and also potentially
tap into more lucrative export markets, particularly in emerging economies.
Speaker 1 (03:24):
Now, would you classify America's approach is more of one
sort of brute strength to spending as much money as
possible to get the best technological performance.
Speaker 2 (03:36):
I think it's true of both to some degree. But
obviously the one key and obvious advantage American AI companies
have over their Chinese counterparts is access to leading edge
AI accelerator technology i e. From the likes of Nvidia
and with the export controls we've seen their Chinese companies
just don't have that, so again they have to work
(03:57):
within the confines of their world currently. But yeah, absolutely,
technology leadership is something that American companies are clearly gunning for,
as well as AGI as it's referred to or artificial
general intelligence plus potentially superintelligence at a later date. Having
said that, leading Chinese companies also have that in their sites,
(04:20):
particularly like Sa Valid Barbar, which will frequently talk about
AGI as being one of its prime objectives for ramping
up x capex this year.
Speaker 1 (04:28):
Asia Centric is produced by Bloomberg Intelligence. We're more than
five hundred experienced analysts and strategists work around the clock
to bring you timely, world class research. Our coverage spans
two hundred market industries, currencies, commodities, and industries, as well
as over two thousand equities and credits. To learn more
about Bloomberg Intelligence, visit BI go on the Bloomberg terminal.
(04:51):
If you like what you're hear, don't forget to subscribe
and share. So American companies are looking or trying to
achieve AGI or art official general intelligence, like for the
lame and what exactly is that?
Speaker 2 (05:04):
I'm not sure there's any dictionary definition of it. But
in crude terms, it's a level of AI that has
a human like or matches human like intelligence. But again,
what exactly is that? So for fans of science fiction,
if you look at something like Blade Runner or back
to two thousand and one AD, we're talking of models
which aren't just intelligent, but maybe have some sentience, have
(05:28):
some degree of consciousness. There are many ways to define it,
but what really defines the models that we see and
understand at the moment, And I think the key technique,
if you like, the key thing that underpins these models
is pattern recognition. Pattern recognition is very powerful and useful technique.
I mean, when we see someone familiar, familiar face in
the street, that's pattern recognition.
Speaker 1 (05:51):
But there are so.
Speaker 2 (05:51):
Many other aspects of human intelligence that the current generation
of models just can't practically replicate. You know, AI is
a very powerful technique that can generate productivity and efficiency
savings and maybe cost savings in certain applications, and even
the potentially high level applications like drug discovery or whatever.
(06:15):
There are potential usage cases there, but again, in no
way does it replicate human intelligence. As we know, it's
very poor at analytical thinking. For example, even though there
are models with supposed reasoning capabilities at the moment, in
no way does that replicate human reasoning in any way.
(06:35):
I'll give you an example. There's been some experimentation done
looking at some of these very high level models which
can solve or allegedly solve high end maths problems. But
when there was looked into in more detail, they're really
just replicating the maths problems that they have been trained
on within their training data. Again, this is an example
(06:57):
of pattern recognition in the way they're doing it blindly,
they're doing at parrot fashion. They're a mimic of what
they have been trained on. They lack any real semantic
understanding of the problem. And I think that's really what
underpins and defines generative AI as we understand it at
the moment. It's very good at understanding the language, the
(07:20):
semantics of language, the relationships between words. If I had
an apple here in my hand in the moment in
English language, there are certain words that will be associated
with it. You know, you could describe an apple as green,
as red, as crispy sweet, whatever it may be. There
are certain words that are associated with the word apple
and other words like the word rectangular or electronic that aren't.
(07:43):
So when these models are trained on potentially billions and
billions of pages of data, they can quite easily work
out the syntax and the relationships between different words, but
they have no real understanding of the semantics the meaning
of those under alliging words. They're not conscious in any way.
They have no understanding of the real world of where
(08:06):
they are, are, feelings, all these other additional aspects that
really define what it is to be a human being.
And don't just take my word for it, look at
the academic research that is there at the moment.
Speaker 1 (08:17):
So, Robert, when do you think that we would reach
this supposed AGI level.
