DeepSeek: Disrupting the AI Industry with Low-Cost, High-Performance Innovation

DeepSeek: Disrupting the AI Industry with Low-Cost, High-Performance Innovation

DeepSeek is an artificial intelligence company based in Hangzhou, China, founded by Liang Wenfeng. On January 20, 2025, the company launched its first chatbot application powered by the DeepSeek-R1 model, available for iOS and Android. By January 27, DeepSeek had surpassed ChatGPT, becoming the most downloaded free app on the U.S. iOS App Store. This milestone has sparked global attention and discussions about its low-cost, high-performance AI model.

Q1: What are DeepSeek’s technological advantages?

DeepSeek-R1 stands out for its exceptional performance, cost efficiency, and open-source model.

• Outstanding Performance:DeepSeek-R1 has demonstrated strong capabilities in real-world applications. In the financial sector, it efficiently processes complex financial risk assessment models, providing accurate mathematical computations to support investment decisions. In software development, DeepSeek-R1 assists in code generation, significantly shortening development cycles while producing highly readable and maintainable code. In a math problem-solving competition organized by a leading AI evaluation institute, DeepSeek-R1 achieved an accuracy rate comparable to OpenAI’s O1 model.
• Cost Efficiency:DeepSeek trained the R1 model with an estimated cost of approximately $6 million, thanks to its unique algorithmic optimizations and resource allocation strategies. By leveraging innovative training algorithms, it reduces unnecessary computational steps and maximizes GPU utilization through efficient resource management. However, this $6 million only accounts for GPU computing costs, excluding expenses for research, infrastructure, and operations. Nevertheless, DeepSeek’s overall infrastructure investment remains significantly lower than that of major AI companies in the U.S.
• Open-Source Collaboration:DeepSeek embraces an open-source approach, making its code publicly accessible for review, use, and modification. This has attracted global developers to contribute and innovate collaboratively. According to open-source community data, the number of projects based on DeepSeek’s open-source code has grown significantly in the past six months, spanning industries such as healthcare, education, and transportation.

Q2: How has DeepSeek impacted the chip market?

DeepSeek’s rise has influenced the chip market in both training and inference, altering demand and competition dynamics across different chip models.

Training Phase

• NVIDIA Chips:NVIDIA’s H800 and H100 are widely used in AI training, but DeepSeek’s focus on cost efficiency and performance has led to a more selective procurement strategy. In response, NVIDIA may introduce more cost-effective chips or optimize the power efficiency of existing models to maintain its market share.
• AMD Chips:AMD’s MI300 series offers high computational power and memory bandwidth, making it theoretically suitable for large-scale model training. However, DeepSeek’s cost-conscious approach may push AMD to further optimize the MI300 series by improving energy efficiency and stability to enhance its competitiveness.
• Intel Chips:Intel’s Gaudi2 chip, known for its unique architecture, performs well in specific AI training scenarios. As DeepSeek explores diverse training solutions, the Gaudi2 chip could become a viable option. This could prompt Intel to increase its market outreach and enhance software compatibility to gain a stronger foothold in the AI training market.

Inference Phase

• Huawei Ascend 910C:DeepSeek has adopted Huawei’s Ascend 910C for inference tasks. As DeepSeek expands its business, Ascend 910C could see broader adoption, encouraging Huawei to accelerate its ecosystem development and attract more developers to build applications on Ascend chips, boosting its competitiveness.
• Google TPU:Google’s TPUs excel in tensor computations. With DeepSeek driving lightweight model inference towards edge computing, Google may optimize TPU performance for edge AI and reduce costs to cater to small and medium-sized enterprises (SMEs) and independent developers.
• Cambricon NPU:Cambricon specializes in AI inference. DeepSeek’s open-source model has encouraged more SMEs and developers to enter the AI space. In response, Cambricon may develop more user-friendly and cost-effective NPU products to meet the needs of these new entrants.
• MediaTek Edge AI Chips:MediaTek’s edge AI chips are power-efficient and cost-effective, making them ideal for smart home applications. As DeepSeek drives AI adoption, MediaTek may increase R&D investments to enhance chip performance and functionality, catering to the growing demand.
• Rockchip RK3588S:The RK3588S chip is widely used in AIoT applications. As DeepSeek explores AI deployment in IoT devices, Rockchip may optimize its software algorithms to enhance compatibility with DeepSeek models, expanding its presence in applications such as smart speakers and industrial automation.

Q3: How has DeepSeek affected memory products?

With the rising trend of local AI model deployment, demand for various memory products has surged, influencing their development trajectory.

