$45 billion valuation, Liang Wenfeng "descending the mountain"

None

Text | Node Finance, Author | Zero Degree

$45 billion.

In early May 2026, this number struck like a heavy hammer, breaking nearly three years of stagnation in the venture capital circle.

Two weeks ago, there were rumors that Alibaba and Tencent planned to spend $20 billion to renew their contracts with DeepSeek. Unexpectedly, the trend quickly shifted. According to sources from the Financial Times, the National Integrated Circuit Industry Investment Fund (National Big Fund) is planning to take the lead in this round of financing, with the valuation directly set at $45 billion.

In half a month, the valuation doubled.

Today, as computing power becomes tight, the funding winter persists, and even AI downturn rumors are nearly exhausted, DeepSeek has declared who truly is the “top dog” in China’s AI circle with a single act. However, Node Finance believes that behind this is not just a few giants playing with numbers on paper, but a complete overhaul of China’s AI underlying logic.

Liang Wenfeng’s “Down the Mountain”

To discuss DeepSeek, one cannot avoid the man named Liang Wenfeng.

For a long time, Liang Wenfeng was like a hermit in the circle. Relying on Fantom Quantitative, this “cash cow,” his early attitude was as firm as a rock: no external funding, no flashy press conferences, and he hardly paid attention to investors.

This confidence of “I have money, so I focus on technology” allowed DeepSeek to avoid impatience, relying solely on algorithm optimization to carve out a path internationally.

So, why did Liang Wenfeng, who had always refused funding, choose to embrace capital in 2026?

Node Finance believes one direct reason is that AI has shifted from a “brainpower activity” to a “heavy industry.”

Early DeepSeek relied on intuition and used exquisite algorithms to dance among Nvidia chips. But by 2026, competition among large models had entered the MasterAgent (L4-level intelligent agent) era, which is not just a contest of algorithms but also a battle over electricity, computing power, and even national-level resource scheduling rights. Fantom, no matter how wealthy, is just a top-tier “private workshop,” unable to provide DeepSeek with the “pass” to influence the entire semiconductor industry chain.

Liang Wenfeng’s “Down the Mountain” is essentially a strategic compromise. He realizes that in the current international environment, it’s no longer possible to be a “recluse.” Not participating means suffocation under the iron curtain of computing power blockade.

An even more important reason is talent.

Recently, core technical backbones like Guo Daya (who transferred from ByteDance) and Luo Fuli (who transferred from Xiaomi) have left one after another. This caused quite a stir in the circle, with some even thinking DeepSeek is “running out of talent.”

But if you look at the author list of DeepSeek V4’s release, the data is quite solid: its core research engineering team is about 270 people, with only around 10 leaving during R&D. Less than 4% turnover, which is a miracle for any large model company—remember, OpenAI’s core talent loss rate in the first two years exceeded 25%.

The reason everyone feels the “talent exodus” is fierce is because DeepSeek’s early team was too lean and too brilliant. When any of these “genius teenagers” is poached by a big company with several times the salary and more prominent titles, the outside world feels like an earthquake.

But from another perspective, the competition for talent is indeed intensifying.

Why raise funds now? It’s actually to give talent a “certain price.” Previously, DeepSeek didn’t raise funds, and employees’ stock options were just “paper wealth,” invisible and intangible. Now, with a valuation of $45 billion, everyone suddenly realizes their shares are truly valuable. Faced with high salaries from big companies, employees will weigh: go to the big company for cash, or stay at DeepSeek and wait for this $45 billion valuation to multiply several times?

Another reason is to counteract the “cash capacity” of big companies. Giants like ByteDance, Tencent, and Xiaomi often lure talent with equity and bonuses. Liang Wenfeng needs this round of financing to transform DeepSeek from a “geek lab” into a “super unicorn” with strong capital backing, regaining initiative in the talent market.

Big Fund Rumored to “Enter the Scene”

According to the Financial Times, the National Integrated Circuit Industry Investment Fund (Big Fund) is in talks with DeepSeek about its first round of financing.

The Big Fund stepping in to invest in a large model company is definitely unprecedented in the circle.

So far, the Big Fund has established three phases of funds. The first is “strategic layout,” solving the “whether it exists” problem; the second is “precise risk mitigation,” focusing on solving “whether it can be blocked” issues, mainly tackling upstream core links like equipment and materials. In May 2024, the third phase of the Big Fund was established, with a scale surpassing the first two combined, expanding into AI chips.

Overall, the previous Big Fund’s investments focused on hardware like chip manufacturing, lithography machines, and etching equipment, emphasizing “strengthening the foundation.” Now, suddenly shifting to push DeepSeek to a valuation of $45 billion is actually a strategic resource reallocation.

Node Finance sees that the reason the Big Fund invests in DeepSeek ultimately comes down to three “cards” that can change the game:

First, it is the “chief tuner” of domestic computing power.

