There is a concern among U.S. tech giants regarding China’s DeepSeek’s rise in the AI sector, especially with companies R1 Model going live.

DeepSeek-R1 Series focuses on advanced reasoning capabilities, including the DeepSeek-R1-Lite-Preview, which is aimed at logical inference and mathematical reasoning. DeepSeek has developed AI models like DeepSeek-V3 that offer comparable or superior performance to those from U.S. tech companies but at a fraction of the cost. This is largely due to its innovative approaches to model architecture and training, which allow for high efficiency with lower-end hardware. The cost-effectiveness of DeepSeek’s models, such as the DeepSeek-R1 series, threatens the business models of companies like Nvidia, OpenAI, and others that rely on high-cost, high-performance solutions.

By releasing its models as open source, DeepSeek is not only democratizing access to advanced AI technologies but also challenges the proprietary models that many U.S. tech firms have built their businesses on. This could potentially lead to a broader adoption of AI across various sectors that previously could not afford such technology, disrupting the market control held by a few major players.

The open-source nature of DeepSeek’s creates possibilities that any advantage gained by U.S. companies through proprietary tech might be short-lived as competitors worldwide can now build upon or directly use these models.

Chinas success in AI could be a geopolitical challenge to U.S. leadership, especially with ongoing trade tensions and export restrictions on advanced chips. DeepSeek’s ability to achieve high performance with less advanced hardware undermines assumptions about the necessity of cutting-edge technology for AI leadership, potentially affecting stock markets and investor confidence in U.S. tech stocks.

DeepSeek is a Chinese artificial intelligence lab that focuses on developing open-source large language models (LLMs). DeepSeek was founded in May 2023 by Liang Wenfeng, who also established High-Flyer, a Chinese hedge fund that uses AI in its trading strategies. It has been disruptive by offering high-performance models at significantly lower costs, challenging the pricing models of other tech giants. This approach has led to a broader impact on the AI industry.

The company’s strategy includes leveraging open-source development, which not only fosters collaboration but also mitigates some of the resource constraints imposed by US sanctions on advanced chip exports to China. This has allowed DeepSeek to innovate in model architecture and training methods, making them more efficient with limited resources.

DeepSeek Coder was Released in November 2023. This was DeepSeek’s first model, designed for coding tasks and available for free to both researchers and commercial users. DeepSeek LLM was launched later in November 2023, it scaled up to 67 billion parameters, aiming to compete with contemporaries like GPT-4. It faced challenges with computational efficiency but introduced a chat version, DeepSeek Chat. DeepSeek-V2 was released in May 2024, it marked a significant step by offering strong performance at a lower cost, igniting a price war in China’s AI model market. DeepSeek-V2 was particularly noted for its performance in various benchmarks, including AlignBench and MT-Bench, where it was competitive with or surpassed models like GPT-4 and LLaMA3-70B. DeepSeek-V3 was introduced in late 2024, this model features an enormous 671 billion parameters, trained on 14.8 trillion tokens, and is considered one of the best open-source AI models available, outperforming many closed-source models in benchmarks. It’s also known for being cost-effective to train and run, with an emphasis on reasoning capabilities.

DeepSeek-R1 Series models focus on advanced reasoning capabilities, including the DeepSeek-R1-Lite-Preview, which aimed at logical inference and mathematical reasoning. The fear is that if DeepSeek can offer similar or better services at much lower costs, the justification for the billions spent on AI by U.S. firms could be questioned, leading to potential market share loss and reduced future earnings.

Galactik Views