Why is the 'National Team' Eager to Invest in a Non-Profitable, Open-Source Company?

DeepSeek is set to raise funds.

This event itself warrants more attention than the amount of funds raised.

Over the past two years, DeepSeek has been the most unique entity in China's AI circle, refraining from fundraising, commercialization, or roadshows. All its code is open-source, and its API pricing is 90% lower than that of all competitors. Founder Liang Wenfeng has been supporting the entire team with the money he earned from his own quantitative fund, never meeting with investors. The industry jokes that DeepSeek is the only company in China that "uses its own money to pursue AI."

Now, this company has suddenly started meeting with investors. And not just any investors—the National Integrated Circuit Industry Investment Fund is negotiating to lead the investment, with internet giants like Tencent and Alibaba waiting in line. The valuation has directly doubled from $20 billion a few weeks ago to $45 billion. The expected fundraising scale is 50 billion RMB, with all participating institutions investing at least 5 billion RMB each. Sources close to the deal have said: Now it's the investors chasing after Liang Wenfeng, not the other way around.

A company with almost no revenue, fully open-source code, and an unproven business model is valued at 300 billion RMB.

What is the logic behind this figure?

Why Raise Funds Now?

There was a simple reason why DeepSeek did not raise funds before: it did not need the money. Liang Wenfeng's High-Flyer Quant manages over 10 billion RMB, with annual profits sufficient to support DeepSeek's R&D team. Not needing external capital means not having to explain the business model to investors, not having to consider exit strategies, and not having to compromise on R&D direction due to profitability timelines. Such freedom is extremely rare among AI startups.

So why open the door now? There are two reasons: one obvious, one fundamental.

The obvious reason is talent retention. DeepSeek has lost five core developers over the past year. The "genius girl" of the post-90s generation, Luo Fuli, was poached by Xiaomi. The core authors of V3 and R1, Wang Bingxuan, have also left. Ruan Chong from the multimodal direction, Wei Haoran from the OCR direction, and Guo Daya, who participated in the development of all models, have all left one after another. This shows the extent of the talent war in the AI industry. ByteDance, Alibaba, and Tencent are offering tens of millions or even hundreds of millions to poach talent. DeepSeek offers high salaries in the industry, but it lacks stock options because, without fundraising, there is no valuation, and without a valuation, equity incentives cannot be used. How do you retain an engineer whom the market is offering 100 million to poach?

The fundamental reason is the ecosystem. Liang Wenfeng is well aware that his team's core strength lies in models and algorithms, but for a large model to truly succeed, technology alone is not enough; an application ecosystem is needed. Who has 1 billion users on the C-side? Tencent. Who has the largest commercial scenarios in e-commerce and cloud computing? Alibaba. Bringing in strategic investors is not for the money but to have DeepSeek's models run in more scenarios. Additionally, bringing in the National Fund has another consideration: AI will inevitably face security regulations in the future, and having state-owned shareholders will make compliance much smoother.

What Does the $45 Billion Valuation Represent?

$45 billion is equivalent to 300 billion RMB. This figure already places it in the absolute first tier among domestic AI companies. Yuezhiyanmian was raising funds at the same time, with a valuation of $20 billion. Zhipu AI's last valuation was around $4 billion. Baichuan Intelligence's valuation is even lower. DeepSeek's valuation is more than double that of all domestic AI unicorns.

However, if you compare it to overseas companies, this figure seems conservative. Anthropic was raising funds at the same time, with a valuation of $900 billion, 20 times that of DeepSeek. OpenAI's latest valuation is $852 billion. DeepSeek's V4 has already approached or even surpassed GPT-5's level in multiple benchmark tests, but its valuation is only one-seventh of OpenAI's.

Why is there this gap? Because DeepSeek has almost no revenue. OpenAI has subscription revenue from ChatGPT Plus, enterprise API revenue, and a deep partnership with Microsoft. Anthropic has strategic investments from Amazon and Google, plus commercialization revenue from Claude. What about DeepSeek? After a 90% API price cut, its pricing is so low that it's almost giving it away. Its open-source models are free, and its commercialization efforts are extremely limited. Investors are not paying for today's revenue but for the technical capabilities of this team.

The core logic behind the $45 billion valuation is this: DeepSeek has created models of nearly equal quality at less than one-tenth the cost of OpenAI, indicating extremely high R&D efficiency. R&D efficiency is the rarest commodity in the AI industry. There may be fewer than ten teams globally capable of creating top-tier large models, and DeepSeek is one of them, and the most efficient one at that. You're not investing in its financial statements today but in its position in the future AI competition.

