In-Depth Analysis: DeepSeek’s “Cost-Saving” Magic and the Pre-IPO “Financial Calculations” Dilemma
Hello everyone, I’m your financial journalist. The hottest topic in the tech and finance circles lately is DeepSeek.
In the past couple of days, I’ve carefully studied two significant pieces of news about DeepSeek: one is the release of a new model version that claims to be more affordable and resource-efficient; the other is the rumor that it is in talks with CITIC Securities, preparing for an IPO (Initial Public Offering).
Putting these two pieces of information together changes the perspective. Previously, when we looked at large models, we focused on “who has the highest performance scores” or “who is the most intelligent.” Now, the focus has shifted to “who can make the most money” and “who has the lowest costs.”
Today, I’ll explain the logic behind this news in plain language, breaking it down for you. We won’t use technical jargon; instead, we’ll discuss what DeepSeek is actually doing and what this means for the entire industry.
---
I. Core Summary: The Transition from “Showcasing Technology” to “Financial Accountability”
In simple terms, DeepSeek is going through a transformation in its identity: **from a “geeky technology company” to a “publicly traded company that has to account for its performance.”
- Technical Aspect: It has released the V4.1 Flash model, whose main selling point is not “intelligence” but “cost-effectiveness.” It has significantly reduced its reliance on expensive hardware (HBM and SSD) and lowered the prices of its API (Application Programming Interface) calls.
- Capital Aspect: It may be about to go public. Going public means it can no longer just say, “I’m investing for the future”; it must answer questions like “How am I making money now?” “Is my cost structure healthy?” “Why would customers be willing to pay in the long term?”
- Industry Aspect: The competition among domestic large models has entered a new phase. In the first half, the focus was on who could develop such models; in the second half, it will be about who can survive in the long term and make a profit.
---
II. Detailed Analysis: A Five-Dimensional Breakdown
1. Going public doesn’t mean immediate profitability, but it does mean “transparency”
Many people think that going public or having an IPO means the company will start making a lot of money right away, or that it’s losing too much money and is in urgent need of funds. However, that’s not necessarily the case.
The news mentions that the Shanghai Stock Exchange has issued guidelines allowing AI companies that “haven’t made much money yet but have impressive technology” to go public. So, DeepSeek’s IPO doesn’t mean it has to be profitable today. The capital market doesn’t deny the value of long-term research and development.
What does going public change, then?
It changes transparency. Before, as a private company, you could say, “I spent 1 billion yuan on GPUs to create a more advanced model next year.” Would investors believe you? Maybe, because you’ve published papers and ranked high. But once you go public, you’re under a microscope:
- How much of the money spent on computing power has actually turned into customer revenue?
- Do customers pay because the service is truly useful, or because of subsidies or free trials?
- Will customers stay if you stop the subsidies?
In other words: Before, it was about “drawing a pie”; now, it’s about showing the actual cash flow and whether you can keep up with your spending.
2. DeepSeek’s Strength: Not “intelligence,” but “efficiency”
The V4.1 Flash model has a key technical feature, but let’s use an analogy to explain it.
Imagine you own a restaurant:
- Old models: Whether you order a bowl of noodles or a fancy meal, the chef has to turn on all the stoves at full power, which is costly in terms of both gas (computing power) and space (storage).
- DeepSeek’s new model: It uses an “asymmetric architecture.” If you only order a simple task (input phase), it uses fewer stoves; for more complex tasks (output phase), it uses more.
What are the benefits?
- Cost savings: The need for expensive HBM and SSD has been reduced by a quarter to half. This means DeepSeek can buy more servers with the same amount of money, or the same servers can handle more tasks.
- Price reduction: With lower costs, it can lower the prices of its API calls.
Why is this important?
For corporate clients, AI models are like utilities. If the cost of using them is too high, companies won’t use them for routine tasks like customer service or coding; they’ll only use them for critical applications. By lowering costs, DeepSeek makes it more feasible for businesses to adopt AI on a large scale.
However, there’s a catch: A low technical cost doesn’t necessarily mean high profits in the financial statements. Research and development errors, data processing, redundant resources for peak times, network maintenance, and compliance costs are all part of the expenses.
The dilemma DeepSeek faces is: The lower the price, the larger the market, but the profit per token sold may be lower. Only when the cost savings outweigh the reduced revenue and the usage volume is sufficient, the business model is sustainable.
3. The Double-Edged Sword of Open Source: Great Influence, but Harder to Make Money
DeepSeek uses open-source software, which means its model code is available for everyone to use.
Benefits: It gains more fame, attracts more developers, and builds a strong ecosystem, making it more attractive.
Disadvantages: More users don’t necessarily mean higher official revenue. For example, you can download DeepSeek’s model and run it on your own servers, and DeepSeek doesn’t earn anything. Or, Alibaba Cloud or Tencent Cloud can host the model, and DeepSeek may only get a small share of the revenue or none at all.
So, the core question for the IPO is: How much of the traffic generated by open source can be converted into direct revenue for DeepSeek? If most users use it for free or through third parties, DeepSeek’s valuation becomes fragile. It must prove that its open-source influence can lead to stable, direct customer engagement and payment.
4. Industry Reconfiguration: From “Competing on GPUs” to “Competing on Efficiency”
If DeepSeek goes public, its impact on the industry will be significant. It will set a new standard: “Just having a good model isn’t enough; you also need to be efficient.”
In the future, the industry will be stratified based on efficiency:
- First Tier (Foundation Layer): Companies like DeepSeek, which have both efficient models and commercialization capabilities. These will attract the most funding.
- Second Tier (Application Layer): Companies that have data specific to certain industries (such as healthcare, law, finance). They may not develop models themselves but use them in their processes, and customers rely on them. These companies will also do well.
- Middle Tier (Most Vulnerable): Companies with no significant technical differences and no unique use cases, relying solely on spending a lot of money on GPUs.
Previously: The competition was about who had the most GPUs; the more GPUs, the higher the valuation.
In the future: The focus will be on GPU utilization, cost per task, and actual business output.
In other words: Before, it was about who had the biggest “house”; now, it’s about who has the highest rental rate, lower utility costs, and stable tenants.
5. Revaluing Talent: Needing Both Technical Knowledge and Business Acumen**
What does this mean for professionals? In the past, an algorithm engineer who could train a model and get high scores was considered top-tier talent. In the future, such talent may become less scarce.
Who will be more valuable?
Those who can integrate “model technology,” “inferencing system optimization,” and “customer needs” into a single profit model. They need to understand technology, deployment, and business strategies to know how to save costs and set prices effectively.
In summary: DeepSeek’s IPO is not just a financing move; it marks a turning point for China’s AI industry from idealism to realism. We’re no longer just celebrating the creation of large models; we’re concerned about whether they can become sustainable businesses.
Here’s a question for you: If DeepSeek goes public, would you pay for its technical efficiency? Do you see the cost savings as a benefit, or are you worried about the loss of revenue due to open source? Feel free to share your thoughts in the comments.