AI Trends

DeepSeek’s RMB 100 Billion Financing Completed: How Perfect World, Hengdian, and Huace Got Involved

36Kr · AIUpdated Nov 19 21:08

AI summary

DeepSeek’s second financing round has been formally registered with the corporate authorities. Citing publicly reported figures, the article says it raised approximately RMB 105.5 billion at a post-money valuation of about RMB 615.5 billion. Tencent, NetEase, Xiaohongshu, Hengdian, Bilibili, Perfect World, and Huace participated through different ownership structures. The article further analyzes the potential value of these entertainment companies’ content data, production scenarios, and distribution channels for multimodal development. Except for the National Artificial Intelligence Industry Investment Fund, funds from external investors were channeled into limited partnership vehicles controlled by Liang Wenfeng, carrying no voting rights and subject to a five-year lock-up period.

Why read this

SelectedAI score 76

The article breaks down the ownership structures through which entertainment companies participated in DeepSeek’s financing and analyzes the potential value of collaboration involving content data and film and television production scenarios.

Topics

Tags

Body · AI translation

On November 17, Hangzhou DeepSeek completed a business registration change. According to Qichacha, it added Hangzhou Chengyan Enterprise Management Consulting Partnership as a shareholder, marking the formal registration of its second financing round.

Based on figures from public reports, Entertainment Capital has found that the round raised approximately RMB 105.5 billion (US$15.8 billion), giving the company a post-money valuation of about RMB 615.5 billion (US$86 billion) and setting a new record for a single AI financing round in China. Chengyan, which serves as the shareholding platform, added 31 partners at once, with registered capital contributions totaling RMB 3.439 billion. It now has 32 partners and holds a 17.26% stake in DeepSeek.

Among the 32 names are industrial companies and state-backed entities such as CATL, China Mobile, CITIC Securities, China Life, SMIC, ChangXin Memory Technologies, Midea, and vivo. Also included are six names from the entertainment and content-platform sectors: Tencent, NetEase, Xiaohongshu, Hengdian, Bilibili, and Perfect World, as well as Huace through a more circuitous route.

The inclusion of midsize entertainment companies such as Perfect World, Hengdian, and Huace alongside so many industry heavyweights is itself a signal worth noting. Could DeepSeek be planning to enter AI video and pursue a multimodal path?

Three tiers of admission prices

The capital contribution percentages of Chengyan’s 32 partners in the second round add up to exactly 100%. By multiplying each percentage by the RMB 105.5 billion raised in this round, Entertainment Capital calculated how much each participant actually contributed: Tencent’s 9.5653% corresponds to approximately RMB 10.09 billion, in line with the RMB 10 billion cited in public reports; an entity affiliated with Lisi Capital’s 7.1740% corresponds to about RMB 7.57 billion; and Boyu’s 3.3479% corresponds to approximately RMB 3.53 billion.

By this measure, the admission prices for entertainment companies fall into three tiers.

The first tier consists of four companies that invested directly through entities they established. Tencent holds 9.5653% of Chengyan through Shanghai Hengxiu, equivalent to approximately RMB 10.1 billion; NetEase holds 2.3913% through Hangzhou Leihuo Wanxiang, equivalent to about RMB 2.5 billion; Hengdian Bohui Venture Capital Fund holds 1.4348%, equivalent to approximately RMB 1.5 billion; and Xiaohongshu holds 0.9565% through Xingyin Information Technology, equivalent to about RMB 1 billion.

The second tier consists of two companies that invested through Lisi Xingshen. Bilibili holds 6.6489% of Lisi Xingshen through Shanghai Huijie Culture, translating into a 0.4770% indirect interest in Chengyan and an investment of approximately RMB 500 million.

The third tier consists of Perfect World and Huace. Like Bilibili, Perfect World holds 0.6649% of Lisi Xingshen, translating into a 0.0477% indirect interest in Chengyan and an investment of approximately RMB 50 million. Huace, meanwhile, holds 49.99% of Jiaxing Puhua Lichuan, which invested for a 2.55% stake in Hangzhou Shixi Zhaoyi No. 1. Shixi Zhaoyi No. 1 then acquired a 3.3479% interest in Chengyan. Tracing through all three layers, Huace’s investment also works out to approximately RMB 50 million.

Tencent contributed the most of all 32 partners, investing RMB 10 billion in each of the two rounds for a cumulative RMB 20 billion. NetEase invested RMB 3 billion in the first round and RMB 2.5 billion in this round, for a total of approximately RMB 5.5 billion.

