DeepSeek's funding round expands to 105.5 billion yuan, with nearly 30 investors

Financing of 105.5 billion yuan! DeepSeek expands its AI hardware circle of friends: ecosystem first, models second. By Zhong Jia, 11 hours ago. It's financing, and also team-building.

Recently, foreign media outlet The Information broke a major piece of news in China's AI circle: DeepSeek's latest funding round has expanded to 105.5 billion yuan, with a valuation directly reaching 615.5 billion yuan.

Beyond the record-breaking funding amount, the report also disclosed a long list of investors. In addition to CATL and China Mobile, it also brought together nearly 30 companies including ChangXin Memory Technologies, SMIC, vivo, Midea, Luxshare Precision, Geely, Meituan, and Xiaohongshu.

Judging from the list, they are not simply making financial investments. These companies cover fields including domestic semiconductors, communications, consumer electronics, smart home appliances, advanced manufacturing, and internet applications—all industries that can connect with AI. In the view of Leitech (ID: leitech), the 105.5 billion yuan in investment that DeepSeek has brought in is not only for buying computing power and stacking parameters, but also a layout for the future of AI—a big game of connecting the software and hardware implementation of AI. Backed by SMIC, ChangXin, China Mobile, and China Telecom, computing infrastructure is not a concern.

Among DeepSeek's key allies will be chip manufacturers and telecom operators. This is not surprising. Supporting the massive computing power of DeepSeek's large models is inseparable from chips and backbone networks.

At present, domestic AI brands generally face a common problem: the prices of Nvidia's high-end chips continue to rise, and due to geopolitical influences, supply to China is restricted, creating huge uncertainty for domestic AI platforms. If DeepSeek wants to maximize model capabilities stably over the long term, it must seek diversified supply solutions in computing infrastructure.

At an internal meeting in July this year, Liang Wenfeng mentioned his views on AI chips. He believed that Nvidia chip prices rise year after year, customers will be forced to seek domestic alternatives, domestic chips can already achieve replacements, and there are no problems with hardware or the ecosystem—the only problem is insufficient production capacity. SMIC represents the strongest force in domestic wafer foundry and is a key link in the domestic AI chip manufacturing chain.

ChangXin Memory Technologies is a leader in the domestic memory industry. At present, a small number of international memory giants dominate the prices of memory, especially HBM, under the AI wave. ChangXin's role in this is, on the one hand, to increase supply and provide ammunition for domestic AI; on the other hand, it competes with international memory giants to curb disorderly memory price increases.

(Image source: ChangXin)

The training and inference of AI models, to a large extent, is a contest of computing power and video memory. Bringing in these two giants can be seen as DeepSeek paving the way in advance for future domestic chip adaptation and customization.

This round's funding list also includes two telecom giants, China Mobile and China Telecom, which control the national backbone network and massive IDC data centers. With their support, DeepSeek has eliminated worries in data transmission and cloud inference. Joining hands with vivo and Midea: seizing terminal entry points and doubling down on AI hardware.

No matter how impressive large models are in the cloud, they ultimately have to land on terminal devices that ordinary users and families are willing to pay for. In DeepSeek's investor list for this round, vivo and Midea are two highly representative brands.

First, mobile phones. In today's phone industry, everyone is competing over AI. vivo laid out its phone AI strategy very early and is currently investing substantial resources. vivo's AI solution for phones is the relatively common combination of device and cloud. vivo's self-developed BlueLM large models, covering different parameter scales, are mainly responsible for local on-device tasks; more complex tasks are handed over to cloud models.

As early as February 2025, vivo announced a deep partnership with DeepSeek, integrating the full version of DeepSeek-R1 directly into the OriginOS system, adding deep thinking capabilities to the built-in AI assistant BlueLM Xiao V. When facing complex logical reasoning, long-text creation, or deep online search, vivo phones will not simply rely on their own cloud computing power, but can also directly call on DeepSeek's capabilities.

(Image source: vivo)

In addition to underlying logical reasoning, AI imaging is also a major focus. Frankly speaking, today's AI imaging is no longer just skin smoothing and beauty enhancement. Large models' semantic understanding of images, generative filling, and frame-by-frame rendering of real-time video are likely to be future competitive focal points. By bringing in DeepSeek's powerful multimodal and reasoning capabilities, vivo can further optimize the algorithms of its self-developed imaging chips, and the entire process from pressing the shutter to producing the final image on a phone may be overturned by AI.

vivo's spending to enter the game clearly shows that it values the inference efficiency and capabilities of DeepSeek's models. For DeepSeek, partnering with a leading phone brand like vivo can also give it access to smartphones, the C-end entry point with the strongest user stickiness, allowing its AI capabilities to land in rich scenarios.

Turning to home appliances, at a Midea air conditioner launch event in March 2025, Leitech (ID: leitech) experienced on site the Midea Fresh Air Machine T6 equipped with large models such as DeepSeek. It can automatically sense multidimensional environmental parameters such as weather, temperature, and air quality with one click, and autonomously adjust air parameters including temperature, humidity, airflow, purification, and freshness, achieving truly intelligent air management.

In the past, smart homes were, to put it nicely, voice-controlled; to put it bluntly, they were voice-activated remote controls, and users had to issue precise commands. But after integrating DeepSeek, home appliances have brains too. Users only need to state their needs in natural language, and the Agent connected to the appliances can automatically adjust device parameters based on the user's body surface temperature, environmental parameters, historical preferences, and more.

