This week’s podcast is about my visit to Baidu.

You can listen to this podcast here, which has the slides and graphics mentioned. Also available at iTunes and Google Podcasts.

Here is the link to the TechMoat Consulting.

Here is the link to our Tech Tours.

——–

Related articles:

• Baidu’s Great Leap Forward: How China’s Search Giant is Conquering GenAI (Tech Strategy)
• Huawei Is Building the New Data Architecture of Generative AI. Plus, Baidu’s New Agent Builder. (Tech Strategy – Podcast 205)
• A Strategy Breakdown of Arm Holdings (1 of 3) (Tech Strategy – Daily Article)

From the Concept Library, concepts for this article are:

• AI Cloud
• Auto

From the Company Library, companies for this article are:

• Baidu AI Cloud

———transcription below

00:05

Welcome, welcome everybody. My name is Jeff Towson and this is the Tech Strategy Podcast from Techmoat Consulting. And the topic for today, lessons, let’s call it updates, from my visit to Baidu headquarters in Beijing. And this was a week or so ago. I had a great time out there. I went to a couple events, a lot of discussion. you know, Baidu is on my short, very short list.

 

00:32

for companies to follow as closely as possible, really because of the AI cloud business.  I don’t really spend that much time thinking about their mobile app business, which is content and search and things like that. No, it’s their cloud business and the fact that they have been pretty much all in on cloud for over a decade, building a full stack AI.

 

00:58

you know, basically a full AI stack plus all the solutions.  They’re not building a couple models, they’re building all of them. They’re building apps, they’re building models, they’re doing semi-conductor, they’re doing the whole thing across the board.

 

01:11

and it’s their major strategic move for a long time and they also have a CEO who has real chops, mean real deep credibility in this particular topic. anyways, I’ve been following them closely so it was kind of a big opportunity to go out and visit and I’ll kind of tell you what I learned in my notes and yeah, things are moving. Things are moving pretty fast and the whole AI game, generative AI is what we’re talking about.

 

01:40

has really changed in the last, let’s call it 12 months.  Not that it wasn’t changing all the time anyways, but yeah, things have changed for Baidu in this regard.  So that will be the topic for today. Let me do my standard disclaimer. Nothing in my podcast, let me try that again.  Nothing in this podcast or my writing or website is investment advice. The numbers and information for me and any guess may be incorrect. The views and opinions expressed may no longer be relevant or accurate. Overall, investing is risky. This is not investment legal or tax advice. Do your own research.

 

02:10

One breath, okay, let’s get into the content. Okay, I don’t really have any concepts for today.  It’s kind of all over the place.  yeah, I mean, this is just the development, rapid advancement of AI cloud and what’s been changing.  know, Baidu is out sort of in the northwest of Beijing, sort of near Peking University, not far from my old office, actually.  A lot of the companies are up there. It’s kind of an interesting space. Didi’s up there.

 

02:41

Oppo was up there. It’s kind of an interesting part of town So anyways, this was sort of Monday

 

02:51

It was a couple great days in Beijing. mean, it was like Xiaomi, JD, Tencent, Baidu. I mean, it was just majors after majors. Super fun. So anyways, we were staying out in Chaoyang, which is, for those familiar, that’s sort of near Sanlitun, Gongti beilu, that area. And I can remember being there 17 years ago.

 

03:14

When like Saint-Luton in particular if you’re that’s a nice shopping center now, it’s really high-end. It’s very fashionable. Well, that used to be sort of a Let’s call it not reputable street

 

03:28

And I remember like, you know, if you were in the main street is called Gongti beilu, but near that is sort of the San Litun Street. If you saw someone you know like at night heading up that street, they were taking a little bit of a walk on the dark side. So now it’s turned into like, you know, this high-end shopping district, but I remember when it was a little bit of a sketchy street area and there was a famous bookstore there called Bookworm where all the, you know.