Speaker 2 (08:24):
Well, I just don't think it can be achieved with
the current generation of models. If you look at the models,
and without exaggerating, there is a new model delivered or
developed literally every other day at the moment, not just
in China, but on a global basis. And if you
look at those models, they're very very specialized in one
individual task or one individual application, whether that be coding,
(08:47):
whether it be image generation, whatever it may be, you know,
and maybe they're pretty good at that individual task. But
in order to generate an AGI system, you need to
be good at everything. You need to be capable of
delivering on any question on antitasks put to you. The
current generation of AI models are nowhere near that. So
(09:09):
what we really need is a major technological breakthrough. We
really need a new generation of models, which we don't
have foresight on at the moment. As and when that
will be achieved, nobody really knows. Could it be in
six months, could it be in one month, Could it
be in three years, could it be never. It's incredibly
difficult to predict. So really, what I'm saying is it's
(09:32):
going to take a major technological breakthrough that may or
may not happen in the future. But what we do
know is the current generation of models can't achieve that.
Let me just quote to you the results of a
survey that was put together by the Association for the
Advancement of Artificial Intelligence. They did a survey of just
under five hundred members back in March of this year.
(09:55):
I should point out that the majority of the respondents,
in fact, over two thirds, we're academics. These academics who
have got no skin in the game. You know, they're
not trying to sell your model or a business. They're
not trying to raise money, but they know this technology
inside out. The majority of respondents, or in fact, seventy
six percent, asserted that the current approach of scaling up
(10:15):
AI is not going to yield AGI. It was basically
unlikely to be achieved on a near term timescale. And
this is the approach that the likes of open Ai
and the Chinese firms are taking at the moment. They're
taking existing generation and models, they're trying to make them
ever bigger. They're taking a scale up approach. So based
on the academic literature that's there, based even on the
(10:38):
likes of someone called Roger Penrose, Roger Penrose is a
Nobel Prize winner in physics, you know, he's of this view.
And I think there is a significant critical mass of
academics out there are of the view as I am,
that the current generation of models cannot achieve AGI. So really,
I think at worst people have got a false hope
(10:59):
of what can be achieved in terms of AI realization
on any reasonable near term timeframe. It's going to take
a major technological breakthrough to achieve this. If and when
that happens, we don't know.
Speaker 1 (11:12):
And why do you think that Chinese approach is much
more pragmatic than say the US approach.
Speaker 2 (11:18):
I think again it's a consequence of the confines they
have to work within because of the in video bands.
So whilst you know Chinese developers and engineers are of
equal competence to the US colleagues, if they don't have
access to leading edge and video chips, it's very hard
for them to compete at the leading edge. And I
think this sort of ties in with one of the
(11:39):
major policies of the Chinese government. This is a strategy
called AI Plus. This was launched in twenty twenty four
and essentially the government is trying to proliferate the application
of AI throughout the economy again, every sector, every geography
of the economy by twenty twenty seven, in order to
generate term cost and efficiency savings. So I think ultimately
(12:03):
to help boost economic productivity throughout the economy. So this
ties in with Chinese economic agenda with their economic objectives.
You know, faced with the challenge as many countries do,
with an aging population, trying to rebalance the economy away
from these previous years of being very heavily reliant on
(12:24):
fixed asset investment, turning to the high tech sector as
a driver, as a government would describe it, of new
productive forces. So the application of AI throughout the economy
in a very pragmatic sense very much is part and
parcel of that overriding government strategy.
Speaker 1 (12:43):
Okay, And also if you look at the Chinese tech companies,
they're not monetizing AI as much as the US counterparts.
Can you go through that?
Speaker 2 (12:51):
That's true, but again go back to what I said
on capex, Chinese companies are spending approximately one tenth in
capex of the US PUS, so whilst their level of
monetization may not be as high, nor are they spending
as much. So again not picking on open Ai, but
there was some very clever analysis done based on Microsoft's
latest quarterly reporting. Because Microsoft will equity account they're holding
(13:18):
in open ai, and within the notes of the accounts,
the loss that was attributable to them was disclosed and
the equity accounting and then you can take that and
gross it up and approximately work out what open ai
is losing per quarter at the moment. So based on
that analysis, per quarter, open ai is losing somewhere between
(13:39):
ten and eleven billion US dollars. Now, as you probably know,
and again there's a lot of data out there in
a moment, the consensus view is that open ai is
going to generate around twelve billion of revenue this year.
So that's substantially ahead of their Chinese counterparts. But as
I said, the level of capex investment, the level of
loss that results from that substantially higher. Now the lower
(14:03):
level of monetization within China, even though it is in
terms of population size obviously of a substantially larger place,
I think it reflects two things. One is the open
source approach within China, that most of these tools are
available for free, free to access.
