My Explanation of Tencent's Big, Revamped Push in AI and Agents (284)

This week’s podcast is about Tencent's revamped approach to AI and agents.

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.

Here is the McKinsey & Co article on arenas.

Here is what Tencent has been doing in AI and agentic products. 

1. Tencent Rebuilt Its Model Building Infrastructure
2. Tencent Kicked Things Off with the Hunyuan Hy3 Preview
3. Additional Models Are Being Developed. I’m Pretty Excited about the World Models.
4. Tencent is Accelerating the Release of New and Upgraded AI Products. And Agents (i.e., Lobsters) Are the Current Focus.

Here is my explanation:

• Pillar 1: Anchor Your AI Agent Strategy on High Value, Scalable Use Cases
• Pillar 2: Balance Aggressive Market Share Growth with Incremental Revenue Capture
• Pillar 3: Prepare for Large, Ongoing Investments and Escalating Industry Competition as the Cost of Dominating this New Economic Arena

---------

I am a consultant and keynote speaker on how to increase digital growth and strengthen digital AI moats.

I am the founder of TechMoat Consulting, a consulting firm specialized in how to increase digital growth and strengthen digital AI moats. Get in touch here.

I write about digital growth and digital AI strategy. With 3 best selling books and +2.9M followers on LinkedIn. You can read my writing at the free email below.

Or read my Moats and Marathons book series, a framework for building and measuring competitive advantages in digital businesses.

This content (articles, podcasts, website info) is not investment, legal or tax 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. This is not investment advice. Investing is risky. Do your own research.

Transcript

00:05

Welcome, welcome everybody. My name is Jeff Towson and this is the Tech Strategy Podcast from Tecmo Consulting. And the topic for today, my explanation of Tencent's big revamped and pretty aggressive push into AI and agents. This has been kind of the biggest news coming out of that company, in my opinion, in the last two to three months. Now it's been going on for a year, but really things have accelerated.

00:34

lots of new products, lots of agent uh products rolling out almost weekly at this point. So I'm going to summarize like one, what do I think they're doing? Why is this actually it's really good strategy and two, what should you start trying of theirs, which you can now do because most of these things are going to be international, not just China. So I'll give you sort of that. I'll probably do this in two podcasts, but let's do the explanation for this one.  Okay, let's see  housekeeping stuff.

01:02

Something I've launched now,  if you want a free copy  of my book, Moats and Marathons, the first one, I have seven books, but the first one's part one, you can go over to jefftowson.com. If you sign up there for our email list,  you can download a free PDF copy of it.  I basically decided to  do that. I'm kind of sick of Amazon for various reasons. So yeah, if you want a free book,  feel free to do that. It's only taken me 10 years. So 10 years worth of thinking on my part.

01:32

It's a free PDF download for a sign up. So yeah, isn't that interesting?  I would ask that you don't share it around if someone wants a copy fine Just go and sign up, but you know, maybe don't share the PDF around if you can I'd appreciate that other stuff I'm going to be in Brazil in a couple weeks  working with some clients there uh any consulting engagements questions meetings uh Workshops things like that. If you're in Brazil, mostly São Paulo a little bit of Rio

02:01

If you're there and you want to talk about stuff, give me a heads up. I'll be in town for a couple of weeks.  Okay. And with that, got to do the, hang on, the disclaimer.  Nothing in this podcast from my writing or website is investment advice. The numbers and information for me and any guests 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. And with that, let's get into the content.

02:28

So I'm a couple weeks behind on podcasts now. I've been running all over and it kind of fell behind. I've caught up on all emails being sent out. I turned out about eight to 10 of them. So I caught up there where I was behind. Now I'm behind in podcasts. So I'll catch up this week, I think. I had a phenomenal couple of weeks.  I was in Beijing, Hangzhou,  Xiamen,  visiting companies mostly. So Alibaba, iQiyi, Tencent, Tsai Niao.

02:55

I was out in Xiamen  last week visiting CATL, which is one of these companies I think the world has not quite  recognized yet. I think they will. They're basically 40 plus percent of all the EV batteries in cars in the world. That's them.  In China, it's more like 50%. They're  a huge player in batteries. It's like them and BYD, but they focus more on, well, one, they do EVs rechargeables, but they also do a lot with data centers.