The significance of Hengdian’s RMB 1.5 billion investment must be viewed in the context of its own financial resources. Hengdian Bohui expanded the fund’s size to RMB 1.65 billion only in October, then immediately put RMB 1.5 billion of it into DeepSeek alone. Within the fund, however, Hengdian-affiliated parties account for 33.32% of Hengdian Bohui’s capital contributions, or about RMB 550 million. Including the personal contribution of Hengdian Group executive Xu Wencai brings the figure to 35.99%, or approximately RMB 594 million. The remainder comes from other Zhejiang investors, including Youngor, Wynca Chemical, and Robam Appliances.

Why these companies?

DeepSeek is not short of money. The second round was originally intended to raise RMB 50 billion, but the target was later more than doubled to RMB 105.5 billion, reportedly because Nvidia chips had become more expensive and the company wanted to reserve ammunition for an industry bubble.

But as everyone knows, many companies are eager to give it money. DeepSeek is the one choosing its shareholders, and what it is selecting is not money.

It is selecting use cases.

By 2026, large language models have reached the point where the ceiling for pure text is already visible. The real battleground is multimodality: video generation, image understanding, and long-form content generation. Compared with text, these capabilities depend even more heavily on three things: high-quality training data, real production environments, and distribution channels.

Entertainment companies happen to possess all three.

Hengdian has filming facilities and production capacity. Not long ago, it also spent RMB 250 million on a partnership with Volcano Engine to bring all of its locations and background-performer data into AI systems. Every stage of film and television production—storyboarding, previsualization, visual effects, dubbing, and editing—is both a use case for multimodal models and a source of labeled data. The RMB 1.5 billion invested by Bohui did not merely purchase a financial interest; it bought a ticket to embed AI into film and television production lines.

Bilibili has a content community and creator ecosystem, and it invested more than any other entertainment participant in this round apart from Tencent and NetEase. The danmu comments, derivative creations, and uploader materials found on video platforms are precisely the kinds of data that video models want most.

Perfect World has game-engine and film production capabilities. Its RMB 50 million investment looks more like a move to reserve a seat: it is also promoting the use of AI to generate game assets, and what it needs is priority access to models rather than returns on equity.

Huace may be the smallest company among them, with an actual investment of approximately RMB 50 million. It has made frequent AI-related moves over the past two years, investing successively in Shuimu Intelligence and Zhiling Xinjing. Its consistent strategy has been to spend modest amounts in exchange for access to technology.

DeepSeek’s multimodal ambitions

Including entertainment companies on its shareholder list may also align with DeepSeek’s technical trajectory over the past year.

In April 2026, DeepSeek released V4-Pro and V4-Flash, both open-weight models under the MIT License supporting a context window of 1 million tokens. An updated version of V4-Flash followed in July, focusing on improved agentic and coding capabilities, while its API entered public beta. The models’ capabilities are moving from “being able to write” toward “being able to act”; the next step will inevitably be “being able to see and generate.”

What this path has always lacked is not computing power, but data and use cases. CATL provides industrial-vision requirements from battery production lines; China Mobile and China Telecom provide network and computing infrastructure; Meituan and others provide scheduling use cases in local services and travel; and JD.com provides retail and supply-chain scenarios. Entertainment companies provide the entire content-production workflow—the part that is hardest for multimodal models to develop on their own.

In other words, a company that only works on text does not need Hengdian, but a company seeking to develop video generation would still want to connect with Hengdian.

This explains a seemingly counterintuitive phenomenon: amid a group of industrial giants worth hundreds of billions of yuan, several film and television companies managed to secure seats. It is not because they invested more money, but because DeepSeek may need their use cases in the future.

This is not an equal relationship

It should be made clear that these entertainment companies have obtained only a financial admission ticket.

This financing round retains the special structure used in the first round: except for the National Artificial Intelligence Industry Investment Fund, none of the external investors’ money is injected directly into DeepSeek. Instead, it goes into limited partnership platforms controlled by Liang Wenfeng, carries no voting rights, and is subject to a five-year lockup period.

This means that Tencent with its RMB 10.1 billion, Hengdian with its RMB 1.5 billion, and Perfect World with its RMB 50 million all carry exactly the same weight in DeepSeek’s governance structure: they provide money, have no say, and cannot sell for five years.

For entertainment companies, the return on this money lies somewhere other than the equity. Hengdian wants the capacity narrative of an “AI film and television production base”; Bilibili and Xiaohongshu want priority access to model capabilities; and Perfect World and Huace want lower content-production costs. Equity appreciation is secondary; getting aboard is what matters most.

Across the two rounds, DeepSeek raised more than RMB 150 billion in less than four months. Reports say the company has hired CITIC Securities to prepare for a listing on the STAR Market, with the process expected to begin within the year.

This article is from the WeChat public account “Entertainment Capital” (ID: yulezibenlun), written by Congnao and published by 36Kr with authorization.

Source

36Kr · AI