 

03:56

journalists and others used to hang out. Well, they’re all gone now. Now it’s all Louis Vuitton and high-end stores. So, I knew I was hanging out there. I jumped in the van, headed out to the Northwest, go up to the Baidu headquarters. First thing we did, which was pretty cool, they were having an event that day anyways. They were releasing a new product, which was called Miao Da, which is basically a new model. And, you know, it was basically a model for developers coding.

 

04:24

basically, generative AI coding model tool.  Okay, fine, mean people are releasing models every week now, it’s kind of hard to keep up.  Let alone with who’s the best in an even week.  So, okay, interesting, and sort of saw the presentations.  The part that actually got me, which I thought was cool, it was all pretty cool, but the part that I thought was different, let’s call it distinctive, was they basically came at this at the point of look,

 

04:52

If you’re coding, all the coding of the world is basically done in English. mean, code is, Java is written in English, you know?  But this is a second language for a lot of people, in particular for Chinese. So, Miao Da, it’s a coding tool, it’s a coding assistant, but you can speak in Chinese.

 

05:11

or you can write it, you can communicate with the app in Chinese, it will then immediately write code, but you don’t have to speak in English anymore if you’re a coder. If you’re more comfortable in Chinese, you can do it in Chinese and it’ll handle the translation over, which was clever.  I was like, okay, that’s pretty neat.  And I think that’s a good little indication of how Baidu is thinking.  They are domestic, right? They are focused on China.

 

05:40

Alibaba Cloud, to a lesser degree, Tencent Cloud, they are trying to be global.  Alibaba Cloud will always talk about where they are in hugging face and who’s using them to build their models and how much Quan is getting adopted.  Baidu is all about China, which is very similar to what they’ve done in search. The Baidu search engine is a China language search engine used in Chinese.

 

06:04

I’m sorry, in Chinese mainland.  So, you can see them going the same thing. They’re doing generative AI for China.  The other thing they’re doing, which I’ll talk about in a bit, is they’re super enterprise focused. That’s where there, you know, they’ve got a very clear strategic pathway, China and enterprise, and everything kind of fits in that.  The B2C story, not as much.

 

06:28

I’m not 100 % sure if that’s a strategic decision or just because maybe, and this is just me making this up, the political complexities of being B2C in China are more, right?  Enterprise is a little, let’s call that more predictable territory.

 

06:48

You’re working with companies’ hand in hand, you’re working on this together, but it’s not like 500, 800 million Chinese are typing in whatever and they’re getting responses on a second-by-second basis. Well, in a country that cares about information flows, curation, things like that, one terrain is easier than the other for sure. So, who knows, that’s a guess.  Okay, so first thing was sort of Miao Da, interesting, cool, bugged out of

 

07:18

pretty quick and wanted you know I’ll give you a little side story there so I was sitting across the table from a woman who was a CNBC reporter at the event as well and we got chatting and one factoid came out of that that was interesting was well how many how many reporters to CNBC have in China one I think there’s a news person as well a TV person but written

 

07:46

As far as I can tell, it’s one for all of China.  Like this is a big deal in my opinion that you go back 10, 15 years, there was a decent number of reporters from all of these major outlets in China. I used to hang out, you see Reuters and BBC and you know, all of them, LA Times, I used to hang out with them a little bit.  They’re all gone.  It is almost impossible to get a visa there now.  It is a very small number of people on the ground.  And I think the net result of that is

 

08:15

the quality of conversation and the quality of understanding in let’s say the US in particular is far less about China than it used to be. I think that creates issues.  We don’t know. Anyways, little factoid.  Okay, okay, let’s go on to the next topic. Now, February, a couple months ago,

 

08:37

There were two pretty interesting announcements, discussions by CEO Robin Lee.  The first one was at the World Government Forum. I think it’s called World Government Forum in Dubai.  Big leaders, they’ll show up and CEO Robin Lee went there and he had some pretty interesting comments, which I think we can see playing out now. One was,

 

09:00

that Baidu was going to be much more, for generative AI, this is all about gen AI, was going to be much more open source than was previously the plan. And that’s 100 % a response to DeepSeek, which he has said, well, not specifically this, but generally speaking, he talked a lot about DeepSeek. The argument is that open source seems to be what people want, especially enterprises, and as I said, they’re living in enterprise.