Speaker 1 (14:19):
Like Deepsek's open source.
Speaker 2 (14:20):
Yeah exactly, and I've got deep Seak on my phone.
I pay nothing for it in order to try and
seed the market and to grow user bases. But the
problem is when every company is employing the same strategy,
it's very hard to differentiate yourself from the competition. And
just to throw another number at you, at the last count,
there were more than five hundred large language models in China.
(14:44):
Given that China was coming from behind. You know, in
approximately March twenty twenty three, when Chat, GPT and open
Ai first started to make the headlines in a meaningful way,
there were no large language models that we were aware
of in China. So they've narrowed the gap rapidly and significantly.
But now the market is flooded with supply and therefore
(15:06):
you're getting a low level of monetization. But I think
there are very clear indications that the large language model
sector in China at least is commoditized already because the
level of product differentiation between the different models that are there,
although some are a little bit better than the other,
the overall level of differentiation is not overly significant and
(15:29):
therefore it's very very hard for companies to gain any
sustainable competitive edge. And if we go back to basic
you know, the ABC of business school and the attributes
you would want to see in any startup company or
any early stage company, having a sustainable competitive edge is
absolutely key to your future growth prospects and your future
(15:51):
profit outlook. So when there are more than five hundred
models that are pretty much similar to each other, it's
very very difficult. So really we need to see a
solidation in the Chinese sector in order to help put
the sector in a better profit footing. But then the
other challenge that the Chinese companies have is the lack
(16:12):
of so called killer apps. There's been a lot of
promises made on AI, particularly from the tech brows in
the US, that AI can potentially cure cancer, that you know,
it's going to be really revolutionary, and whilst it's obviously
a useful productivity and efficiency tool, I think it's a
(16:32):
very very long way away from fulfilling any of those promises,
let alone AGI. And this is the problem. So if
you look at the AI apps that are there at
the moment, there are no pure play AI apps in
the China market. What we're seeing is really AI being
used to augment or improve the existing apps that are
out there, and by and large, it's not having a
(16:56):
significant impact on the revenue generation in the sector. I'll
give you an example. If you look at the total
monetization in the Chinese AI chatbots sector over the last
twelve months, and I'm going to quote you a number
on the iOS system, so it doesn't include Android, but
my point is going to be made anyway. It was
less than one million US dollars. One million US dollar
(17:19):
revenue return in a sector of about twenty or thirty
chatbots is an extremely low level of monetization. So monetization
what it's all about at the moment. I think the
real winners of this are not so much to private
sector companies. It's the Chinese economy as a whole, again
tying into the government strategy of proliferating AI within the economy.
(17:43):
It's the productivity gains. It's the Chinese economy. It's the
Chinese consumers. It's small to medium sized Chinese businesses which
will ultimately be the winners of China's overall investment in AI,
not its private sector companies. The private sector companies are
effectively subsidizing this national rollout.
Speaker 1 (18:03):
Well, that's quite interesting because in the US, customers, you know,
both enterprise and individuals are still willing to pay for
the chet GPT service. They're paying twenty to one hundred
dollars a month depending on which level. And you just
mentioned Open AI is going to make what thirteen billion
dollars this year. I think Anthropic is going to make
a few billion dollars as well. Could you say that
(18:24):
the Chinese models don't perform as well as the American ones.
Speaker 2 (18:27):
Well, you might think that we'll come back to that,
but as I said, sure, the level of monetization that
open ai is substantially higher by an order of magnitude
or several orders of magnitude versus the entire Chinese AI sector.
These are hard numbers that is not up for debate.
But as I said, the level of KAPEX investment in
(18:48):
R and D investment in America is also substantially higher.
And as I quoted to you, open Ai, based on
this calculation, is losing ten or eleven billion US dollars
per quarter. So whilst it may be annualizing a run
rate of around twelve billion per year, it's still losing
a substantial level of investment and that is coming at
(19:11):
a huge financial cost. And also what we don't know
is what open ai is capitalizing on its balance sheet
at the moment. So in cash terms, the US companies
are burning through substantial amounts even if their monetization is
at a more advanced stage. In other words, in pure
financial terms, they're still burning through a ginormous level of
(19:36):
cash and generating a substantial level of loss. So therefore
I think, you know, China spending less money and they're
focused more on a pragmatic near term application of AI
proliferation of AI in terms of real world applications, I
think that on balance is more likely to generate a
payback in the near term than the US approach.
