By World Model Workshop
DeepSeek's financing faces another twist.
On April 22, Reuters cited The Information, reporting that Alibaba and Tencent are in talks to invest in DeepSeek, with the valuation expectation raised to over $20 billion.
Just a few earlier, Reuters had reported that DeepSeek was seeking at least $300 million in financing, targeting a valuation of no less than $10 billion.
Within just a few days, DeepSeek's valuation expectation has doubled.
As of the time of writing, DeepSeek has not commented, and Reuters also stated that it has not independently verified all the details.
But the market signals are real.
Having been established for nearly two years, DeepSeek has always relied on funding from its parent company, Huanfang Quantitative, and has never accepted external investment.
Founder Liang Wenfeng is one of the most well-known independents in China's AI circle—not aligning with any camp, not raising funds, not going public.
Now, a crack has emerged.
When Alibaba and Tencent, two tech giants, appear at the negotiating table at the same time, what exactly are they competing for?
Whose money will DeepSeek ultimately take?
DeepSeek's Market Price
To answer what Alibaba and Tencent are competing for, we must first answer another question:
What exactly is the significance of $20 billion valuation in China's large model landscape?
Moonlight Intelligence, which is considering an IPO in Hong Kong, has a latest round valuation of about $18 billion;
MiniMax had an IPO valuation of about $6.5 billion when it went public in Hong Kong in January 2026;
Zhipu had an IPO valuation of about $6.7 billion when it went public in Hong Kong at the end of 2025.
If DeepSeek's round is indeed negotiated at a $20 billion valuation, then it would be among the most expensive tier in China's startup large model companies.
Considering the actual global influence of its V3 and R1 models, this figure seems more like a starting point for negotiation rather than a ceiling.
As the valuation rises, how much will DeepSeek raise?
In the public report on April 17, DeepSeek was originally discussing at least $300 million with at least $10 billion valuation, equivalent to a financing proportion of about 3%.
If calculated similarly, at a $20 billion valuation, selling 3% equity would correspond to about $600 million in financing; 5% would correspond to about $1 billion; 10% would correspond to about $2 billion.
So, is this financing amount large or small for DeepSeek?
$300 million to $600 million is not a small amount for DeepSeek, but it is not at the level of not needing money.
Reuters emphasized in the report:
As cutting-edge models, inference, and agent systems become increasingly capital-intensive, DeepSeek's initiation of its first external financing itself indicates that it is facing higher capital expenditure pressure.
These few billion dollars are more like opening the external financing channel and replenishing a round of ammunition, rather than securing war funds for the next few years in one go.
If it ultimately reaches $800 million to over $1 billion, the nature would be different.
This scale is already a very large financing round among China's startup large model companies.
For comparison, MiniMax raised about $619 million in its Hong Kong IPO.
If DeepSeek's single financing round approaches or exceeds this level, it would be enough to show that the market already regards it as a top-tier scarce asset among China's startup large models.
Regardless of the final range, the money itself is not the goal.
For Liang Wenfeng, what truly matters are two things:
Providing a market-based pricing for the core employees' stock options to stop the crisis of core talent loss;
Exchanging for non-financial resources that pure financial investment cannot fully provide, such as computing power discounts, cloud customer channels, and supply chain synergies, to support the rising capital demands in the competition of inference models and agents.
Alibaba and Tencent's Entry
No matter how high the valuation, it cannot stop the continuous stream of investment institutions wanting to contact Liang Wenfeng.
DeepSeek remains one of the most difficult projects to invest in within China's AI investment circle.
An investor close to DeepSeek commented: This is not a target where you can get in just because you can afford the price. In Liang Wenfeng's screening criteria, money is the least important factor.
According to media reports, DeepSeek has previously rejected investment proposals from China's top venture capital institutions and tech giants.
In the eyes of the capital market, DeepSeek has always been a sought-after scarce asset. This is why even Alibaba and Tencent want to squeeze into the negotiating table.
The answer does not lie in DeepSeek needing them, but in them needing DeepSeek.
China's large model industry is entering a stage of ecosystem warfare, entry point warfare, and resource warfare. Both Alibaba and Tencent are unwilling to be absent from the key position of DeepSeek.
If they can invest, DeepSeek can become part of the giant's ecosystem synergy: model capabilities, cloud resources, enterprise customers, and distribution interfaces all have the opportunity to form联动.
But if they fail to invest, or if another force takes the position first, DeepSeek could become a relatively independent yet strong enough variable, weakening the giants' control in the AI era in turn.
This type of investment has both offensive implications and obvious defensive characteristics.
But the deeper anxiety comes from the countdown of scarcity.
Zhipu and MiniMax have already gone public, Moonlight Intelligence is sprinting towards a Hong Kong IPO, and model manufacturers like StepFun and Baichuan Intelligence are also accelerating their capitalization.
DeepSeek is the only head general-purpose large model among the "Six Tigers" that has never accepted external financing.
Missing it means missing the last ticket to board in this cycle.
Once it accepts other capital or goes public, Alibaba and Tencent will lose the opportunity to use shareholder status to exchange for premium cloud customers.
For two giants with a combined market capitalization of over $800 billion, an investment of a few billion dollars is just a fraction of their annual capital expenditure.
But if the opponent secures DeepSeek, what will be lost is the ecosystem dominance of an era.