 

09:30

You know, the idea of running everything on the cloud, the idea of having a subscription to a closed source model, that doesn’t appear to be the way things are moving. It appears to be, this is so fundamental to your operation that you want to download and run these things locally. At a minimum, you want open source. Ideally, you want to run these things locally.

 

09:52

So, and he basically said the rate of innovation, the rate of development, the participation rate is faster there. So, we could call that the rate of innovation and adoption. I thought that was interesting. He talked about DeepSeek and basically talked about the cost. Now people have made these, there’s a lot of argument about how much did DeepSeek really cost? Was it 5 % of, know, GPT for…

 

10:20

Okay, I don’t know. I don’t really think that’s the key thing. The key thing is you can run DeepSeek and get very good results with dramatically less compute power being required. That is 100 % true. So, this idea of open source.

 

10:40

Models that you can have significant control over that cost a lot less in terms of compute power I mean remember last year like the discussion was all about who can build the biggest, you know cluster

 

10:54

AI builds, I’m sorry, X builds colossus.  It was all about whoever’s going to have the most horsepower, the most cluster is going to win.  Well, DeepSeek changed that discussion. So, I think you’re going to hear this come out from Robin later that open source and low cost are big parameters in terms of what they’re doing.  The other thing I thought he thought, he said that I thought was interesting that just, this was about DeepSeek as well.

 

11:21

that this is really unpredictable and big innovations and changes to the game can come from anywhere.  DeepSeek was a tiny little team.  In 18 months, their big model that shocked everyone a couple months ago, that wasn’t their first model, that was like their fourth model.  They had a visual model, they had some other models, but that was the one that got everyone’s real attention is the reasoning model. Okay.

 

11:49

So interesting.  He also talked to look; this is at the World Forum. He also talked a little bit about robotaxis, which by do has been kind of rocking and rolling.  There, you know, they’re autonomous vehicles. Apollo Go.  It’s getting deployed in China, five, six major cities.  Lots of them are running around.  There’s sixth generation Apollo Go.  Apollo Go is now on the road and it’s, I mean, it’s L four autonomous driving.

 

12:18

And he had a comment that basically autonomous driving is at least 10 times safer than humans. And he had the numbers to back that up. So that was the world forum. OK, interesting.  A week or two after they had the earnings call for fourth quarter 2024, which is really annual. So, end of year 2024 comments.  He had a couple interesting comments there too as well.

 

12:44

The one I think that got people’s attention was he basically said Baidu is now an AI first company. Traditionally they are an internet company. They are a suite of mobile apps of which searches their crown jewel.  He’s basically saying they’re an AI first company. They’re no longer an internet company. Okay, is that just terminology or is there, how deep does that go? I don’t know, but that was, it was a big announcement even though I’m not totally sure what it means in practice.

 

13:13

He talked about AI cloud growth in terms of revenue. Remember, well, maybe you don’t remember, about a year ago when Gen.AI was really just taking off, Baidu was really the first group to put up significant revenue growth from generative AI. Not only did they have the products deployed, but they also showed revenue growth early on. Well, AI cloud over the last 12 months grew 25%, roughly, in revenue.

 

13:42

So, it’s a real business, right?  He’s mentioned that Ernie, their main engine, their LLMs, it’s going to go open source. So, Ernie 4.5 is going to go open source and Ernie Bot, basically that’s like ChatGPT, it’s going to be free.

 

14:01

So, I think that feeds right into kind of what we here was just saying like open source, low cost, that appears to be a shift because a year ago he was not talking at all about making earning these things open source. In fact, you could see him leaning to proprietary and closed.  So big change there.  What else?  Basically, said they’re going to stay with their full AI stack approach. They’re going to focus on enterprise solutions.

 

14:31

And they didn’t say China, but I would say China in there as well. So, all kind of reaffirming what we’ve heard. Okay, back to the Baidu visit.  So, I sit down and my standard question is always, you know, what are the big strategic priorities? There’s always like, you know, tons and tons of projects, tons of initiatives, fine. What are the big ones? Like where is the money going? Where is the focus going? Where are the, what are the projects that the best people all want to be on, right?