Speaker 1 (19:58):
So could you make an rument that the American companies
are better at monetizing AI, but for the wider economy,
China's approach could actually be better in the long run
because it's cheaper, open source, and it's proliferating across the
economy at much quicker scale and speed.
Speaker 2 (20:17):
Yes, but again the higher monetization in the US reflects
their greater focus on close source, the fact that most
of them are being charged for whereas out of choice,
the Chinese have not gone for that. So I suppose
if the Chinese gone down the same business model and
adopted closed source, undoubtedly we would see a higher level
of monetization in the moment. But I think, you know,
(20:39):
relative to the much lower levels of R and D
and CAPEX expenditure in China. And also we shouldn't forget
with the substantial level of state support that the Chinese
companies get at the moment. I mean there's a national
data center network as well, so there is you know,
a substantial level of subsidization in there as well, which.
Speaker 1 (20:56):
Is positive for the wider economy absolutely.
Speaker 2 (20:59):
So again I come back to the central point. I
think it's the Chinese economy as a whole which will
be the main beneficiary of its national AI strategy at
the moment. And come back to the performance of the models,
if I may, I'm just going to quote to you
some numbers from a US benchmarking service. This is a
company called live bench so it's actually based out of
the US where they look at the performance metrics across
(21:23):
a range of measures, whether it's coding, language, mathematics, across
all the world's leading large language models. Now, unsurprisingly, US
models are leading the pack. The top ranked Chinese model
is deep seek, which based on the third of November,
was ranked at number thirteen globally. So you know, point
number one, the US models are leading. But if you
(21:45):
look at their overall scores, Open AI or chut GPT
five's high model is scoring on its overall score about
ten percent higher, only ten percent higher than deep Seek.
So it's a better model. It's doing better, but it's
not an order of magnitude better. So I think if
you take everything into consideration, looking at the much lower spender,
(22:06):
the Chinese companies are incurring the much lower level of
loss capital investment and then more pragmatic focus on trying
to drive the application, the real word application of AI
in their economy. The net net I would say they
stand a higher chance of generating economic returns in the
near term than their uspers do.
Speaker 1 (22:29):
Okay, And when do you think these Chinese companies like
deep Seek and you know some of the other ones
from Tensen or by Doo, when do you think they'll
start making money?
Speaker 2 (22:40):
So, based on a thembers in a moment, we don't
foresee any of them making sustainable profits on at least
the next three years. And I think for many of them,
by doing particular, it's very hard to see that they'll
ever achieve that. In all honesty, so state the obvious.
The companies are best place to monetize and to at
least try and make some return. In terms of the
private sector companies in the moment on the infrastructure side,
(23:01):
which would be the huaweis and the s MI cs.
Now Huawei is a private company, so we don't have
great visibility on the numbers, similar to what's going on
in the US, the server companies in video, et cetera.
But if you take a company like deep Seek, I
mean it's really i would say, set up on altruistic motives,
charitable motives. It's there to support that again, the proliferation
(23:26):
of AI across the economy. It's not a profit seeking company.
There's obviously a substantial level of investment going into them.
They're sitting on a big cashhile profit isn't their primary motive.
They're set up effectively on a charitable basis. They're there
to do it for the good of the country, and
therefore I think it's probably at this point highly likely
that a company like deep seek would ever come to
(23:47):
the market. But then if you look at the peers
ten to out and Ali Barbar, I think Ali Baba
its main business is really on the cloud computing side.
They are monetizing it. There is some additional revenue coming in,
but again factoring in the cutthroat competitive dynamics in China's
cloud sector, which are very very different to the US.
(24:08):
It's a far more fragmented market. The level of margin
and cash generation coming through in that business is not
big enough to move the needle on an e commerce
goodliath like Ali Barber. And in fact, the very very
high CAPEX requirements, even though they're substantially lower than the uspers,
are still putting a lot of pressure on Ali Barbar's
(24:30):
free cash flow. So in pure cash terms, Ali Barbar's
cloud business is loss making at this point in time.
And I think to end on a positive on this,
the company that is really best place to at least
monetize and to gain some economic return on its AA
investments is really ten Cent because rather than focus on
(24:51):
subsidizing the state role out of AI, which is really
I would argue what Ali Barbara is doing, they're more
focused on the internal application applying AI to to for example,
within the games business, in terms of accelerating the games development,
in terms of improving the ad placement. I think the
results are already there if you look at the last
few quarters numbers, the margin enhancement that's bringing, I think
(25:15):
that's more likely to generate a return great.
