Abandoning 94% of users? Do you understand DeepSeek's strategy?

Recently, news that “DeepSeek users have plummeted by 94% in six months” has caused a stir online. Many people saw the headline and felt a pang of anxiety: is another AI star about to fall from grace?

To be honest, when I first saw this figure, I was quite shocked. However, after spending some time in the AI industry, you come to understand a fundamental principle: focusing solely on user numbers is as superficial as judging someone's health based solely on their weight.

Today, let's play detective and peel back the layers to uncover the true story behind DeepSeek. What lies beneath may reveal the new rules of the game for the second half of the AI industry.

Bidding farewell to the “traffic frenzy,” AI enters the “value-driven” era

First, we need to shift our perspective: the AI industry is no longer in its “wild growth phase” where companies burn through cash to acquire users.

Looking back at 2023, that was the “first half” of AI, where various large models competed fiercely, with the focus on who had more users and greater brand recognition. Everyone was scrambling to expand their market share. It felt like the early days of the internet, where the loudest voice won.

But now, the tide has turned. The AI competition has entered a more brutal and pragmatic “second half.” What are they competing over? Two words: commercialization and scenario value.

In simple terms, no matter how advanced your technology is, can it ultimately solve real-world problems? Can it help people accomplish more with less time? Most importantly, can it generate revenue to sustain itself?

ChatGPT, Doubao, and DeepSeek, the general-purpose chatbots we're discussing today, all face the same awkward dilemma: insufficient user retention. Many people are just curious and use them for a few minutes, but once the novelty wears off, the app might end up gathering dust in the corner of their phone.

If a product cannot deeply integrate into your work and life like WeChat or Photoshop, then the so-called user base is just a fleeting trend—it comes quickly and disappears just as fast.

Therefore, using “declining user numbers” to dismiss DeepSeek is like judging a marathon runner by the speed of a 100-meter dash—it completely misses the mark. What the market truly cares about is no longer how many “casual” users you have, but whether you can create tangible value in specific scenarios and build a self-sustaining, healthy ecosystem.

User “purification”: From a jack-of-all-trades to a professional tool

So the question arises: Where did DeepSeek's users go? The answer might surprise you: They didn't “leave”; they were actively “purified.”

DeepSeek is quietly undergoing a transformative shift: from a “jack-of-all-trades” that can discuss anything, to a “super productivity tool” that helps professionals tackle the tough, tedious, and technical tasks.

This transformation will inevitably filter out a large number of casual users who only want to chat, but those who remain are core users of extremely high value. Let's look at a few real-life examples to get a sense of this.

Scenario One: The “Secret Weapon” for E-commerce Professionals

If you work in e-commerce, you know the pain: to analyze a best-selling competitor product, you have to manually copy and paste the title, five-point description, and detailed product description, then study every word and sentence. An entire afternoon can pass, and your eyes are strained from staring at the screen.

Now? Pros are using the “DeepSeek + Seller Genius” combo. With one click, you can extract all competitor information. DeepSeek acts like a seasoned operations director, generating a report in seconds: copywriting quality score, underutilized keywords, SEO optimization suggestions, and even a clear analysis of the competitor's core strengths and weaknesses.

This isn't just about saving time; it's using AI's “god's-eye view” to help you with operations, hitting every target precisely.

Scenario Two: The Marketer's “External Brain”

Take market research or consulting professionals, for example. In the past, to write an industry report, they might have to sift through hundreds of web pages, manually screen and organize information, and pull all-nighters—only to have the report criticized by their boss as “not in-depth enough.”

Now, they simply input their request, such as “Analyze user preferences in the 2024 new energy vehicle market,” into DeepSeek. AI can instantly transform publicly available information across the web into your “personal think tank,” seamlessly handling data extraction, cleaning, and insight generation.

Some users have reported that this process saves them at least three hours daily, and the reports they submit are logically sound and data-rich, even impressing even the most critical supervisors with their professionalism.

Scenario 3: The “inspiration accelerator” for product managers

What's even more amazing is in the product manager community. In the past, if you had a vague idea, such as “I want to create a fitness app targeting 996 office workers who don't have time,” you would have to go through lengthy market research and endless brainstorming sessions.

Now, you can share this idea with DeepSeek. What it can do might blow your mind:

It might automatically scrape negative reviews of competitors across the web and tell you, “Don't make the social features too heavy; users find them annoying”; it will combine user pain points to recommend a few core features: “Try fragmented video courses and add a sleep monitoring feature”; or it can directly generate several product prototype sketches in different styles for your reference.

What used to require a team to hold two meetings and conduct three days of market research can now be clarified by AI in just half an hour.

Got it? DeepSeek isn't losing users; it's precisely “capturing” them. It's ditching those ‘optional’ chatters and instead targeting e-commerce professionals, marketers, and product managers—the “golden customers” willing to pay for efficiency. These are the true goldmine for AI commercialization.

Revealing the ace up its sleeve: 5.6% cost, 545% profit margin—that's the real game-changer.

If deepening its focus on specific scenarios is DeepSeek's tactical approach, then its powerful technology and formidable cost control capabilities are the strategic trump cards that enable it to play this game.

In the AI industry, widely recognized as a “money-eating beast,” whoever can break free from the cycle of burning cash and achieve profitability first will survive and emerge victorious.

This year, DeepSeek unveiled two trump cards that would make all competitors nervous:

The first trump card: extreme cost control.

According to public data, the API call price for the DeepSeek-V3 model is only 5.6% of that of industry giant GPT-4o!

What does this mean? It means that a developer or a small company looking to use DeepSeek's technology to develop their own AI applications can now do so at a cost that's practically rock-bottom. What was once an AI innovation only accessible to large corporations is now within reach for a small team of three or five people, or even a single tech expert. This significantly lowers the barrier to innovation and attracts a steady stream of developers to DeepSeek's ecosystem.

The second ace up its sleeve: remarkable profitability.

DeepSeek has publicly disclosed the theoretical cost-profit margin of its inference system—a staggering 545%!

This figure completely overturns the external stereotype that “large models = losing money to make a splash.” It proves with data that DeepSeek's technical architecture is not only top-notch but also completely viable commercially, with strong self-sustaining capabilities.

While other players are still struggling with high server and operational costs, DeepSeek has quietly found the perfect closed loop from technology to profit.

On one hand, its extremely low usage costs attract countless developers; on the other hand, its astonishingly high profit margin ensures sufficient resources. The combination of these two factors has formed an impenetrable technological and commercial moat.

So, returning to our original question, is DeepSeek really not viable?

On the contrary. While the outside world is still debating that 94% figure, it has already completed a seamless transition from pursuing breadth to pursuing depth, and has confidently revealed its strongest trump card.

In the second half of AI, it's not about who shouts the loudest, but who has the deepest roots.

Some have figured this out, while others are still confused. What about you? Do you think the future of AI belongs to the traffic giants casting a wide net, or the efficiency experts with deep roots? Feel free to share your thoughts in the comments section.