Chinese artificial intelligence startup DeepSeek has commenced its first external fundraising initiative, a strategic move aimed primarily at establishing a market valuation benchmark for the company. According to a report by the South China Morning Post, this limited funding round is designed to provide employees with greater clarity on the value of their stock options, thereby mitigating the risk of departures to well-funded competitors in an environment of aggressive talent poaching across China's AI sector.
DeepSeek, based in Hangzhou, is not in urgent need of capital, as its parent company, the quantitative hedge fund High-Flyer, was China's second most profitable hedge fund in 2024, according to industry data. The startup intends to sell no more than 3% of its equity, preferring to target state-backed and industrial investors, including local government guidance funds and affiliated tech platforms that can offer strategic resources such as AI infrastructure. Private, market-oriented funds are expected to be included in subsequent financing rounds, given the capital-intensive nature of AI development.
The Strategic Importance of Valuation
The primary objective of this funding round is to set a clear valuation for DeepSeek, which will serve as a reference point for employee stock options. In China's intensely competitive AI talent market, startups often lose key personnel to larger tech giants like Alibaba, Tencent, and Baidu, which can offer higher salaries and more lucrative equity packages. By establishing a transparent valuation, DeepSeek hopes to give its employees confidence in the long-term value of their holdings, reducing the incentive to leave.
DeepSeek was founded in 2023 by Liang Wenfeng, a prominent figure in China's quantitative finance world. High-Flyer, which Liang co-founded, has been a major beneficiary of China's AI boom, using machine learning models to generate trading strategies. DeepSeek itself focuses on developing large language models (LLMs) and generative AI technologies, positioning itself as a competitor to Baidu's Ernie Bot and Alibaba's Tongyi Qianwen. The startup has gained attention for its open-source models, such as DeepSeek-Coder, which targets code generation tasks, and DeepSeek-V2, a general-purpose LLM that has shown competitive performance against leading models like GPT-4.
Access to Chips: The Real Bottleneck
The funding round comes at a time when DeepSeek faces significant challenges beyond capital. Liang Wenfeng has publicly stated that the company's main constraint is not money but rather access to advanced semiconductors, due to US export restrictions imposed on China. These restrictions, which target high-performance AI chips like Nvidia's A100 and H100, have forced Chinese AI startups to rely on less powerful alternatives or stockpile existing inventory. DeepSeek, like its peers, has been seeking alternative sources of chips, including those from domestic manufacturers like Huawei, but the performance gap remains a critical bottleneck.
The Information earlier reported that DeepSeek was aiming to raise at least $300 million at a valuation of at least $10 billion, and that Tencent Holdings and Alibaba Group were in discussions to invest at a valuation of over $20 billion. However, one of the investors who spoke to the Post said DeepSeek's estimated valuation could exceed $100 billion, a figure that reflects the high expectations for China's AI market. The wide range of valuations highlights the uncertainty in the sector, where investors are betting on future breakthroughs while grappling with geopolitical risks.
Background: China's AI Landscape
China's AI industry has experienced explosive growth over the past five years, fueled by government support, a vast pool of technical talent, and abundant data. However, the sector is also characterized by intense competition and high employee turnover. Companies like ByteDance, Meituan, and Xiaomi have all made significant investments in AI, while traditional tech giants are pouring billions into developing their own LLMs. In this environment, retaining top researchers and engineers is crucial for any startup's survival.
DeepSeek's approach of selling a small equity stake to state-backed investors is not unique. Many Chinese AI startups, including SenseTime, Megvii, and 4Paradigm, have similarly sought strategic investors that can provide not just capital but also access to government contracts, data, and infrastructure. Local government guidance funds, which are often managed by municipal or provincial governments, have become a key source of funding for tech companies, as they align with the national push for self-sufficiency in AI.
The startup's parent, High-Flyer, has been a major player in China's quantitative trading space, managing over 100 billion yuan ($14 billion) in assets at its peak. The hedge fund has been particularly successful in using machine learning models to generate alpha, with its flagship product, "Hengxin 8," delivering double-digit returns in 2023. This financial backing gives DeepSeek a considerable advantage over other AI startups that must constantly raise external capital.
DeepSeek's Technical Roadmap
DeepSeek has released several open-source models that have garnered attention in the AI community. In early 2024, the company introduced DeepSeek-V2, a 236-billion-parameter mixture-of-experts model that achieved state-of-the-art results on multiple benchmarks, including MMLU and HumanEval. The model's architecture uses a novel attention mechanism called "Multi-Head Latent Attention" to improve efficiency. DeepSeek also released DeepSeek-Coder, a specialized model for code generation that has been adopted by developers for tasks like bug fixing and code completion.
The startup's focus on open-source is a deliberate strategy to build goodwill and attract talent. By sharing its models publicly, DeepSeek can showcase its technical prowess and invite contributions from the global research community, which can help improve its models more rapidly. This approach mirrors that of Meta's LLaMA series and Mistral AI's open-source models, which have gained widespread adoption.
DeepSeek has also invested heavily in training infrastructure. The company operates a cluster of thousands of Nvidia A100 GPUs, which it obtained before the latest US export restrictions took effect. However, as these chips age and demand for compute grows, DeepSeek will need to secure newer, more powerful hardware. This is where strategic investors can help—by providing access to cloud computing resources or facilitating partnerships with chip suppliers.
Market Reactions and Future Prospects
The news of DeepSeek's funding round has generated significant buzz in the tech and investment communities. Analysts see the move as a sign that Chinese AI startups are maturing and seeking to professionalize their operations. The decision to involve state-backed investors also indicates a desire to align with government priorities, such as achieving technological independence and building a domestic AI ecosystem.
However, challenges remain. The US is expected to tighten export controls further, potentially restricting even less advanced chips. This could force DeepSeek and other Chinese AI companies to rely on domestic alternatives like Huawei's Ascend 910B or Cambricon's MLU series, which still lag behind Nvidia in performance. Additionally, the global AI talent shortage means that even with better stock options, DeepSeek will face constant pressure to retain its top people.
DeepSeek's valuation, which could reach $100 billion or more, reflects both the high potential of its technology and the frothy nature of the AI market. For comparison, OpenAI was valued at $80 billion in early 2024, while Anthropic reached $18 billion. While DeepSeek's valuation may seem inflated given its shorter track record, investors are betting that China's massive market and government support will enable it to become a major global player.
The company's parent, High-Flyer, has the financial resources to support DeepSeek through its early stages. But as the AI sector becomes more capital-intensive, with training runs costing tens of millions of dollars, even well-funded startups will need to tap external sources. The current funding round is likely just the first of many, as DeepSeek scales up its model training and explores commercial applications.
In the coming months, DeepSeek is expected to launch new models focused on multimodal capabilities, including image and video generation. It is also reportedly building a platform to allow businesses to fine-tune its models for specific use cases, such as customer service, content moderation, and financial analysis. These product developments could help DeepSeek generate revenue and attract additional investment.
As of early 2025, DeepSeek employs approximately 500 people, many of whom are PhDs from top Chinese universities like Tsinghua and Peking. The company has also hired several researchers from overseas, including former employees of Google Brain and Microsoft Research. This talent concentration underscores the startup's ambition to compete at the highest level of AI research.
The broader AI landscape in China remains fragmented, with dozens of startups vying for dominance. DeepSeek's strategy of combining top-tier research, open-source engagement, and strategic government partnerships could give it a unique edge. However, it must navigate the geopolitical minefield of chip sanctions and ensure that its valuation does not become a distraction from building real-world products.
Source:Silicon UK News