For DeepSeek, home appliances are likewise an important entry point covering a huge consumer group, forming a differentiation from the phone scenario. At the same time, if DeepSeek wants its AI capabilities and C-end scenarios to break out of web-based dialogue and integrate into ordinary users' daily lives, it must continuously strengthen its presence in consumer electronics such as phones and home appliances. From this perspective, DeepSeek's deep collaboration with vivo and Midea is a key step in connecting the AI software and hardware ecosystem. Luxshare, Meituan, Xiaohongshu: AI will eventually go to the real world to level up

There are also several companies on DeepSeek's investment list that appear to have little to do with AI large models, such as contract manufacturer Luxshare Precision, and internet service providers Meituan and Xiaohongshu. But in fact, their connection to AI is closer than outsiders imagine.

As a smart manufacturing giant, Luxshare Precision's entry represents the implementation of large models in the industrial internet field. Yield detection, robotic arm scheduling, and supply chain automation in factories all require strong data analysis and visual understanding capabilities. In recent years, robots going to work in factories has become a popular trend; the reason those robotics companies do this is to let robots go to the real world to level up and test their real capabilities in the physical world. DeepSeek has not directly participated in the robotics business, but it has invested in many related companies, including its earlier stake in Unitree Robotics and its recent lead investment in Winverse AI.

One can imagine that with DeepSeek's brain plus Luxshare Precision's manufacturing scenarios, future smart factories may widely deploy self-evolving AI robots. Luxshare Precision has also been deeply involved in the AI industry in recent years; it already has a data center hardware business, producing high-speed copper cables, optical modules, optical connections, and other components that are important parts of AI servers.

As for Meituan and Xiaohongshu, they hold relatively high-quality high-frequency scenario data in the domestic market. Take Meituan: its scarcity lies in owning massive sales and service data from merchants in the offline local life sector; Xiaohongshu is a community with a huge user base, with large amounts of real interaction data and strong consumption conversion capability.

(Image source: Xiaohongshu)

If they cooperate, DeepSeek can obtain high-quality corpus to make its models better understand users' real needs, a characteristic that lab data does not have. At the same time, Meituan and Xiaohongshu can also use DeepSeek to reconstruct their search, recommendation, and intelligent customer service. Meituan and Xiaohongshu's participation in DeepSeek's investment brings more possibilities for future cooperation. CATL, Geely: Penetrating new energy mobility scenarios

In this investment list of DeepSeek, CATL and Geely are the other two companies worth introducing separately. Their participation reflects the forms in which AI lands in the new energy and automotive industries.

Frankly speaking, today's cars increasingly resemble smart devices with four wheels. Geely's entry mainly targets two directions: smart cockpits and high-level intelligent driving.

First, the cockpit. Although current in-car voice assistants can complete basic commands such as opening and closing windows and navigation, the experience often suffers when encountering complex continuous conversations or vague intentions. With the integration of DeepSeek's large model, the in-car system can better understand users' natural semantics and continuous context.

Next, intelligent driving. The current technical route relies heavily on massive video data processing and cloud model training. DeepSeek's capabilities in multimodal understanding and logical reasoning can support Geely's intelligent driving algorithms.

(Image source: Geely)

As for CATL, outsiders mainly see it as a battery giant, but the R&D and production of power batteries are already inseparable from AI participation. At the R&D level, the reasoning capabilities of large models can help CATL shorten the verification cycle for battery materials and reduce trial-and-error costs.

At the manufacturing level, CATL not only has huge battery production lines but is also laying out unmanned mines and heavy-truck battery-swapping networks. In these complex physical scenarios, AI models can play a role in optimizing equipment scheduling efficiency and improving the yield of battery production lines.

To put it plainly, compared with phones and internet applications, putting AI models into cars and battery production lines faces higher reliability requirements and lower fault tolerance. But this is also the inevitable path for large models to go deep into the real economy. DeepSeek's binding with these two leading companies is, to a certain extent, aimed at obtaining more real-world data in mobility and manufacturing, so that the models can truly be implemented in specific links of the real industries. The substance behind the hundred-billion financing: DeepSeek expands its AI circle of friends

Looking back at this 105.5 billion financing list, frankly speaking, the funds may be only the most basic part of this cooperation. With DeepSeek's current scale, if it wants to maintain its lead in the next stage of competition, it cannot rely on fighting alone. This list covering nearly 30 companies is, to put it plainly, the circle of friends DeepSeek has assembled for itself across the entire AI industry chain.

In Leitech's (ID: leitech) view, the development of large models has now entered the deep-water zone of competing over implementation. On the one hand, bottlenecks in underlying hardware and rising costs are problems commonly faced by domestic AI. Bringing in SMIC and ChangXin Memory is equivalent to DeepSeek preparing a safeguard for its long-term future development. Once the external hardware supply chain fluctuates, with the support of these domestic semiconductor giants, DeepSeek can at least have alternatives and hold its basic position.

On the other hand, AI cannot live only in cloud servers, nor can it rely solely on web versions and API services to support an ever-growing valuation. If large models want to continue evolving, they need to absorb real-world data outside the laboratory; if AI technology wants to expand its commercialization tracks, it must also further penetrate ordinary consumers' terminal devices. vivo and Midea provide phone and home appliance terminals closer to users, Xiaohongshu and Meituan provide rich life consumption and social interaction data, while Geely, CATL, and Luxshare Precision possess mobility and physical manufacturing scenarios.

Overall, DeepSeek has not gone into the field itself to make phones, home appliances, or robots; through financing, it has formed communities of interest with leading players in various industries.

We can also see that in the next stage of AI industry competition, the focus will be on who can truly implement technology in specific scenarios where ordinary people are willing to pay. At present, DeepSeek has already made preparations in advance.

Image source for this article: 123RF licensed image library. Infringement report Return Share Related articles All comments