03:24

because obviously data centers require a tremendous amount of, well, a lot of things, chips, fiber optic cables, transformers, but also a lot of sort of energy storage systems of which batteries are kind of the go-to in addition to other things. So anyways, I'm going to write that up in the next couple of days. I'll send that out. Fascinating company. Xiamen is awesome. If you've never been to Xiamen, it's really a fun city. It's on my short list. When I used to live in Shanghai, that was where you go for the weekends.

03:53

You fly down to Xiamen, it's a couple hours.  People don't really know about it, but there were, in China, historically, there were sort of three  islands on the coast that were sort of controlled by foreign powers. So  the British had Hong Kong, right? Island right across the way, port city, British controlled for a long time.  Macau  Island controlled by the Portuguese. The other one was Xiamen. And I forget who controlled that one, was it?

04:22

the Dutch. I it was the Dutch maybe.  So it's another one of these islands with a historic European city. And now the city is just beautiful. mean, mountains fantastic.  There aren't a lot of tech companies there like their CATL, which is the mothership. uh May 2, which is  the selfie app. They're based there, which is kind of funny. But it's actually kind of an interesting city to watch because certain cities are so attractive for lifestyle  that they can get the best tech workers to come live there.

04:52

like Chongqing, Shanghai, Hangzhou. Hangzhou is a bit of a stretch for some people.  I would put Xiamen on that list. So it might be an interesting tech hub over time.  The contrast to that is like Guangzhou, which has always wanted to be a tech hub, but people don't really like to live in Guangzhou. It's not the greatest city. Anyways, keep an eye on Xiamen. It's interesting. I'm going to write all that up. That'll be fun.  So anyways, that's been a couple of weeks, and then uh I go to Jakarta in a couple of days. Some tech stuff going on there.

05:22

bounce back, Brazil. uh Yeah, it's going to be crazy. June's been crazy. And then in the process, my girlfriend had a knee surgery.  So I've been playing caretaker between all this. Anyways, that's a long explanation for why I'm behind with podcasts for two to three weeks now. I'll catch up pretty quick. Okay, with that unrelated topic, let's get into the actual content,  which is what is Tencent doing?

05:48

And I kind of gave you a heads up over the last couple of weeks that I'm going to be focusing a lot more on Tencent. I  think Tencent, Alibaba are going to be major players  in AI and agents internationally, not just China.  Huawei is a big deal on the hardware side.  Baidu is very big, but it's mostly domestic. These are the two I'm watching more than any others  coming out of China.

06:13

You could put DeepSeek on there. You could put ByteDance with their Seedance app on there. A couple others, not quite as,  you know, they're not playing across the board like these two companies are. And they have these massive domestic ecosystems, which lets them deploy AI tools in a very powerful way, which is kind of what I'm talking about today.  Okay, so what’s the question? The question is, look, what is their AI and agent strategy?

06:41

because over the past six months, they have clearly changed focus and they're given a big push right now and it's pretty well thought out. And everything I'm saying here is stuff that they have published in the last really month. So, I'm kind of summarizing what management has said and others have said.  Obviously none of this is inside in any way, or form because this is a public company.  So yeah, nothing on here is investment advice. Everything I'm going to say is totally public. All right.

07:10

They did over about six months, they basically revamped their approach to building models. Now they're playing across the whole tech stack, apps, because they have the ecosystem massive in China, but also they're focusing externally. They have the models, you know, they're building their Hunyuan model, they have their Yuan Bao, which is an app.  And in addition to their core models, Hunyuan is like their GPT. They stopped calling it Hunyuan, now they call it HY3.

07:39

because I think they want to be more international.  That's their GPT. But they've also got partnerships with DeepSeek. So when you're using one of their products, you can click between which model you'd like to use, DeepSeek or HY3. And they're doing the same thing with agents, where they have their internal agent, which is called QClaw and WorkBuddy, and I'll talk about that.  But you can also, and that's kind of based on OpenClaw, which is a Western open-source framework.

08:04

but you can also click between that and Hermes. So, you can see they're sort of playing across the board. They have their own, but you don't have to use them. Okay, so that's models, and then below that you can get to sort of data ecosystem, and then you can get down to servers, and you can get down to chips, and they're definitely going all the way down the tech stack, but you don't hear as much about them in chips as you do say. I mean, definitely Huawei's the major one everyone's watching, but you could also put Alibaba there as well.

08:33

And Tencent's there as well, but you don't hear about it as much. Within all of this, they've been rebuilding over six months their foundation model. How do you build models? So, they basically, I don't know if they fired everybody, I don't think they fired everybody. They built the whole team over. So, the LLM team is, yeah, they just have a new team. All native researchers and engineers who do that. So, they basically change the people in a fairly...

