Waking up this morning was exciting; the moment I opened my eyes, I saw such a significant event. By the time I got up, it was already 1 AM Beijing time, but I still managed to catch up with several industry friends to discuss the impact of the upcoming products.
Let’s talk about why this event is so significant.
First, it marks a substantial reduction in the trial-and-error cost of products, allowing for large-scale experimentation:
<ul>
<li>
We have already seen from the supply chain that Tencent has ordered 100,000 to 200,000 H20 units, and now the WeChat version of Deepseek has a clear purpose.
</li>
<li>
Each H20 can support 500 full-version Deepseek users simultaneously, which means that 100,000 to 200,000 H20 units can support 50 million to 100 million concurrent users (concurrent users, not DAU), which basically meets the usage of the first batch of Deepseek users on WeChat and exceeds the concurrent users of ChatGPT.
</li>
<li>
Currently, the cost of H20 is less than $10,000, corresponding to a cluster cost of $15,000. Assuming 100,000 to 200,000 cards, that’s about $2 billion; according to AWS's latest depreciation definition, GPUs can be used for 5 years, with an average annual depreciation cost of $400 million.
</li>
<li>
$400 million is equivalent to 1% of Tencent's Non-GAAP profit this year; to undertake such a significant project only requires 1% of the profit, whereas previously using an equivalent model from OpenAI would have cost 5-10%. Therefore, Deepseek has a trial-and-error cost that is acceptable, but GPT-4o is a cost that Chinese companies cannot accept, not to mention O1/O3.
</li>
<li>
The $400 million trial-and-error cost is even lower than the operational trial-and-error cost when Weishi was launched 7 years ago (I was one of the earliest employees at Weishi), while Tencent's profit back then was only 30% of what it is now. Deepseek has already reduced costs to a level where giants can fully experiment without affecting their core business.
</li>
</ul>
This might also explain why it is happening now and not earlier, essentially offering everyone a GPT-4o+O1 at a 5-10 times cheaper price; you just need to accept it...
In addition, Tencent has already begun to apply Deepseek to knowledge bases (IMA) and marketing scenarios. Today, a friend mentioned to me that domestic e-commerce platforms are also starting to use Deepseek to optimize personnel, including customer service automation, negative review management, copywriting optimization, and long-tail keywords. Moreover, domestic ERP companies serving e-commerce sellers have begun to productize these functionalities.
In our article &#34;The Verification Logic of AI's Grand Narrative,&#34; we discussed Jevons Paradox:
<ul>
<li>“However, in B2B scenarios, customers' willingness to pay is inherently stronger; better models will lead to increased usage, and the elasticity brought by price may be limited. For example, the Salesforce Agentforce product currently has a common customer discount of 20-30%, and reducing model costs is unlikely to encourage customers to pay 10% to stimulate more volume. However, the improvement in model capability can not only increase usage but also lead to a higher ASP.”</li>
<li>&#34;But in most 2C scenarios, a cheaper price means lower trial and error costs, which can cover more customers, and there is no problem with that.&#34;</li>
</ul>
So after drastic cost reductions in 2C, this Jevons Paradox opportunity, which is ten times better than 2B, has emerged.
Let's talk about what major changes there will be at the product level.
First, the WeChat version of Deepseek may be much more user-friendly than Deepseek itself:
<ul>
<li>
Friends who are testing it can already search for public account articles in real-time.
</li>
<li>
Those familiar with the domestic 2C ecosystem know that WeChat public accounts and Xiaohongshu content essentially form a closed loop, while the content of WeChat public accounts is dozens of times larger than that of Xiaohongshu, naturally making it of much higher quality than public search.
</li>
<li>
Of course, it is still a testing version, and there are continuous improvement opportunities.
</li>
</ul>
In the PC scenario, the WeChat version of Deepseek can naturally extend to the browser scenario:
<ul>
<li>
The next step for ChatGPT and Doubao is likely to be developing a browser, as browsers are the easiest to carry Agents, including the latest Operator scenarios that will also land in browsers. DeepResearch is also just search and browsing that users are not aware of.
</li>
<li>
The current WeChat search browser is already very frequent on PCs, and WeChat itself is the most direct representative product of a dialogue box. The WeChat version of Deepseek + AI CoT Agent will greatly simplify future interaction interfaces; future browsers or even operating systems may just be a single window.
</li>
<li>
This is the largest traffic entry point at the front end, the core distribution position.
</li>
</ul>
In the mobile scenario, according to WeChat's tone, it won't be very aggressive; it will be more about feeling the way forward, and it won't be too late to launch when technology and scenarios mature further. It's just about figuring out in what form and experience to interact.
Some scenarios can be done first:
<ul>
<li>
For example, an Agent can help you read and track your various public account subscriptions, then summarize a daily or weekly report, and even allow for interaction. Further extension could turn it into a personal knowledge base.
</li>
<li>
For instance, it could manage group operations and be integrated into WeChat for Work; many e-commerce operations are already doing this, which is very suitable for the current WeChat for Work + video account e-commerce scenario.
</li>
<li>
Another example is the recent New Year scenario, where you need to send greetings one by one. Could AI rewrite sincere blessings in my tone based on the chat dialogues and relationship tags of different people? On the receiving end of the blessings, it would be hard to feel whether it was you or the Agent.
</li>
</ul>
As for the ordinary user group chat Agent or personal assistant that we want more, it is unlikely to happen quickly due to WeChat's high compliance requirements, and there is no rush.
I speculate that in Long Ge's vision, WeChat Agent is not just a tool for improving efficiency; it is more of a lever for building content and tool ecosystems, allowing everyone to provide creativity and improve efficiency for others, serving everyone.
This might be what WeChat wants to do.
Source: Consensus Crusher, Original title: &#34;WeChat + Deepseek: A Turning Point for 2C Applications&#34;

TCTZF

TCEHY

00700

The depreciation cost of WeChat integrating Deepseek is far lower than the cost of integrating an equivalent model from OpenAI, which means that in most 2C scenarios, the product trial and error cost is greatly reduced, allowing for coverage of more customers. Analysis suggests that in the future, WeChat Agent will not only be a tool for improving efficiency but also a means to build a content ecosystem and tool ecosystem

- The article discusses the significant impact of Tencent's Deepseek product, highlighting its reduced trial and error costs, enabling broader testing.  
- With the ability to support up to 100 million simultaneous users, Deepseek's implementation is seen as a strategic move to enhance efficiency in various sectors, including e-commerce and customer service.  
- The author suggests that this development represents a pivotal moment for consumer applications, potentially transforming user interactions and content ecosystems within WeChat.

WeChat + Deepseek: The Turning Point for 2C Applications