 

14:58

And gives you a pretty good annotation of the direction of the company.  you know, the Robotaxi, it’s not really Robotaxi, it’s autonomous driving. mean, they’re basically providing software to OEMs for the most part.  Now they have their Apollo Go, but they’re also, you know, in the software business and providing it to companies. So not just Robotaxi, I don’t know why I keep saying that.  So that’s a big priority.  Cool.

 

15:29

AI Cloud, big surprise, that’s kind of where we’re talking about, Enterprise Focus, which I mentioned. Then sort of talked about Ernie Bot. Okay, that’s interesting.  mean, okay, AI Cloud is full cloud solution, compute, connectivity, storage, AI tools, models, all of that.  The full suite. But then when talk Ernie Bot, you’re really talking about apps and you’re talking about use cases. And that’s really interesting because…

 

15:57

That’s where you, you know, when I try to study what’s actually happening on the ground, if I’m looking at a company, can look at the rev, I can go deep into the company because they’re public.

 

16:08

Generative AI, we don’t really have any of that yet. I mean there aren’t any big public companies that have long track records that you can study.  So, I spend a lot of time looking at use cases and trying to figure out which one is going to matter for customers and which aren’t. That’s kind of my where the rubber hits the road is to use cases.  So that’s Ernie Bot.  And it appears they are putting Ernie Bot into everything.

 

16:33

Like they’re playing across the whole board. We’re doing the full AI tech stack.  We’re doing all the models, not just the big ones. We’re doing all of them.  And we’re doing all the apps. We’re doing all the use cases. So, all the industries, manufacturing, transportation, finance, health care.

 

16:54

I asked for some examples. talked about manufacturing is a big one because they’re enterprise focused. And if you’re enterprise focused on China, manufacturing is huge, but also financial services, their insurance.  Trip.com is using AI Cloud. So that was in her. Oh, not AI Cloud, but using some of their services.  So, you can sort of break this down by industry, but you can also break it down by sort of.

 

17:20

specific services like AI agents for customer service, things like that. They’re playing across the board as far as I can tell. And all this sort of circles a question which is, how do you win in this business? Like everything’s happening. You can’t predict too much so you have to play across the board. But you kind of got to know what winning looks like. And I’m more and more thinking, I’m thinking there’s four things you have got to have.

 

17:49

And I think two of them are probably decisive. This is a guess. Okay, so the four things.  Chips, LLMs, data, and customer relationship, know, significant value there. So, a year ago, everyone was talking about chips. Whoever had the most compute was the best. You know, a year ago…

 

18:12

Colossus, the Twitter X supercomputer, also an X-Men if you read comic books.  They were the best. They had the most GPUs. They had the most everything. They were big.  But that was what people were competing on. Then it’s like, well…

 

18:30

If you can distill the big models down into simpler models that can be then used easily like DeepSeek, know, distilling quantizing models, you can make them much smaller. it’s sort of like the analogy I’ve always heard is it’s like the big, big models, the teacher, but you can have a student be taught who knows most of what the teacher knows much cheaper and much simpler. like, okay, clauses as a teacher, but you DeepSeek would be the student.

 

18:58

So, there’s lots of ways. So okay, the chips thing is there. That’s number one.  The LLMs, the models, you know, again, there is at least this month a tendency that look, these things, people like to own their own models. They really do like to have them in house.  They’re so fundamental to business.  You know, any business with any size is trying to bring that in house.

 

19:22

Now smaller businesses, fine, they’re going to use a cloud service, but that also seems to be moving that direction. And that brings you to data, which to me is one of the two places I think maybe you can win.  I sort of put data in different categories.  One would be, okay, do you have proprietary data? Do you have data that others don’t have? Now for businesses,

 

19:45

This is where you have to live, right? You don’t give anyone your internal data or your customer data.  And in fact, this may be one of the few defenses that businesses have against things like automation and AIs.  We have the data and you don’t because we’ve been in this business forever. So, data, fine.  But at the AI cloud level, okay, could it be proprietary?  Maybe.  Doesn’t seem like that’s the place to go.