09:03

significant way and they describe it as quote young and energetic.  So, the people changed. They re-engineered the system and processes for how they do pre-training and reinforcement learning which is kind of the main thing when you're building models right.  After that you're going to update and retrain but it's that pre-training and reinforcement that matters.  And they focused a lot on improving the quality of their data sets. So, they have a lot of specialized experts, data collection, cleaning, synthesis.

09:32

And this is actually kind of a big deal. There's this question of how do you increase the performance of your model? And everything I've ever heard from everybody who builds these things is, you can increase the power of your model, but if you increase the power, the quality of the data sets, you're going to get a bigger bang for your dollar. So data quality probably matters as much, if not more,  in a lot of the incremental gains in performance.

10:00

So that's kind of their restructured R &D team for model building. And they've got a new process of this, and they detailed this out in some press releases. OK, so that was kind of the first thing that happened. That was six months. The first sort of kickoff of this new system was Hunyuan 3 preview. They call it HY3 preview, which came out a couple of weeks ago. This was like, we're going to use this system for the first time, and we're going to start with relatively small model.

10:29

not a massive 500 billion parameter model. They're going to focus on something smaller, which is not a bad way to start when you're doing something new. And the thing they've done, which I think their approach is really interesting on this, basically as far as I can tell, it's a focus on performance  and cost. So they are not, this was not a, this is the Uber model that does everything and it's expensive. No, this is  very practical.

10:59

good performance, but highly pragmatic and useful  at a good cost. So they're using this, you the goal is deployment in use cases, not  let's impress everyone with our capabilities and hit all the big crazy metrics people are tracking. No, it's very,  I'd say it's very focused. So uh mixture of experts, right? That's kind of what you'd expect. uh You know, that's what DeepSeek is doing. That's what a lot of these sort of architectures are.

11:29

that are trying to come in at a very competitive price point,  but actually have some pretty useful, I mean, for those of you who aren't familiar, like  mixture of extra basically means you have a  big model, let's say 100, 200 billion parameters,  but when you run a query through it, it's only using part of the model. So  it's a mixed, it's like a team of experts and you send the query to the right expert for that question. So  you might only be using 10 % of the parameters for a living.

11:58

given query. And you can allocate that dynamically. DeepSeek really rocked the world with their mixture of extras model a year ago. And that's what they're doing. And it's been released in late April, this HY3 preview they're calling it. It's open source, low cost. And within that, it does a lot of stuff. It does images, videos, 3D, auto modalities. It does tags. But

12:27

You know, it looks like it's mostly focused on  complex reasoning, coding,  long context understanding,  and that it's suitable for agents. See, there's a very, there's very much a practical approach. This is not like a video generation heavy model. You know, it's something you can deploy. So, you know, they say this is a quote, quote, proven to be very useful. So I think that's an interesting approach. I'll talk about why I think they're doing this.

12:56

So, okay, you got a highly efficient MOE language model. Within that, the key is the expert routing and the, what they call hierarchical detention, right? You got to send  the query to the right person and sort of give the right hierarchy of attention.  That's how you balance efficiency, i.e. and how deep the reasoning is going to go. It's all about that sort of routing,  expert routing, basically.  Parameters, 295 billion.

13:26

but activates about 20 billion per token. So about 10%, which is pretty common for MOEs. Within that, you got 192 experts within the MOE structure. And the context window is 256,000 tokens. So that's a really interesting structure. And when I read something like that and I listen, I'm like, this is clearly a good strategy. This is not the same.

13:54

This is clearly a well thought out business strategy. This is not,  let's impress everyone with capabilities.  Let's build something super usable.  The cost, about  six cents to 18 cents per million tokens. That's pretty good. That's 10 % of what you see in a lot of these premier ones coming out of these proprietary ones. It isn’t deep seek, but it's pretty close.

14:24

Low cost.  All right. Now they released that in late April. They put it up on open router temporarily. It kind of hit number one on all the boards. Hey, it's number one for coding.  Hey, it's number one for  what was it? Multi-chain. I have a couple of them, a couple of these categories. And what management said during the release of this, I think is pretty interesting. They basically said, look, this is about solving real world problems. So the metrics they're using are not these sort of

14:54

You know, these crazy  rankings you see about which one can do better.  The rankings, the assessments they're using are all based on real world use cases. So they want this thing to hit high on the rankings in practical uses, not these sort of random benchmarks everyone's looking at right now. And  within that, coding is obviously a big one. And then internally, they started deploying this within, know,  Tencent has...