The Big Fund has invested in semiconductors for many years, with the biggest concern being that domestic chips are “hard to produce but not easy to use.”

Huawei Ascend, Cambrian, Biren… these domestic AI chips have improved hardware metrics, but in terms of ecology and compatibility, they are still overshadowed by Nvidia. Many model developers prefer CUDA ecosystem for convenience.

DeepSeek is different; it’s a “hard nut.” Its strength lies in extreme algorithm optimization, enabling domestically produced chips, which are somewhat awkward at the software level, to deliver performance far beyond specifications in real-world applications. The Big Fund’s investment in DeepSeek is actually backing a super engine capable of “driving” the domestic semiconductor industry chain. If DeepSeek can run L4-level intelligent agents (MasterAgent) on domestic computing platforms, it’s equivalent to issuing a “pass” for domestic chips to enter the global market.

Second, it is a “troublemaker” that breaks competition.

In the large model field, if you follow OpenAI’s lead by stacking more computing power and pouring in dollars, you will always be a follower.

DeepSeek is valued by the national team because it proves another possibility: achieving bigger results with less money and fewer resources.

This asymmetric “algorithm-for-compute” approach aligns perfectly with our current survival logic under extreme pressure. The Big Fund’s lead in financing aims to establish this “low-power, high-intelligence” technical path as the mainstream of domestic AI, avoiding ineffective internal competition in the red ocean of computing power.

Third, it is the “final piece” of the ecological closed loop.

Previously, China’s AI industry was fragmented: Big Fund invested in chips, Alibaba and Tencent developed applications, startups built models.

This time, with the Big Fund leading and internet giants following, it’s like extending a hand to connect scattered pearls into a necklace.

At the bottom: Big Fund guarantees the priority of domestic chips.

In the middle: DeepSeek provides a full-stack self-developed core brain.

At the top: internet giants supply data and find outlets for models in scenarios.

This “state team guaranteed + geek leadership + giant implementation” combo is what the Big Fund truly hopes to see—a secure full industry chain.

Therefore, what the Big Fund invests in is not just a chatbot, but a soft- and hardware integrated commander that can force China’s AI to carve out an oasis in the “desert of computing power.”

The $45 billion valuation buys an opportunity to truly “revive” the domestic semiconductor industry. In today’s game, this opportunity is indeed an invaluable “treasure.”

Behind the $45 billion: not just US-China rivalry, but a “surgical stitching” of the ecosystem

Many interpret DeepSeek’s rise as a “battle” between China and the US. But Node Finance thinks this view is somewhat superficial.

In 2026, the US holds hegemonic control over B200 chips, trying to trap Chinese AI in a “resource-poor zone.” DeepSeek’s positioning is precisely China’s strongest response. The Big Fund’s leading role in this round of financing is not about DeepSeek writing poetry, but about its capability.

DeepSeek is completing an unprecedented “ecosystem stitching”:

Upward, it “nurtures” computing power: During the critical period when domestic chips are still struggling, DeepSeek uses top-tier algorithm optimization to force domestic chips to deliver unexpectedly strong real-world performance. If top-tier cards are unavailable, it uses top-tier logic to squeeze every residual value from domestic chips. This is not just cost-saving; it’s a “lifeline” for China’s domestic computing industry.

Downward, it “builds” a closed loop: Previously, China’s AI ecosystem was fragmented—chips are chips, models are models, applications are applications. Now, with the Big Fund leading investment and internet giants following, a full-chain integration is achieved: Big Fund guarantees the base, DeepSeek provides the brain, and internet giants supply scenarios and data.

This approach creates a unified, battle-ready “command center” from previously scattered domestic resources.

Within this $45 billion framework, there is a subtle wisdom.

Node Finance believes that the participation of internet giants is no longer just “land grabbing,” but more of an ecological complementarity. Big companies have hundreds of millions of user entrances and enterprise scenarios, which are the most needed practical training grounds for a “geek lab” like DeepSeek. Meanwhile, the entry of the national team injects a strong confidence boost into this industry full of uncertainties.

For Liang Wenfeng, this round of financing is a “coming of age.” He bids farewell to the pure laboratory era and steps into the most complex arena of interests. This shift may lack some of the geek’s aloofness, but it gains resilience for survival amid great power competition.

Is $45 billion expensive? If it’s just for a model, yes, it’s exorbitant; but if it’s for an AI ecosystem capable of self-evolution under extreme blockade, this price actually embodies a future national destiny.

In the storm of 2026, DeepSeek has already secured the heaviest chip. Whether it can truly become that “Great Wall of Digital China” to counter Silicon Valley depends not only on its algorithm depth but also on its resource integration ability amid this capital wave.

After all, in the game called “infinite competition” of AI, surviving and becoming irreplaceable is the only true goal.