Does this logic sound familiar? It's the same logic investors used when they invested in early-stage Amazon or Tesla. The companies weren't profitable or were even losing money, but investors saw their efficiency, their barriers, and their position in the race. The $45 billion valuation is not pricing a financial statement but the scarcity of "China's strongest large model team."

What Does the National Fund Leading the Investment Mean?

The most noteworthy aspect of this fundraising is not the amount but the identity of the lead investor.

The National Fund is the "national team" capital in China's semiconductor industry. The third phase of the fund has a registered capital of 344 billion RMB, making it the largest national-level industrial investment fund in the country. Its past investment directions have always been hardware-focused, such as SMIC, Yangtze Memory, and Naura, all in chip manufacturing, equipment, and materials—the bottleneck sectors. If this investment materializes, it will be the first time the National Fund has invested in a large model company. From hardware to software, from chips to models, the national team's investment logic is expanding.

The significance of this signal is greater than the $45 billion valuation itself. It indicates that at the national strategic level, the importance of AI large models has been elevated to the same level as chip manufacturing. The National Fund invested in chips before because China was being constrained by the U.S. Now it's investing in large models because the second half of the AI competition is not just about hardware but also about models and ecosystems. Having chips alone is not enough; you also need models to run on them.

Moreover, the National Fund leading the investment also implies protection. You can recall the recent incident where the National Development and Reform Commission prohibited Manus from being acquired by foreign investors. Once a Chinese AI company is defined as a "national strategic asset," its financing, mergers and acquisitions, and technology transfers will all be subject to security reviews. The National Fund investing in DeepSeek, in a sense, places a "national-level" seal on the company. Any future foreign investment or acquisition of DeepSeek will have to undergo a review similar to that of Manus. This is not just an investment but a signal: DeepSeek is a national asset, not a commodity freely tradable on the market.

What Is Liang Wenfeng Thinking?

A few days before the fundraising news broke, DeepSeek made an interesting move: equity restructuring. On April 27, its registered capital increased from 10 million to 15 million RMB, and Liang Wenfeng increased his direct stake from 1% to 34% through capital injection, bringing his total control to 84.29% including indirect stakes. It is also reported that he will invest in this round of fundraising in his personal capacity.

This move has a clear meaning in the primary market: the founder is consolidating control before fundraising. Even after the 50 billion RMB fundraising, Liang Wenfeng will remain the absolute controlling shareholder. He is bringing in the National Fund, Tencent, and Alibaba but not relinquishing control. He wants the money and the ecosystem, but he still holds the steering wheel.

This is consistent with DeepSeek's style all along. Liang Wenfeng is a person with strong judgment and execution (execution power) in technical directions. The fact that V3's training cost is only one-tenth of OpenAI's is not accidental; it's because he chose a different technical route (MoE architecture + FP8 training) from Silicon Valley from the beginning and stuck to it unswerving (unwaveringly). Such a person will not change the R&D direction just because he has raised a round of funds. This is also what investors are betting on. When you invest in DeepSeek, you're not investing in a founder who "listens to investors" but in one who "won't listen to you even if you invest, but he will probably make the right decisions."

In the primary market, many founders, after taking investors' money, start running to the investors' rhythm, doing things that look good in the short term but are useless in the long term to boost the next round's valuation. Liang Wenfeng has not made this mistake so far because he did not need investors before. Now that he is starting to take money, whether he can maintain this independence is the most noteworthy aspect of this deal.

Is $45 Billion Expensive?

Finally, to answer the question at the beginning: Is DeepSeek worth $45 billion?

If you use traditional valuation methods—P/E ratio, P/S ratio, DCF—this figure makes no sense. A company with almost no revenue being valued at 300 billion RMB would make any due diligence professional shake their head.

But AI companies' valuations have never been calculated using traditional methods. OpenAI was valued at $80 billion before it was profitable, and Anthropic was valued at $900 billion before it was profitable. What this industry prices is not today's financial numbers but your position in the future AI landscape. There may be fewer than ten teams globally capable of creating top-tier large models: OpenAI, Anthropic, Google DeepMind, Meta FAIR, xAI, DeepSeek, Zhipu, and Yuezhiyanmian. Having a spot on this list is itself worth hundreds of billions of dollars because once you drop off this list, you'll never get back on.

$45 billion buys DeepSeek's position on this list. As long as it stays on the list, this price is not expensive. But if its core team continues to churn (drain away), the technological breakthroughs of the next-generation model slow down, and the open-source strategy fails to translate into commercial value, $45 billion will become a very heavy figure.