15:22

I don't even know the number, 150 different products,  AI tools, apps that they use,  that they offer. It's great. mean, UNBOW, QQ, WorkBuddy, IMA, IMA, Tencent News,  they're basically starting to deploy it within those. Now that gets you a couple things.  One,  it adds value immediately. It's very use case focused. Two, it gets you a nice feedback loop  that you can use to improve these models.

15:49

And one of the things they talk about in their new revamp process is what they're calling co-design.  These new apps and models are co-designed between the product teams, like the group that oversees QQ or Tencent News,  and the LLM teams. So all of these products are being co-designed,  deployed into their own internal products, so highly use-focused, and then get a tight feedback loop to improve them. Anyways, that all sounds like good business strategy, mate.

16:19

Okay, so that's kind of what's going on.  were two, those are the two big announcements. The third big announcement, and then I'll sort of give you my take on what this all means. The third big announcement is, look, they've been doing other models as well.  HY3  is kind of the most pushed one right now, but they're doing world models and  3D models.  3D models, you give a text, you give a photo, it creates a three-dimensional image.

16:44

That's kind of fun. You can build digital avatars and things like that. Obviously it's good for gaming.  And then they have world models, which  are kind of a big topic right now. And that's basically the idea that like, human intelligence is not based on words beyond a certain point. Words help us with reasoning and thinking  and writing. But when we walk around the physical world, our brain is not thinking in terms of words at all. It's images.

17:14

perception and the world that exists out there is not limited to what we can see. The physics of a building or a bird or a I don't know, avalanche, they exist whether we are viewing it or not. So world models sort of exist in the world of physics and images and all, but also far beyond what any user might view. So anyways, they're rolling out world model teams, which are

17:42

Really fascinating. So the Hunyuan 3D World Model 2.0, which they're calling HY World 2.0. Now obviously if you're the world's biggest gaming company, world models are a big deal.  So  I'm kind of watching them closely  to see what they release here. They're also starting to roll out some robotic  embodied AI models, robots. They have a laboratory that they're doing.  I won't talk about that today. Anyways, interesting stuff.

18:11

last point and then I'll give you my take on this. Yeah, they are absolutely accelerating  the cadence  of product releases. Like the product upgrades, either these are new AI native products or these are existing products that they're upgrading. The cadence of this is crazy. It's almost every week I'm hearing something.  And within all of this, the big push is agents.

18:39

That is clearly the biggest focus for them this year. And within agents, their flagship is WorkBuddy, but they also talk about QClaw, CodeBuddy. So you'll see a lot of news in the last month about Tencent and OpenClaw, which is the open source agent model out of the West.  They've basically adapted it and built on top of it in China, which is called QClaw, but it's going into other things as well. A lot of funny news about that. uh

19:07

The symbol for open claw is a lobster claw. So in China, it's referred to raising lobsters. Like that was how it was described. Like everyone said, oh, you're raising lobsters. But here's the thing. Chinese don't really eat lobster. Chinese eat crayfish, especially on the street. You can get them on the street. So now they're basically calling it crayfish in the Chinese language.

19:33

Yeah, it's kind of funny. So are you raising crayfish? They don't really say lobsters. And the alternative to that is  Hermes, which is another agent platform. That's referred to as horses. So it's this idea that like,  look, we're growing horses and lobsters.  You'll hear this in China. It's pretty funny.  OK, that's kind of my first point. That's a summary of what they've released.  Let me sort of give you the business explanation on this. What do I think their strategy is?  Business digital strategy, not just

20:02

the technical bells and whistles of these things.  Okay, so  there's a problem at the center of this. Let  me summarize what they've done. They've rebuilt their model building infrastructure and processes, people, processes, data sets, blah, blah. They've kicked off their new process with  HY3. Interesting.  Now they're doing other models using the same process. World models are the one I'm kind of personally excited about.

