DeepSeek's new funding round approaches 100 billion yuan, plans IPO for early 2027

[[**DeepSeek's new funding round approaches 100 billion yuan, with a valuation of about 500 billion yuan and plans for an IPO in early 2027. Tencent and CATL led the investment, with the latter aiming for computing-power and electricity coordination, pulling large-model competition onto the battlefield of energy efficiency.**

**Key points**

**1. Subscription demand doubled, paving the way for an IPO**

The original fundraising target was 50 billion yuan, but actual subscriptions exceeded 80 billion, and the term sheet points to nearly 100 billion yuan. The post-money valuation is about 500 billion yuan, and the deal is expected to close within October, after which the corporate structure will be reorganized in preparation for an IPO in early 2027.

**2. Liang Wenfeng maintains firm control of the company through the structure**

All external investors hold indirect stakes through limited partnerships, enjoying only dividend rights and preemptive subscription rights, with no voting rights and no board seats, plus a five-year lock-up period. Liang Wenfeng controls a total of about 84% of the equity and almost all voting rights.

**3. CATL makes a heavy bet on computing-power and electricity coordination**

It has invested in DeepSeek for two consecutive rounds, with a cumulative total of more than 10.5 billion yuan. At the same time, it has taken a stake in VNET and increased capital in Zhongheng Electric, integrating green power, energy storage, and power distribution capabilities to provide a stable, low-electricity-price solution for DeepSeek's gigawatt-scale data center in Ulanqab, reducing the 60%-70% electricity cost in AI operations.

**4. The divergence between Chinese and U.S. large-model paths is obvious**

In the United States, Anthropic and OpenAI rely on debt and sky-high valuations to support computing expansion, with subscription models heavily subsidized, pushing revenue per megawatt down by about $36 million. DeepSeek is tied to industrial capital and domestic chips, taking a low-cost route of integrating the energy chain.

**5. China's AI competition enters the "water, electricity, coal" integration stage**

In the second half of the large-model race, what is being contested is not model accuracy but the ability to link green power, domestic chips, and the data center chain into a closed loop. Every turn of a wind turbine in Ulanqab turns green power into billions of tokens; whoever controls this conversion chain will win the right to speak in the next decade. 2026-10-08 17:51 DeepSeek stands at the door of an IPO Old Shu Quick View

This article comes from the WeChat public account: Old Shu's Notes, author: Old Shu, original title: "DeepSeek stands at the door of an IPO"

In the past two days, several media outlets have released news that has left the entire venture capital circle unable to sit still.

According to Bloomberg, DeepSeek's new funding round is close to being finalized, with a scale of at least 80 billion yuan, and according to the signed investment term sheet, the final figure may approach 100 billion.

Moreover, DeepSeek's initial fundraising target was only 50 billion, but amid the booming industry, investors pushed subscriptions to twice the original target. Notably, among the investor list, the two familiar names that committed the most capital were Tencent and CATL.

And once this money is settled, DeepSeek's next step is to restructure the company's architecture, paving the way for an IPO in early 2027.

After all, just over a year ago, DeepSeek was still publicly denying rumors, saying that the funding news circulating in the market was all rumors, and founder Liang Wenfeng himself repeatedly said he did not lack money.

Now, from financial investors to industrial capital, those who want to give money seem to be lining up again.

A 50 billion pot cannot hold 100 billion in money

The ins and outs of this funding round are worth sorting through from the beginning.

In June this year, DeepSeek had just completed the first round of external funding since its founding, with a scale of about 51 billion yuan and a post-money valuation of about 350 billion yuan.

In fact, the list of contributors in that round was already a bit unusual. Liang Wenfeng himself put in about 20 billion, Tencent contributed 10 billion to become the largest external investor, the CATL system followed with about 5 billion, JD.com and IDG each put in about 3 billion, and the National Artificial Intelligence Industry Investment Fund also joined.

A month later, DeepSeek's second funding round, launched in July this year, pushed the target valuation to about 500 billion yuan. The original plan was to raise 50 billion, but subscriptions surged all the way to more than 80 billion, and the signed term sheet also points to 100 billion. According to people familiar with the matter cited by Bloomberg, the deal is expected to be completed within October.

Why are capitalists scrambling like this? The report card delivered by "Liang Sheng" deserves much of the credit.

On September 10, DeepSeek released V4.1-Flash, a 55.2-billion-parameter MoE architecture, scoring 90.9 on the GPQA Diamond evaluation. A Bloomberg Intelligence research report shows that after the model was launched, the benchmark gap between top Chinese and U.S. models has narrowed to about 3%, while at the beginning of the year that figure was still 15%.

DeepSeek's revenue side is also exploding. According to a late-September report by The Information, DeepSeek's annualized revenue has reached about $1 billion, while a few months earlier that figure was less than $500 million.

What is even more rare is that despite the influx of external investors, Liang Wenfeng's control over the company remains solid.

According to disclosures by several media outlets, all external investors will hold indirect stakes in DeepSeek through limited partnerships, receiving dividend rights and preemptive subscription rights, with no operating voting rights, no board seats, and a five-year lock-up period, while Liang Wenfeng himself controls a total of about 84% of the equity and almost all voting rights.