 

20:11

Is it just large? Is it accumulated data? Is it about who has the biggest library of data? Now the problem there is these models require so much data in the inference process that you can’t store it. No, you basically have to tap into a stream of data in real time and do that. So, what you really need is real time data and ideally real time updating data.

 

20:41

where we’re plugged into this river of information, we’re running our inference on it, that’s real-time data that you have to use, because it’s impossible to store most of it.  But it also needs to be updating all the time. It can’t be stagnant, because then other people will have it. And this is something I’ve heard Baidu say, this idea of real-time updating data. And if you think about search engines, that’s really what they run on.

 

21:05

All the web pages, you what does a Baidu search engine tell you? It tells you where to find your information on the web because it indexes all the web pages.  But the web pages change all the time. So, it’s this game of like, yeah, you can copy some of what we’re doing, but the game is always changing and updating. And we’re able to do that. So, this idea of real-time updating data is something I’m looking at. And then the other idea here is what Robin Lee’s been talking about for almost two years now, which is…

 

21:34

the feedback loop between the intelligence of models, specifically industry-specific models, like an LLM models just for manufacturing, LLM models just for financial services, the feedback loop between that and usage.

 

21:52

which is I think why they’re focused on enterprise. I think that’s actually the core strategy is to build this feedback, this flywheel between industry adoption and the intelligence of their industry specific models. So, within that data piece, it looks to me like there’s maybe two winning hands. And then the fourth bucket would be customer relationships, client relationships, value added. I think that’s where most businesses actually live.  I think it’s this idea that

 

22:22

you know, getting customers is difficult, getting them to switch is difficult.  If you are very good at that piece and you have a strong value proposition, that creates a lot of switching costs, it creates a lot of familiarity.  And if you’re doing the standard, we always advance, we always improve our customer value over and over, which I always refer to as a marathon.  That’s a very strong position.  So, for me, I’m thinking the data piece and I’m thinking the

 

22:50

the customer relationship is ultimately where you can win.  Now maybe there’s other places, but that’s sort of my working diagnosis right now. We’ll see.  Anyways, from there the conversation went over to the autonomous vehicles, which I think I was under appreciating how, you know, sort of focused they are on that.

 

23:12

And this a lot of this trip was me sort of catching up to speed on autonomous driving, which hasn’t really been my area, but I know it’s sort of got to be. So, I’ve been reading about these companies a lot.  Yeah, they’re in, they say 50 models of cars. I mentioned they’re doing software, right?  The OEMs do the hardware, they do the software. And it’s a pick and choose model, right? You can, you know, if you just want the mapping.

 

23:38

solution. If you want voice commands you can pick and choose. I believe Huawei in contrast which is in the same space. You know that’s an all-in-one package right. They do the whole thing and that’s what they do. They’ll buy the whole system. Okay so in 50 different models that’s kind of impressive.  They have a mapping deal with Tesla but they don’t have I thought they were going to do an autonomous driving deal with Tesla.

 

24:07

because I don’t think foreign companies are going to be allowed to do autonomous driving in China. You’re going to have to have a partner.  I thought it would be Baidu.  Apparently it’s a mapping deal. thought it was…  Again, this appears to be all China focused. I thought maybe Apollo Go would be going international, which they might, but still mostly a China story at this point as far as I can tell.

 

24:32

Anyways, that was kind of my main takeaways. Interesting.  think maybe valuable to you in this is try Ernie 4.5 or try X1. That’s their reasoning model. They’re both going to be open source soon.  The price is low. Yeah, I think it’s worth keeping an eye on.  For me, looking at the business, you know, what’s the next step?  What’s what do I need to keep my eye on?