Speaker 1 (25:19):
And I wanted to also ask a bigger, broader question.
Now I know this is this is a bit of
a loaded question, but who's going to win the AI race.
And what does this even mean?
Speaker 2 (25:30):
Well, exactly what does it mean? The world? We're in
a bifurcated system. You know, China and America are already
in many respects across their economies operate in separate and
different worlds, whether that's cultural, economically. The words of globalization
at this point are behind us, aren't they And that
will only continue going forward. And at the end of
(25:51):
the day, just because someone's model is bigger or better,
what does that really mean? I mean again, I've quoted
you the numbers based on a US benchmarking service. American
models are roughly ten percent better. But it comes at
a substantial cost. But so what. At the end of
the day, any investment needs to generate a return. You
(26:11):
need a return on investment unless it's done on purely
charitable basis. And as I say, I think China's lower cost,
more pragmatic strategy is far more likely to generate a
near term return or drive productivity savings within its economy.
Given you know, ultimately China is you know, the government
(26:34):
has a hand in everything. There is a very high
level of state direction here and state oversight as enshrined
by the AI plus strategy. So I think Chinese investors,
maybe not the private sector ones, but I think based
on that strategy, they're more likely to generate a return
than the US pers.
Speaker 1 (26:55):
Robert's been an intriguing conversation. Thanks for joining.
Speaker 2 (26:58):
That's great, John, thanks very much, pleasure to be on.
Speaker 1 (27:01):
You've been listening to Age Eccentric from Bloomberg Intelligence. I'm
John Lead, Hong Kong. You can listen to all our
episodes on Apple Podcasts, Spotify or review Listen and this
podcast was produced and edited by Clara Chen. Thanks for listening.
Hosts And Creators
John Lee
Tom Corbett
Popular Podcasts
On Purpose with Jay Shetty
I’m Jay Shetty host of On Purpose the worlds #1 Mental Health podcast and I’m so grateful you found us. I started this podcast 5 years ago to invite you into conversations and workshops that are designed to help make you happier, healthier and more healed. I believe that when you (yes you) feel seen, heard and understood you’re able to deal with relationship struggles, work challenges and life’s ups and downs with more ease and grace. I interview experts, celebrities, thought leaders and athletes so that we can grow our mindset, build better habits and uncover a side of them we’ve never seen before. New episodes every Monday and Friday. Your support means the world to me and I don’t take it for granted — click the follow button and leave a review to help us spread the love with On Purpose. I can’t wait for you to listen to your first or 500th episode!
Ruthie's Table 4
For more than 30 years The River Cafe in London, has been the home-from-home of artists, architects, designers, actors, collectors, writers, activists, and politicians. Michael Caine, Glenn Close, JJ Abrams, Steve McQueen, Victoria and David Beckham, and Lily Allen, are just some of the people who love to call The River Cafe home. On River Cafe Table 4, Rogers sits down with her customers—who have become friends—to talk about food memories. Table 4 explores how food impacts every aspect of our lives. “Foods is politics, food is cultural, food is how you express love, food is about your heritage, it defines who you and who you want to be,” says Rogers. Each week, Rogers invites her guest to reminisce about family suppers and first dates, what they cook, how they eat when performing, the restaurants they choose, and what food they seek when they need comfort. And to punctuate each episode of Table 4, guests such as Ralph Fiennes, Emily Blunt, and Alfonso Cuarón, read their favourite recipe from one of the best-selling River Cafe cookbooks. Table 4 itself, is situated near The River Cafe’s open kitchen, close to the bright pink wood-fired oven and next to the glossy yellow pass, where Ruthie oversees the restaurant. You are invited to take a seat at this intimate table and join the conversation. For more information, recipes, and ingredients, go to https://shoptherivercafe.co.uk/ Web: https://rivercafe.co.uk/ Instagram: www.instagram.com/therivercafelondon/ Facebook: https://en-gb.facebook.com/therivercafelondon/ For more podcasts from iHeartRadio, visit the iheartradio app, apple podcasts, or wherever you listen to your favorite shows. Learn more about your ad-choices at https://www.iheartpodcastnetwork.com
The Joe Rogan Experience
The official podcast of comedian Joe Rogan.