Will DeepSeek Take the Giants' Money?
The anxiety of Alibaba and Tencent is real, but DeepSeek's sense of urgency cannot be ignored either.
Liang Wenfeng's stance is clearly changing: from long-term rejection of external funds to opening a window for external financing for the first time.
Behind this may not be just a funding issue, but rather the simultaneous pressure of time, talent, and infrastructure beginning to approach.
First, there is model iteration pressure. The flagship model V4 has been delayed multiple times and has not yet been released;
Second, there is the talent defense battle. Without external valuation, the incentive of stock options for core talent will be significantly discounted;
Finally, there is the heavy burden of infrastructure. Adapting V4 to domestic chips requires substantial engineering resources, which Huanfang alone cannot support.
But Liang Wenfeng's negotiating posture is still opening a crack in the door, not begging to enter.
The $20 billion valuation expectation, simultaneous contact with two giants, and delayed finalization of details—this ambiguity itself shows that he is waiting for an optimal solution.
This also means that what Liang Wenfeng cares about is not whether to accept investment from Alibaba or Tencent, but rather which kind of money to accept, and to what extent.
The first possibility is the most ideal and also the scarcest: take only financial investment, without giving up directional control.
That is to say, the giants can take a stake, but only as minority shareholders, without deep business binding, without requiring exclusive cooperation, without locking DeepSeek into a certain ecosystem in advance.
For DeepSeek, this is the most desired situation. The money is obtained, relationships are established, but the outside world will still regard it as a relatively independent model company.
The advantage of this structure is that it both replenishes ammunition and minimizes damage to the independence that is most valuable in the valuation.
The second possibility is more realistic and also more dangerous:表面上拿的是财务投资，实际上拿到的是资源绑定 (On the surface, it's financial investment, but in reality, it's resource binding).
Because the giants' money is rarely just money; it usually also comes with cloud resources, customer channels, interface capabilities, distribution entry points, and future cooperation expectations.
In the short term, this binding is very tempting.
The problem is, once this cooperation deepens, the market will quickly redefine DeepSeek. Is it still an independent model company, or has it already become a key piece in a giant's AI puzzle?
DeepSeek certainly knows what the giants' money can bring: faster resource对接, greater commercialization imagination, a thicker safety net.
But it also knows what the giants' money will take away: the outside world's imagination of its independence.
This is the most difficult part of this financing round.
Once investment is accepted, DeepSeek's previously undefined state will begin to change.
Who is the Better Choice?
If the money from Alibaba and Tencent both carries the risk of ecosystem lock-in, is there a better choice in the market?
In theory, state-owned or national team capital that brings resources but does not exert strong control is closer to this balance point.
The advantages of state-owned capital are obvious.
Large scale of funds, long cycle, not pursuing short-term exits;
Low willingness to intervene in technical routes, more tolerant of DeepSeek's open-source strategy;
It can also bring computing power support, industrial synergy, government project resources, and stronger policy coordination capabilities.
But the disadvantages of state-owned capital are equally obvious.
Long decision-making processes, high compliance requirements, may not keep up with DeepSeek's technical schedule.
More subtly, the national team label may inversely limit its international layout, causing overseas developers and enterprise customers to have doubts about its data security and neutrality.
In the current public information, there is no clear report indicating that state-owned funds are contacting DeepSeek.
But in February 2025, Reuters disclosed that state-background funds such as China Investment Corporation and the National Social Security Fund had expressed interest in investing in DeepSeek.
This clue indicates that the national team is not without willingness.
Then, what about pure financial investors?
Top institutions like Sequoia China and Hillhouse indeed have money and professionalism, but the real shortcoming of pure VCs lies in resources.
DeepSeek does not lack money; it lacks the infrastructure and industrial resources that should arrive together with the money.
According to industry insiders, over the past few months, many top VCs have tried to contact Liang Wenfeng, and most did not even get the chance for formal due diligence.
Taking VC money, DeepSeek would still face commercialization requirements, which actually deviates from Liang Wenfeng's original intention of "doing research" when founding DeepSeek.
In the end, this is not a simple problem of choosing money, but choosing an identity.
Taking the giants' money, DeepSeek may quickly obtain resources, accelerate commercialization, and survive the fierce competition, but the price is the gradual erosion of independence.
Taking the national team's money, it may maintain technical route freedom and open-source ideals, but it may also, due to insufficient industrial synergy, be half a step slow in the commercialization competition, eventually becoming a national strategic asset.
Two kinds of money, two destinies.
There is no perfect answer, only a choice after weighing.
Window is Counting Down
Given DeepSeek's technical reputation and current market heat, it can always find willing buyers.
But this seller's position has an expiration date.
In the next 6 to 12 months, manufacturers like StepFun and Moonlight Intelligence may successively land on the capital market. The scarcity of unlisted head general-purpose large models is rapidly diluting.
Once these targets are all securitized, capital's gaze will turn to the secondary market, and the premium space in the private market will be compressed.
As more global head AI companies advance financing or other capitalization actions, the pricing anchor of global AI assets may continue to change, and the valuation system of domestic related companies also faces recalibration pressure.
$20 billion is an exciting starting point today, but it may be a price that needs to be defended tomorrow.
Liang Wenfeng's crack has appeared, but the door is only open a crack.
The countdown for this decision—how wide this door will eventually open and who will be let in first—has already started.
