Deep Dive
Shownotes Transcript
Capital Client Group, Inc.
When you're making financial decisions for a company, growth is great until manual work slows you down. Enter Intuit Enterprise Suite. This next-level solution integrates and streamlines all business management tools, improving effectiveness while cutting costs and overhead so your business keeps growing. Learn more at Intuit.com slash enterprise. Bloomberg Audio Studios. Podcasts. Radio. News.
Welcome to the Bloomberg Daybreak Asia podcast. I'm Doug Krisner. On today's episode, we'll be taking a deep dive into DeepSeek. This is a Chinese artificial intelligence startup, and its AI model known as R1 appears to rival the performance of U.S. products like ChatGPT. Perhaps its biggest selling point is that the model was developed at a fraction of the cost of competing products.
And that could make high valuations of some U.S. AI-related companies a little tough to justify. Let me give you one example. Today, in U.S. trading, shares in the AI chip leader NVIDIA were down nearly 17%. And that led to a loss in the NASDAQ 100 of 3%. In a moment, we'll be speaking with Jamie Cox. He is managing partner at Harris Financial Group. But let's begin in Hong Kong with Bloomberg Tech editor Vlad Savov.
It's always a pleasure to chat with you, Vlad. This feels like a major inflection point in the development of AI models. First, let me get your reaction and what you're hearing among Hong Kongers now that follow the tech space.
Well, the interesting thing about DeepSeek is it's a company that has rocketed to global renown and recognition today and over the past week, but it has been known about ever since its beginning in 2023. It's one that our Bloomberg Intelligence colleagues have been taking a close look at because it is developing such a hot area of new software as AI. It has been developing its own models. So DeepSeek
People here have been aware of it. It has been on the radar. But really, the thing that has happened is it hit a certain threshold of recognition. Part of it really was Silicon Valley people, venture capitalists. They took an interest. They tried it out and they found out that it is indeed, like you mentioned, as competitive as the company itself says.
It has posted some benchmark results on its website. It has claimed already that it competes with OpenAI, but frankly, every AI model developer claims that they compete or they can rival OpenAI. The amazing thing and the really, really impressive thing is that the people who know this stuff, which I can't claim that I'm an expert in it,
they all recognize it as a breakthrough. So we need to recognize that too. And it's also worth mentioning here that the company has put out a research paper of its own, and they've said that the costs of training this model are less than $6 million. Now, that excludes a whole bunch of costs, so we can't be exactly sure about how much more affordable it is for
this small company to develop it, but it is very clearly a difference maker when it comes to our assessment of how much it costs to develop a world-leading AI model. I think to be fair, our analyst here in New York, Mandeep Singh, pointed out that R1 was developed using output from OpenAI
Anthropic and Llama to train the models. So it's almost as though AI is training AI a little bit. And I think that what may have precipitated or at least engendered a lot of the selling that we saw in the US in the last session was the rating on the Apple App Store. I think R1 rose to the top level.
Can we agree though that the current Western approach to AI, I'm talking here about the reliance on those high-end chips, extensive computing power and vast amounts of electricity, is that being challenged to some extent? Yes, we can agree that it's challenged. We cannot agree that it's a foregone conclusion that it's going to be going away.
One thing to bear in mind here, which I think is very important, is that DeepSeq's model has delivered an impressive chatbot. It only works with text. So what we have is a breakthrough and an impressive achievement, but it's within one medium. One of the things that you have to bear in mind is that the more advanced chips, the ones that are making NVIDIA still a multi-trillion dollar company, even with its massive share price drop,
The thing that drives it is that you want those chips in order to get to the next stage of AI, specifically generative AI. So at the moment, we're talking about generating text, but the next evolution is going to be generating images, generating video, is stuff that open AI and
other services have already made major advances in. And that's where DeepSeek is probably going to need a lot more hardware than is used just for tech. So it may be too much to say that DeepSeek has been able to engineer a way around the export bans that were put in place during the Biden administration, the bans that basically kept China from accessing some of those advanced semiconductors manufactured by NVIDIA. We can't really articulate that, can we?
Yes, this is correct. Now, there are two things to say here. Firstly, those bans and those trade curbs, everyone in the industry said this is only going to encourage China and Chinese companies to focus on more efficient methods. So this is exactly what has been predicted by the industry. They said China will focus on more light, more small scale models, and DeepSeek is the fruit of those efforts.
That being said, we cannot say that the sanctions are not effective because DeepSeek's CEO and founder has himself said the thing that DeepSeek requires right now is not more money, is not more investment, it's actually access to those most advanced chips. The embargo, as he calls it, the ban on accessing those NVIDIA chips is the thing that's holding DeepSeek back.
Are we likely to see some type of fragmentation within the artificial intelligence industry, which is to say that you have certain products that have been produced at low cost being rolled out at lower prices, and maybe those products will see greater adoption, and some of the more sophisticated elements of artificial intelligence you'll have to pay a premium for, and that segment of the market will be controlled by virtualization.
very few companies? Yes, I can very much see that. One of the things that has been the trend over the past few months is actually OpenAI's revenue generation has accelerated. And it's one of the trends that we're seeing. And if you had asked me to predict what will be the big trend in AI in 2025, it is exactly this, taking all that investment, taking all that developed technology and turning it into products that you can sell subscriptions or in other ways entice people to spend money on them.