• Consumer PC Memory:Deploying large models locally requires high-performance hardware, particularly for models with 32B+ parameters, which demand at least 24GB of GPU memory, larger RAM capacity, and better heat dissipation/electromagnetic shielding. DDR SDRAM memory, including DDR4 and DDR5, is seeing a shift in demand due to DeepSeek-driven local deployment trends. Users now prefer high-frequency, large-capacity DDR memory for smooth AI operations, prompting manufacturers to accelerate DDR technology advancements. DDR5 availability is increasing, and its pricing fluctuates based on demand and technology maturity.
• Server Memory:DeepSeek’s AI training and inference tasks require vast computational resources and data storage. This increases demand for high-performance, high-capacity server memory, particularly DDR SDRAM with ECC functionality, which ensures data accuracy and system stability—crucial for large-scale AI computations. ChipKill-enabled memory is also gaining popularity due to its enhanced fault tolerance. Memory manufacturers are ramping up R&D efforts to develop advanced server memory solutions.
• Laptop Memory:Although laptops have relatively weaker hardware performance, DeepSeek’s lightweight model development and mobile application expansion are influencing laptop memory demand. Users favor compact, high-capacity, high-speed, low-power, and efficient cooling memory solutions. While 16GB remains the mainstream laptop memory capacity, 32GB adoption is gradually increasing. As DeepSeek-powered mobile applications gain traction, laptop memory will continue to evolve to meet AI-driven requirements.
• VRAM (Video Memory):VRAM is crucial for AI computations, as GPUs frequently read and write data during DeepSeek training and inference processes. High-bandwidth, large-capacity VRAM, such as GDDR6 and GDDR6X, is essential for high-performance graphics cards, driving increased demand. Memory manufacturers are continuously innovating to enhance VRAM performance for AI workloads.

Q4: How has DeepSeek affected global tech companies and AI investment trends?

On January 27, 2025, NVIDIA’s stock price dropped by approximately 17%, wiping out $60 billion in market value. Other tech giants, including Microsoft and Alphabet (Google’s parent company), also experienced stock declines.

• Impact on Tech Companies:Microsoft has heavily invested in AI, with its Azure cloud services providing AI computing power. DeepSeek’s rise could prompt some Azure customers to switch to DeepSeek, impacting Microsoft’s cloud market share. In response, Microsoft may increase investments in open-source AI projects and leverage its technological strengths to develop more competitive AI products and services. Similarly, Google, which has long been at the forefront of AI research, faces competitive pressure. Google may shift its AI R&D focus towards model optimization and edge computing to adapt to market changes.
• Impact on AI Investment Models:Traditional AI model training requires massive financial investments. DeepSeek’s cost-effective success is prompting investors to reassess their funding strategies. Recent AI startup investments emphasize cost control and technical innovation rather than sheer capital scale. This shift is creating opportunities for more startups with innovative, low-cost AI solutions, fostering greater diversity in AI development.

Q5: What does DeepSeek’s success mean for AI adoption?

DeepSeek’s low-cost model and open-source approach are accelerating AI adoption.

• Lowering Barriers to AI Application:DeepSeek’s affordability enables SMEs and independent developers to build and deploy AI solutions. For example, Great Wall Motors has integrated DeepSeek technology into its “Coffee Intelligence” system, enhancing smart driving experiences. In healthcare, several companies have leveraged DeepSeek for medical imaging diagnostics, reducing costs and improving efficiency. The number of SMEs developing AI applications with DeepSeek has surged in the past six months.
• Advancing AI Democratization:DeepSeek’s open-source model fosters widespread AI innovation. The growing number of projects based on DeepSeek’s code demonstrates its impact across multiple sectors, driving AI’s real-world adoption.

Q6: How is DeepSeek reshaping global AI competition?

DeepSeek’s rise marks a significant breakthrough for China in AI, reshaping the global competitive landscape.

• Providing New Industry Insights:Its low-cost, high-performance model serves as a blueprint for AI startups worldwide. Many companies are adjusting their R&D strategies, prioritizing algorithm optimization and resource efficiency to enhance AI models while reducing costs.
• Promoting Market Diversification:The AI market is shifting from being dominated by tech giants to a more diversified ecosystem, with startups from India, Southeast Asia, and beyond adopting DeepSeek’s approach to develop AI solutions tailored to their local markets.
• Driving AI Democratization:AI is no longer exclusive to major tech corporations. DeepSeek’s success has fueled discussions on AI democratization, with its case being widely referenced in international AI conferences.

Latest Developments

DeepSeek recently announced partnerships with multiple research institutions to explore AI applications in environmental protection. Additionally, the company plans to release an upgraded DeepSeek-R2 model within the next six months, with substantial improvements in performance and generalization, expanding its applicability across industries.