In the author's view, this structural design is an important reason why Tencent and others are willing to line up to enter.

Capital has come in, but the helm of the ship remains firmly in the founder's hands, consistent with Tencent's long-standing financial-investment approach. In addition, binding the most capable domestic large model into its own ecosystem gives WeChat's entry point with over a billion users the kind of engine it needs.

CATL Enters the Fray, Aiming at Compute-Power Synergy

Interestingly, in DeepSeek's funding list, the most intriguing name is CATL.

Why would a battery manufacturer make two consecutive rounds of heavy bets on a large-model company? The answer lies in Ulanqab, Inner Mongolia.

According to media reports, DeepSeek is building a gigawatt-scale data center in Ulanqab, with power equivalent to a nuclear generating unit, and plans to deploy at least 160,000 Huawei Ascend 950DT chips. The center is expected to gradually come into operation between the end of 2027 and early 2028, and once completed it will be one of the largest known Huawei AI chip clusters.

If such a data center runs at full load 24 hours a day, the electricity burden is easy to imagine.

A research report by Changjiang Securities shows that electricity accounts for as much as 60% to 70% of the operating costs of AI large models. Gao Wen, an academician of the Chinese Academy of Engineering, also did the math at this year's World Artificial Intelligence Conference: for computing power to achieve long-term sustainable development, electricity costs should not exceed 0.3 yuan per kWh.

Against this backdrop, whoever can push electricity costs down holds the largest cost component of large models in their grip. Thus, many of CATL's moves this year have revolved almost entirely around compute-power synergy.

In May this year, CATL spent about 6.4 billion yuan to take a stake in Vianet, becoming the largest shareholder of this veteran IDC enterprise and taking control of more than 50 data centers across more than 30 cities.

In August, CATL put in another 4.1 billion yuan to increase capital in the parent company of Zhongheng Electric. As a leading enterprise in high-voltage direct-current power supply for data centers, Zhongheng Electric has a domestic market share of about 31%, ranking first in the industry.

Coupled with two consecutive rounds of funding for DeepSeek, CATL's cumulative investment along the compute-power synergy line has exceeded 10.5 billion yuan.

Green power, energy storage, power distribution, machine rooms—CATL has a presence in every link of this industrial chain. What it wants to sell to DeepSeek is not just battery cells, but a full set of solutions to keep a gigawatt-scale data center running steadily beside a wind farm.

US Peers: Trading Debt for Time

Turning the large-model lens to the United States across the ocean, one can see a set of contrasts entirely different from DeepSeek.

At present, Anthropic is sprinting toward an IPO before Thanksgiving at a valuation of up to $2 trillion, carrying huge losses on its books and relying on up to $42 billion in loans from Broadcom and $60 billion in chip financing from a banking syndicate to support its computing expansion.

As for OpenAI, while speeding up GPT-6 by 50%, it is quietly tightening subscription quotas, and users have even found that its subscription quota is only one-fifth that of Claude.

Clearly, current paid subscriptions for large models are still far from a sound business model. Compared with enormous computing consumption, users' spending is a drop in the bucket for large-model companies.

SemiAnalysis once did the math and concluded that large-model subscriptions are a highly subsidized product. Taking Anthropic as an example, its subscription revenue accounts for only about 10% of total revenue but may consume more than 40% of inference computing power, dragging down Anthropic's blended revenue per megawatt by about $36 million.

This is perhaps the fundamental reason why leading large-model companies guard against users all day long, especially non-core users, and frequently ban accounts. The playbook of US leaders is clear: trade debt and sky-high valuations for time, and wait until the computing-power moat is built first.

Compared with OpenAI and its peers, DeepSeek is taking another path.

It does not borrow money; it takes money from industrial capital and binds itself to two trump cards: power and domestic chips.

Other players on the same track are similar. Moonshot AI has just completed the final round of private financing at a valuation of about $50 billion. In August, when ARR broke through $1 billion, Moonshot AI set its listing for the first quarter of 2027, choosing Hong Kong as the venue.

Chinese large models are collectively heading toward the capital market, and they are also pulling battery makers, power grids, and domestic chip makers into the game. The real signal of DeepSeek's 80 billion yuan financing is a declaration that in the second half of China's AI competition, what matters is not whose model is smarter, but who can integrate "water, electricity, and coal" into a chain.

Every time a wind turbine in Ulanqab turns, a kilowatt-hour of green power flows into the machine room and becomes billions of tokens. Power becomes computing power, and computing power becomes intelligence. Whoever can stand firm on this conversion chain will hold the power of discourse in the next decade.

From this perspective, DeepSeek has pulled the battle over large models from the dimension of cutting-edge technological capability toward competition over ways of energy utilization. In the past, almost every rewriting of the energy structure rewrote the coordinates of civilization—so it was with the steam engine, so it was with the power grid, and today's computing power will be no exception. Channel: Finance and Economics. This content is published with the author's authorization. The views represent only the author and do not represent the position of Huxiu. If you have objections or complaints about this manuscript, please contact tougao@huxiu.com. Everyone who is changing and wants to change the world is on the Huxiu APP[ ]]]