That is still going to continue to be the case. The premium products OpenAI sells a $200 per month pro product.
There is an audience for that. But that being said, the deep seek approach, it releases a software as an open source software. So you as an individual can take that and piggyback on top of it and decide that you want to build your own specific chatbot. And a good analogy that I have for this is when you think about banking, you have the traditional banks, the ones with branches and all those overheads and costs.
And then you have the Digibanks, the ones that are mobile only, app only. And because of that efficiency that those other banks have, they can offer you better rates and better offers. So to go back to the hardware software issue, it seems like Chinese AI engineers have found a way to work with the software.
given the limitations of the hardware that they have been forced to use because of those US export controls. But I'm wondering whether on the semiconductor industry side of things in China, that we may see some sort of breakthrough when it comes to advanced AI chips in the near future. Is that a possibility? Yes, there's two things to say here. Firstly, from what little we know about DeepSeek, and bear in mind, it's a very, very young startup. It came out of an AI hedge fund, AI-driven hedge fund.
So we still don't know that much about it, but from interviews with the founder, he has said that most of his engineers are graduates from China's universities. They are entirely domestic Chinese engineers. So credit to the company's educational system. Now on the semiconductor front, there are technical limitations and challenges. The classic one is EUV machines that only ASML makes and is prevented from selling to China.
Solving that issue, we haven't seen much evidence of China doing it. But then again, it may be something along the lines of deep seek. We never saw deep seek. We were only aware that it's building a model. We never saw it doing a breakthrough until today, until it was actually ready to come out there. Vlad, we'll leave it there. Thank you so much. Vlad Savov, Bloomberg Tech Editor in Asia, joining us from Hong Kong here on the Daybreak Asia podcast.
Capital Ideas, conversations with Mike Gitlin from Capital Group, features our top investment professionals sharing what drives them in today's market. Get stories and actionable insights. Subscribe wherever you get your podcasts. Capital Client Group, Inc. This show is sponsored by BetterHelp. BetterHelp has been revolutionary in connecting people to mental health services. Using BetterHelp can be as easy as opening your laptop or your phone and clicking a button, and the session begins.
Clients are able to choose in what way they would like to communicate with me, whether video or on the phone or chat texting. BetterHelp is there when you need it, and that's what makes all the difference. Visit betterhelp.com slash podbusiness to get 10% off your first month. Therapists were compensated.
Welcome back to the Daybreak Asia podcast. I'm Doug Krisner. We continue with our deep dive on DeepSeek. News of the Chinese startup's new chatbot R1 sparked a sell-off in many AI-related names globally, and it wiped out nearly a trillion dollars in market cap from U.S. and European tech names.
So now a key question is on CapEx spending plans for some of America's biggest companies. Joining me now for a closer look is Jamie Cox. He is managing partner at Harris Financial Group. Jamie, thank you so much for being with us. Certainly was an interesting day where AI was concerned. How do you make sense of the market reaction to this deep seek story? I think the market's reaction was great.
quite positive. I mean, there were only eight Dow stocks that were down today. 60% of the S&P 500 companies were up. So this was very much an isolated day for chip stocks. This was sort of the moment in time where you see disruption take hold. And when you start to see things like
AI disruption and the possibility that maybe companies are over-investing in AI infrastructure, you have to sit back and think a minute and ask yourself if NVIDIA can legitimately sell $30,000 chips or not. So I think it made a lot of investors think, but a lot of investors have been worried about overvaluation in these stocks anyway. So I think that the deep-seek
situation sort of precipitated was the precipitating event for things that may have happened anyway. So I think that's one thing to consider. The second is, I actually think this is a good thing. It's very important that we not get complacent in our technology prowess in the United States. We need to realize that these technologies can be created at scale all over the world. And what
deep seek basically did was kick off what i believe will be an arms race maybe a sputnik moment for the united states where we get really serious and the and the prospect of competition from around the world will actually propel ai development forward so i'm actually very positive both on the the development of ai and actually in in
Ironically, even though the stocks really took a beating today, I actually think that it further illustrates the need to build out AI infrastructure because what we're seeing now is disruption in basically AI 1.0, which will bring about the possibility of AI 2.0, which will make it universally usable and much more widely adopted than what we see right now. So I'm pretty excited about what we see. This is just a moment in time in the market, and we'll be looking back on this as one of the best
best buying opportunities in these stocks in quite some time. So one of the things that I found very compelling here is that DeepSeek's chatbot, known as R1, was apparently trained using less than cutting-edge semiconductors, and that, in turn, required much less energy
energy consumption. So I hear what you're saying when it comes to the sell-off that we had in certain semiconductor names today, but I was also struck by the fact that Constellation Energy was down 20%. How does the DeepSeek story change the conversation around power?