20:32

That's all in place. Now they're accelerating the release of new  AI native products and upgraded existing products.  And within that, agents is the centerpiece right now. OK. The problem, the business problem in all of this, which you actually hear this on Twitter and stuff in the last couple of weeks a lot. Look, these things have costs associated with them. Like,

20:58

There was kind of a default way of approaching this, which is how software has always been done,  which is, you know, software traditionally was a lot of fixed cost. You build Microsoft Word, it's a fixed cost upfront and then to oh some degree ongoing. And then your costs are pretty much done and you can give it away for free. And a lot of  business strategy and pricing strategies based on the fact that

21:23

Traditional software had a very low close to zero marginal production cost You know, you can just give away copies of an eBook It doesn't cost anything. I can read the eBook ten times a hundred times doesn't cost anything or Marginal cost is very small. Okay And when you have that type of approach you see companies going for market share We got to push hard. We're going to give this away for free. We're going to you know

21:54

lose some money for a couple years, but we're going to end up with the market and we'll find some way to monetize later. That's kind of been the default strategy for traditional software  and internet-based software and even SaaS to a large degree. You can give away. There's always a uh freemium level. And maybe to go beyond that, you could do like YouTube where they have ads at a certain point or Facebook added ads at a certain point.  Okay.

22:22

The cost structure of intelligence doesn't work that way. It turns out the cost structure of AI-based intelligence and agents has a significant variable cost.  And that means you can't really give that stuff away for free. And that's what a lot of companies are kind of realizing right now is,  we fired a bunch of our software people and we gave free use to AI and AI agents to our staff,  and our costs have gone through the roof.

22:52

Yeah, because this is the architecture.  The usage and the cost structure of generative AI  is different than traditional software. And I've talked about this before. I did some  kind of wordy, struggling podcasts and articles a year ago talking about what is the cost of maintaining correctness. You we have an AI app.

23:17

where you're using it to check your documents, you're using it to get a medical diagnosis, whatever.  What is the cost of providing that intelligence on an ongoing basis? People put in queries, they get responses.  And then what is the additional cost of  maintaining the correctness? Because you can see the quality of  an AI answer  decrease over time.  These are non-deterministic.

23:46

So you ask AI something 10 times, you're going to get 10 different answers. And over a year,  the cost of keeping it accurate can be significant. So these generative AI tools, they look like a mixture of traditional software economics and services. Okay, so one of the things that the Tencent management has talked about  is, actually they got asked this, but one of the questions was,  are you just going to give this away for free?

24:16

your tools and go for market share, which is kind what AI, OpenAI has been doing.  Or are you going to be more like Anthropic where, look, you're not going to do that game. You're going to go for a small pool of high power users who are willing to pay. You know, let's say software developers who pay a lot for the coding tools. You know,  what is your focus?  And  their answer,  I'm paraphrasing, was kind of what I just said. Look, the cost structure

24:45

of intelligence and these tools, it's not the same. The variable costs are significant.  The cost of maintaining correctness, that's my phrase,  in some cases can be significant.  So you're looking at a different business model. And I was looking into this a year ago because I was like, look, most competitive advantages and modes  follow from the cost structure, half of them. OK, this is a new cost structure, which means the modes are going to be different.

25:14

I haven't talked a lot about motes based on variable costs.  Most of the economies of scale I talk about, those are about fixed costs. Okay, you can have motes based on variable costs.  Toyota is an example of that. Okay, so I'm going to actually end up writing a lot more about what would a mote be in intelligence, and we're going to end up talking a lot about variable costs, because  you use something more, you pay more. Now, why does this matter?  One,  as soon as you use generative AI,

25:44

The amount of compute you're using increases dramatically. That's why everyone's building these massive data centers.  This is not something you can run on your laptop. mean,  so the compute is huge. The amount of data that must be ingested and processed and then stored in some form  is huge. The amount of power required is very big. So the variable costs are a big deal. And I'm going to talk about SpaceX next week,  because they did their IPO filing.

26:15

which was kind of weird. I know if I would recommend reading it. But the part that was fascinating  is they detailed out their strategy for XAI and Grok,  and they talk about the cost per token as the main variable they have to drive down over time.  So they're talking about how do you drive the cost down? And they talk basically about the cost of the model, the cost of the  compute infrastructure, and the cost of the power infrastructure. As the three main drivers,

26:44

of the cost of these things, the cost per token. And then cost per token is the primary performance driver. So I'll go into that next week, but it's pretty interesting. Okay, so that's the normal. People are using intelligence. You've given it to the coders. The costs are going through the roof. That's all going to get much worse once agents start doing compute on their own. know, a human will go on and ask a question and get an answer, maybe load up a document.

