DeepSeek V3 Update: Leading a New Paradigm of AI Algorithm and Computing Power

Recently, DeepSeek released the latest V3 version update - DeepSeek-V3-0324. This model has 68.5 billion parameters and has made significant improvements in coding capabilities, UI design, and reasoning abilities.

At the recently concluded 2025 GTC conference, NVIDIA CEO Jensen Huang highly praised DeepSeek's achievements. He pointed out that the market's previous belief that DeepSeek's efficient model would reduce demand for chips was incorrect; in fact, future computing demand will only continue to increase.

DeepSeek, as a representative product of algorithm breakthroughs, has sparked thoughts on the role of Computing Power and Algorithm in the development of the AI industry, particularly in its relationship with chip supply.

The Synergistic Development of Computing Power and Algorithm

In the field of AI, the enhancement of Computing Power provides the operating foundation for more complex Algorithms, enabling models to handle larger datasets and learn more intricate patterns. At the same time, the optimization of Algorithms can utilize Computing Power more efficiently, improving the utilization efficiency of computing resources.

The collaborative relationship between Computing Power and Algorithm is reshaping the AI industry landscape:

1. Technical route differentiation: Some companies focus on building ultra-large Computing Power clusters, while others are dedicated to optimizing Algorithm efficiency, forming different technical schools.

2. Industrial Chain Restructuring: Some companies have become leaders in AI Computing Power through ecosystems, while cloud service providers reduce deployment thresholds through elastic Computing Power services.

3. Resource allocation adjustment: Enterprises seek a balance between hardware infrastructure investment and efficient algorithm development.

4. The Rise of Open Source Communities: Open source models enable the sharing of algorithm innovations and Computing Power optimization results, accelerating technological iteration and diffusion.

Technical Innovations of DeepSeek

The success of DeepSeek is inseparable from its technological innovations. Below is a brief explanation of its main technological innovations:

Model Architecture Optimization

DeepSeek adopts a combination architecture of Transformer + MOE (Mixture of Experts) and introduces a Multi-Head Latent Attention (MLA) mechanism. This architecture is like an efficient team, where the Transformer handles regular tasks, the MOE acts as an expert group for specific problems, and the MLA mechanism allows the model to flexibly focus on important details.

Methodological Innovation

DeepSeek has proposed the FP8 mixed precision training framework, which can dynamically select the appropriate computing power based on the needs during the training process, improving training speed and reducing memory usage while ensuring model accuracy.

Improvement in Inference Efficiency

During the inference phase, DeepSeek introduced Multi-token Prediction (MTP) technology, which can predict multiple Tokens at once, greatly improving inference speed and reducing costs.

Reinforcement Learning Algorithm Breakthrough

DeepSeek's new reinforcement learning algorithm GRPO (Generalized Reward-Penalized Optimization) optimizes the model training process, achieving a balance between performance and cost while ensuring performance improvements and reducing unnecessary Computing Power.

These innovations have formed a complete technological system, reducing the computing power requirements across the entire chain from training to inference, allowing ordinary consumer-grade graphics cards to run powerful AI models, significantly lowering the threshold for AI applications.

Impact on Chip Suppliers

DeepSeek optimizes algorithms through a company's PTX (Parallel Thread Execution) layer, which is actually more deeply integrated with the ecosystem of hardware vendors. This optimization may expand the overall market size on one hand, while on the other hand, it may alter the market's demand structure for high-end chips.

Significance to China's AI Industry

DeepSeek's algorithm optimization provides a technological breakthrough path for China's AI industry. Against the backdrop of limited high-end chips, the idea of "software compensating for hardware" reduces reliance on top imported chips.

Upstream, efficient algorithms reduce the pressure on computing power demands, enabling computing power service providers to extend hardware usage cycles through software optimization, thereby improving return on investment. Downstream, the optimized open-source models lower the barriers to AI application development, allowing numerous small and medium enterprises to develop competitive applications based on the DeepSeek model.

The Profound Impact of Web3+AI

Decentralized AI Infrastructure

DeepSeek's innovation provides new momentum for Web3 AI infrastructure. The MoE architecture is suitable for distributed deployment, and the FP8 training framework reduces the demand for high-end Computing Power resources, all of which contribute to the construction of a decentralized AI inference network.

Multi-Agent System

DeepSeek's technological innovations provide possibilities for multi-agent system applications in the Web3 domain, including:

1. Intelligent Trading Strategy Optimization: By coordinating multiple specialized agents, it helps users achieve higher returns.

2. Automated execution of smart contracts: Multiple agents work together to achieve automation of more complex business logic.

3. Personalized Portfolio Management: AI finds the best staking or liquidity provision opportunities in real-time based on user needs.

DeepSeek has opened up a differentiated development path for China's AI industry through algorithm innovation, lowering application barriers, promoting the integration of Web3 and AI, reducing dependence on high-end chips, and empowering financial innovation. These impacts are reshaping the landscape of the digital economy. The future development of AI will be a competition of collaborative optimization between Computing Power and Algorithm, and innovators like DeepSeek are redefining the rules of this competition with Chinese wisdom.