So there's two parts to that question. I'll address the power one first. There have been reports of inability for companies to get adequate power sources to data centers. So companies like Microsoft, Apple,
Amazon are putting in personal nuclear reactors to actually power data centers. And companies like Constellation Energy provide those services. So if the data centers are not going to be as plentiful and therefore the power sources will not be needed, that's why you see these knock-on effects. It's one of the reasons why the power story had been one of the sort of
to play the AI story without actually buying the tech stocks. So that's why they're sort of connected together in a trade. I don't believe that either story will turn out to be, the sell-off today will be the right one. I think it's actually the wrong choice. But to go back to talk about the clever technology or the clever engineering that made this possible, where the R1,
learning module is able to use just a very small change, but it's very, very impactful. So if you think about like disruptive technologies, you have incumbent players who basically iterate. They have built this fantastic
and they just iterate small in small pieces. Whereas these disruptive players come around and they actually change the entire game. And that's what DeepSeek has done. They basically took the AI engine, which was ever present at OpenAI, where the technology is ready for any question on any topic.
Whereas in the deep seek infrastructure, it would basically take the input and only turn on the pieces of the infrastructure that were needed to answer the question or to iterate. So it basically didn't have the entire system running
all the time. And that will be readily adopted by all AI providers. For sure, Beta and for sure, OpenAI will adopt that clever engineering and they too will have it. And so the reason why I think that this is just the beginning is that now that infrastructure that was being built to handle an ever present AI infrastructure now could be used for other things.
And I think that's why we're going to see this AI infrastructure actually blossom as a result of this disruption not be torn apart. It is a critical week for tech earnings this week. Four of the MAG-7 are reporting. That's Apple, Microsoft, Tesla, and Meta Platforms. What are your expectations here? And what do these companies need to say about the future in order for investors to really kind of not lose heart?
Well, two of the four of them had a very good day today. I mean, Apple and Meta really bucked the trend, and it was largely because they're consumers of these more expensive chips. And I think if anything that lowers the infrastructure cost accrues directly to
to players like Meta or Apple. Apple's a unique story because there have been some concerns about the inability to produce or sell the iPhone. Maybe sales had declined, maybe people weren't upgrading at the same rate, but the story in Apple is less hardware these days and definitely more about the ancillary services. So I think Apple's going to do quite well and I think
I mean, the stock definitely had a fantastic day today on the prospects of them having higher margins because they're spending less on their data infrastructure. Meta, the same thing. I think Microsoft kind of got caught up in the AI given its investment in the open AI platform. But I think that Microsoft is actually of the three, the best platform.
positioned to continue to grow earnings year over year. And so I think three of the four are doing well. I don't follow Tesla that much, but I feel like Elon Musk may be in a better position to understand whether or not this deep seek technology is actually, you know, whether or not the technology being deployed or the
the things that the company has said that they've been able to do are actual reality because Tesla has one of the largest AI infrastructures powered by all the automobiles that Tesla makes. So their neural net
has been around for some time and has been generating AI content for generation of software with the cars. So I think maybe we might learn a thing or two in the conference call that may be related to DeepSeek and some of this AI disruption. I think that will be the story for the Tesla conference call this year. Jamie, before I let you go, we also have a Fed decision on Wednesday.
Powell News Conference right after we get the decision. Is it likely that we're going to get kind of a hawkish message right now? No, I think the Fed is on hold. They may discuss, as they did in the press conference in December, the need to be cautious about
lowering rates too fast. But it's going to be interesting to watch over the next couple of months because the president is going to put enormous pressure on the Fed to do something about mortgage rates. That has been a particular concern to President Trump, and he is going to start beating the drum on that very loudly. So I have a feeling we'll start to see the Fed start to talk about how they're going to deal with the balance sheet.
and how they could help consumers on the mortgage side. So I don't think this Fed meeting will be much on the Fed funds rate in terms of news. I mean, the Fed fund futures basically are 99% that they hold. What is not
built into the model is what they talk about on the balance sheet. And there has been runoff at a fairly decent clip over the last year and a half. And I think that may be ready to slow down. And if the Fed would slow its mortgage-backed securities runoff, I think that might help mortgage rates some. So I think that's where you're going to see some of the action. And it would not surprise me at all if the president is out
you know, with comments in and around the Fed meeting that would, you know, really encourage them to be more aggressive in helping mortgage rates decline. Jamie, we'll leave it there. Thank you so much for being with us. Jamie Cox there, managing partner at Harris Financial Group, joining from Richmond, Virginia, here on the Daybreak Asia podcast.
Thanks for listening to today's episode of the Bloomberg Daybreak Asia Edition podcast. Each weekday, we look at the stories shaping markets, finance, and geopolitics in the Asia Pacific. You can find us on Apple, Spotify, the Bloomberg Podcast YouTube channel, or anywhere else you listen. Join us again tomorrow for insight on the market moves from Hong Kong to Singapore and Australia. I'm Doug Krisner, and this is Bloomberg.
World-class journalists, global leaders, influential thinkers, cutting-edge data. January 20th to 23rd in Davos, Conversations at Bloomberg House will provide the context you need on the biggest stories that will shape the year ahead.
From AI and the future of tech to geopolitics, markets, sustainability, inequality, and more. Join us in person or watch live. Visit BloombergLive.com slash BloombergHouseDavos to learn more.