My Explanation for Tencent’s More Focused and Aggressive Strategy in AI and Agents (3 of 4)

Posted on by

In Part 2, I summarized Tencent’s accelerating approach to AI and agentic products. Which had 4 events.

  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.

In this Part 3, I want to give my assessment of this. Which I think follows from the different nature of the compute at the center of AI and agents.

Then in Part 4, I’ll focus on Tencent’s new agentic products, which are a big focus right now (among others).

Why You Can’t Give AI Agents Away for Free. The High Variable Costs of Intelligence.

During the management Q&A, there were some interesting comments about how OpenAI is going for mass adoption while Anthropic is going more for a small pool of high intent power users (like coding).

And Tencent management made an important point about how the economics of intelligence are different than the economics of traditional software. In particular, the variable costs can be significantly higher.

I have written a lot about how the economics of AI are different than software. I called this the “costs of correctness”. My article on this is here.

Unlike traditional software, you can’t just build a GenAI service and then let everyone use it for free. The compute and other costs required for intelligence are not mostly fixed. The marginal costs of production are not close to zero.

Productivity tools like Gmail and Excel can be built once and then mostly given away for free. Your costs are mostly just storage. So, you build it once and then use it over and over for free (mostly).

But in GenAI, every new inquiry costs money.

Dramatically more compute is required. Dramatically more data must be ingested and processed. Everyone talks about the token costs of different models. DeepSeek made a big impact because it was so much cheaper. Variable costs are a big deal in generative AI, whether it be LLMs, image generation, or video generation.

And agents are starting to make this problem 1,000x worse.

Agents don’t use GenAI episodically like humans. They run continuously and send endless context and tokens back and forth. And the number of agents is growing exponentially. They will soon dwarf the number of humans.

Mobile carriers like Huawei talk about how mobile networks must now be built to serve both 8 billion humans and 800 billion agents. Where traffic will shift from being episodic with a focus on download to continuous with both surging uploads (context and sensing) and downloads (tokens).

Here is my article on agentic-native networks.

So, the variable costs of GenAI, and especially agents, are going to be significant.

And users of AI and Agentic products will need to pay. Otherwise, you can’t scale these products. Intelligence will not be free. The logic and economics of software have changed.

So Anthropic had it mostly right. You need to focus on niches where users will pay. And that means you need to focus on high value use cases.

And I think you can see this in Tencent’s revised approach to model building and AI / Agentic-first products (in Part 1).

Here’s how I would summarize Tencent’s current approach.

Pillar 1: Anchor Your AI Agent Strategy on High Value, Scalable Use Cases

These would definitely include work productivity (which they are targeting with WorkBuddy) and coding (which they are targeting with CodeBuddy).

I think this is why the LLM teams are co-designing new products with the products teams.

It that gets you close to the customers. And to what they actually value (and will pay for). That is different than focusing on building impressive capabilities. Note: Baidu has a similar approach and is really focused on industry and especially manufacturing AI use cases.

I think you can also see this focus in how they now measure model performance. One of their main points made during the release of Hy3 was they were not evaluating performance based on generic leaderboards. They are measuring model performance based on effectiveness in specific tasks and uses cases, not just generic performance. There was a lot of discussion about the refocusing of evaluations from widely gamed benchmarks to real world use cases.

Finally, Tencent is definitely building user feedback loops and intelligence feedback loops into these products.

Usage and data from users goes right back to designers. And given the vast distribution and user base of Tencent, this should be a very powerful feedback loop for rapidly improving these products. Note: Models don’t just benefit from feedback to fix problems. The data itself makes the model smarter.

Pillar 2: Balance Aggressive Market Share Growth with Incremental Revenue Capture

As discussed in Part 2, the ROI for Tencent’s AI-related initiatives appears to include both marketshare and incremental revenue. And not just market share and adoption. Which is what a lot of model and app builders are focusing on.

The focus on incremental revenue makes sense given the variable costs. These products need to grow the revenue. Which forces you to focus on high value use cases (i.e., what people will pay for).

I like how they are balancing marketshare with incremental revenue.

Pillar 3: Prepare for Large, Ongoing Investments and Escalating Industry Competition as the Cost of Dominating this New Economic Arena

McKinsey & Co has a good paper about the new “arenas” of competition. They argue that there are 18 powerful industries that are transforming the business landscape. These arenas have both high growth and dynamism. And they will capture an outsized portion of the economy’s economic expansion.

These arenas include EVs, ecommerce, consumer electronics, software, video and audio entertainment and other sectors.

And they include cloud services and information-enabled business services. So definitely AI and Agentic native products. And AI and Agents applied to quite a few of Tencent’s core businesses.

McKinsey argues that these industries have different economics and competitive dynamics.

  • They are massive potential markets with big long-term growth. long-term. These are big opportunities that will continue to grow. That means a long-term fight.
  • They have escalatory competition. The technology is rapidly advancing. You have to keep building and rebuilding products and capabilities. You have to stay on the frontier. Which means large, ongoing investments.

For Tencent, this means they need to be prepared for and committed to a long-term ongoing fight. Lots of new capability building. Lots of new product development. And lots of ongoing investment.

This is what I thought about when I dug into Tencent’s revised approach to AI and Agents. The approach plus the recent financial returns look to me like a financially sustainable approach to building and investing in this area long-term.

Financial analysts, of course, will be trying to calculate the ROI for all this investment (which includes staffing and capex).

My assessment of Tencent’s answer to that is that the return will be a mix of long-term growth (and value) and near-term gains. But that’s just my guess.

Here is McKinsey’s report on arenas. It’s a pretty good read.

***

Ok. That’s my take. In Part 4 (last part), I’ll go into their Agent Products, including WorkBuddy and QClaw.

Cheers, Jeff

Note: This article is not investment advice. The information and opinions 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.

——-

Related articles:

From the Concept Library, concepts for this article are:

  • AI Agents
  • GenAI

From the Company Library, companies for this article are:

  • Tencent

———-

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.

Comments are closed.