27:13

there'll be an episodic back and forth with the AI. Agents live in this 24-7. They are sending queries back and forth all the time. So the usage of an agent of these tools is like 1,000 times higher than a human being. And we're getting close to the world where the internet and mobile networks have to be built for 8 billion humans and 800 billion agents. Because once one agent

27:41

works, you can make 10 of them with Ctrl C, Ctrl V,  and they can all start querying the models. Now the real tipping point is what happens when one agent can create another agent on its own? That's how you go from a million agents to a billion to a trillion  very quickly.  Well, they're all going to be running variable costs. So this whole problem is going to explode. We're not there yet, but  we've seen the first shoe drop.

28:09

with humans are using a lot of this stuff and the cost structure surprise people. The other shoe's going to drop when agents take off. So anyways, that's your fundamental problem.  And  based on that problem, you can kind of see 10 cents approach. And this is my assessment. This is me talking. So here's how I would summarize  what they're doing. Three pillars. Pillar number one.

28:34

They are anchoring their AI and especially their agent strategy  on high valuable,  scalable  use cases. They are focusing on areas and niches where the value is so high that people will pay for it. That is the only way you can scale these things up. If you're barely breaking even on your cost structure because the variable cost keeps getting out of control,  or if you're losing money,

29:03

The last thing you want to do is for that thing to scale up dramatically because your problem gets 10 times worse. So they're starting. Let's focus on the high value use cases that basically you can monetize immediately and then we'll scale up those. We will anchor our strategy there. OK. What are those? Well, they've already kind of work productivity. You know, and the agent they're talking about here is work buddy.  They've launched three. Now it's

29:33

Like I literally wrote this two weeks ago. They've launched three main agents, WorkBuddy, CodeBuddy, QClaw. They've already got five now. Like I just checked today and there's like two new ones. There's a data agent and there's one about design tools. Like they're moving real fast. Okay, but pillar number one, high value,  scalable use cases, start with work productivity, that's WorkBuddy, which is basically an agent you run at work.

29:59

sitting at your desk that does everything for you. runs on your laptop.  It can run on the cloud in your business and you can access it with your laptop  or you can access it with your phone. And specifically you can use WeChat.  The next podcast I'll probably do that in the next day or two. I'll just talk about these agent tools like WorkBuddy. Okay. Get you close to the customers. You can see they're doing that. But yeah, that's where they're starting. The other one they launched immediately was CodeBuddy.

30:28

Same thing, we're focusing on software coders, giving them an agent, uh high value use case that you can charge for, we scale it up. The third one is QClaw.  That's sort of a personal assistant.  It's sort of a general for everybody case. I'm not totally sure that's a high value use case.  That's basically built on OpenClaw.  That one is international already. All three of these are going to be available internationally.

30:58

I think you can use all of them right now. I'll check that. Like two weeks ago, you couldn't. I think you can now. So I'll put the links in there in the next podcast when I talk about these so you can go try them.  My priority is work buddy. Like I'm building work buddy into what I'm doing right now.  So, okay. That's kind of pillar number one. Pillar number two. If that's your basic product strategy, what are your KPIs?

31:27

What are your metrics? And their metrics appear to be market share, growth, and incremental revenue growth. That third one was interesting because if we were talking about traditional software,  we'd be talking about the other two. We'd talk about how fast are you growing? What's your market share? You got to be the number one video sharing platform YouTube because you got to get network effects and win. We don't care if you're making money right now.

31:57

Now you can't do that here. It's got to be market share, growth,  and incremental revenue.  Every time you grow and expand market share, you have to increase your revenue  to cover your increasing costs. Now, what is not in the KPIs as far as I can tell, uh profitability. So usually you'll hear this discussion as,  are you going for market share and growth or are you going for profits?

32:25

They've kind of split the question. We're not going for profits as far as I can tell. We are going for market share growth and incremental revenue growth because we have to cover at a minimum  our rising variable costs. I thought that was quite clever.  Okay, so they're balancing market share and incremental revenue. Pillar number three, last one.

32:49

There was a really cool  study written by McKinsey a year or two ago, multiple, they've actually written multiple times about the same subject, called  the arenas of growth.  They call these the arenas of competition. And they basically say, there are 18 powerful industries that are transforming business almost everywhere.

33:14

And they're very different than what we would consider other industries, even other high growth industries. No, no, these are  transformative industries.  And that would be something like EVs,  e-commerce, consumer electronics, software, audio, video entertainment, in that, yes, they have high growth. Yes, but they're also sort of  very dynamic.