 

25:01

I you can follow the models forever because they keep changing. I’m really looking for sort of not cars on the road. I’m looking for the number of OEM deals they have.  How many models is your solution plugged into? Because there’s some pretty good design and manufacturing and other switching costs that get built into that. I think that’s a good metric of success in that area.  For the AI cloud, it’s obviously enterprise adoption.  That’s the key thing. Unfortunately, that number is hard to get at.

 

25:30

You know, it’s hard to know, even if they tell you we got all these clients, you don’t really know what’s in those contracts. So that’s a bit of a black box to an outsider. You can also look at developer adoption, which is sort of a fuzzy metric, but you can generally get a read on that.

 

25:50

like API calls is not a bad one. You can see how many people are just tapping into it on a regular basis.  And they do release, or at least they talk about the growth in the number of API calls they had this year by last year, which is a couple hundred percent, something like that.  Anyways, that’s kind of where I am and we’ll see what happens next.  I’m going to spend time, my focus going forward is not really to rate one model versus another, because one, I don’t really know how to do that. And two, they change all the time.  I’m looking for use cases. Like that’s really where I’m

 

26:20

looking for like, you know, where the rubber hits the road, you know, what is valuable at the end of the day for a business.

 

26:28

What are they going to invest their time and their money in because it’ll make a difference with their customers or their competitors? I’m looking for how it changes business strategy at the end of the day.  So that’s really use cases for me.  So that was the visit. I hope that was helpful, a little bit interesting.  It’s pretty fun campus to bounce around. know, it’s a couple of years ago, I can remember years ago going out there and it was a little bit sleepy. Like you didn’t get the sort of charged up feeling because, you know, Baidu was kind of…

 

26:57

not off the radar, but people weren’t paying attention to it very much for a long time.  It was doing well obviously, but it wasn’t sort of these, it wasn’t on the frontier. It was the very successful search engine. It is absolutely on the frontier now. So yeah, it’s kind of a fun place to be.  And that’s it for the content for today. Pretty short one today.  Fun stuff.

 

27:21

I’m all packed up. I’m going to SongKran. It starts in 48 hours.  So, I got this. This is a big water fight of Thailand. I got the squirt guns. I got the glasses. I got the earplugs because you can get water. I’ve had an ear infection before one of these things because you get hit by river water for two days.  Yeah, I’m all packed up. I’m ready to go.  I’ve got a GoPro all charged up with an extra SD card.  I’m going to mount it on the front of the water pistol.

 

27:51

so that the video will look like Call of Duty.  Got that?  This is out of my fifth or sixth Songkran, so I’m pretty excited about it. I got a whole system now. So, when it me and the girlfriend are flying there tomorrow, yeah, we’ll see. That’ll be great.  I’ll post some videos because that’s really fun.  Other stuff, you know what? I watched a Netflix show this week. It was so good.  There’s a show called Million Dollar Secret.

 

28:19

which is new, it’s been a couple weeks.  And it’s one of these game shows where you put people in a mansion and one people got a million dollars and you got to figure out who it is and if you figure it out it switches to another person. So, it’s all about sort of deceit and cunning and investigating other people.  It was so good.  We watched that thing in about three days. Like it was, you couldn’t stop watching it. Every episode was like twists and turns and even the end was awesome. Like.

 

28:48

I don’t watch a lot of those shows but this one was great. I’m going to totally watch this. It was a million-dollar secret. Just watch the first episode. You’ll get it. You’ll be like, oh this is really interesting. Because everyone’s kind of figuring it out together. Anyways, okay, random thing. That’s it for me. Have a great week. I will talk to you next week. Bye bye.

 

———–

I write, speak and consult about how to win (and not lose) in digital strategy and transformation.

I am the founder of TechMoat Consulting, a boutique consulting firm that helps retailers, brands, and technology companies exploit digital change to grow faster, innovate better and build digital moats. Get in touch here.

My book series Moats and Marathons is one-of-a-kind framework for building and measuring competitive advantages in digital businesses.

Note: This content (articles, podcasts, website info) is not investment advice. The information and opinions from me and any guests may be incorrect. The numbers and information may be wrong. The views expressed may no longer be relevant or accurate. Investing is risky. Do your own research.