33:44

If you're going after, and they're going to be a disproportionate amount of all the economic value in the business world  is going to go into these 18 sectors.  They're changing so much. If you're going after one of these, your competitive approach has to be different. Yes, you have to go for growth.  Yes, you have to sort of be very dynamic because the capabilities

34:12

you need to compete on these, keep evolving. You can't just launch a product in EVs and then focus on growth. No, you have to create an entirely new product 12 months later  and an entirely new product because the tech underneath is changing itself,  which means, OK, you're going for one of the arenas. That means you're going for high growth and it means you want to capture outsize profits. Fine. To play that game,

34:41

you have to be prepared to invest and reinvest and reinvest in large amounts for a long, long time. You can't just be good at video generation today.  You have to be good in 12 months, 24 months, because the capabilities keep advancing. So this is a long-term game with lots and lots of spending required to stay at the frontier. That's how you play if you want to do this. So I think the third pillar for Tencent is, look,

35:10

They  are prepared, as far as I can tell, for large  ongoing  investments in industries that are going to continually escalate in terms of the competitive dynamic and the capabilities. You are playing on a rapidly advancing frontier,  and that is going to be very expensive. So you have to be prepared for the large ongoing investments.

35:40

And at a minimum, to play this game well, I think you need two things. You either need to make sure that your variable costs are covered, which we've talked about, and you probably need an external source of money beyond these products. Now Tencent has all its other businesses which make a lot of money.  Other companies like OpenAI have to continually raise money to play this game. So you're going to need both of those. Incremental revenue as you grow plus probably constant

36:09

cash infusions into this space for a long time. And if you're not prepared to do that, you probably don't want to play in this space because you'll spend a lot of money for two years. You'll run out of dry powder and then you'll get beat. So this is kind of a game for the big boys.  Anyways, there's a good McKinsey paper about this called the 12 arenas, the 18 arenas of competition. It's pretty good, but it's a long term, very expensive fight on the frontier of technology and capabilities.

36:40

But if you win it, yeah, you're going to win huge. So that's a good way to think about  what they're doing here. Anyways, that's sort of the third pillar.  I'll put a link to the McKinsey paper online. It's pretty good. They've kept writing more and more about this over time. They keep updating it. so yeah, it's kind of a fun space. I'm happy to see that a lot of what I do sits in those pillars. Actually, that made me feel pretty good,  those arenas.  I'm like, oh, most of what I do is in this space. That's good, because  according to them, at least, that's where

37:10

A big portion of the profits of the entire business world are going over the next five to 10 years. Okay. It's a hard game, but at least it's a game you can win. Like if you win, there's a payoff. All right, that's pretty much it. I'll put those in the notes. I'll put the three pillars. I'll put sort of my assessment of what's going on.  I'll do another podcast about this probably tomorrow or the next day and talk about the specific tools, the agents that they're rolling out and where you can try them and why I think they're important.

37:40

Anyways, that'll be it for that.  One last disclosure here just to sort of dot my I's and cross my T's.  My business, which is called Asia Tech Services,  has had in the past 12 months a consulting relationship with Tencent. So  should disclose that. But this article, everything I'm saying, this report,  not paid, sponsored, anything like that. But yeah, there was a  relationship. Actually, you have to disclose. If it's been within 24 months, I think you have to mention it. But anyways, in the last 12 months, there was something.

38:10

Okay, that is it for this. As for me, I'm doing pretty well. Life's going on. It's such a fascinating time. I feel kind of guilty that I get to spend my days doing this. It's so interesting. I kind of get to do this for a living. It's really kind of amazing. So many things are changing so quickly. I get to learn about really cool companies all the time. And occasionally you fly around. I'm going to bounce out to Rio in a...

38:38

Two weeks. I've got some stuff in Sao Paulo, but you know, I used to spend as a rule in life I used to spend one to two months in Rio every December January and it was just Really nice part of life. I haven't done that really in a three or four years Just I've been locked down. So I'm going to sneak away to Rio next after we finish up in Sao Paulo I'm going to sneak away there for a weekend and hang out if you're in Leblon Take a look check and look for me in the Starbucks. That's where I usually sit and type

39:07

And because it's too hot outside obviously so I'm always in that level on Starbucks. I've written like half my books there. Anyways that's it for me.  I hope everyone is doing well and I will talk to you probably in a couple days.  Bye bye.