DeepSeek Leads a New Paradigm in AI: From Computing Power Competition to Algorithm Innovation

Recently, DeepSeek released the latest V3 version update - DeepSeek-V3-0324 on the Hugging Face platform. This model, with 685 billion parameters, has significant improvements in coding ability, UI design, and reasoning capabilities.

At the recently concluded 2025 GTC conference, NVIDIA CEO Jensen Huang gave high praise to DeepSeek. He emphasized that the market's previous belief that DeepSeek's efficient model would reduce the demand for chips was incorrect; future computing demand will only increase, not decrease.

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

The Symbiotic Evolution of Computing Power and Algorithm

In the field of AI, the improvement of Computing Power provides the foundation for more complex Algorithms to run, allowing models to process larger amounts of data and learn more complex patterns; while the optimization of Algorithms can utilize Computing Power more efficiently, enhancing the utilization efficiency of computing resources.

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

• Technical route differentiation: Some companies pursue the construction of ultra-large Computing Power clusters, while others focus on optimizing Algorithm efficiency, forming different technical sects.
• Industry Chain Reconstruction: Chip manufacturers become leaders in AI Computing Power through ecosystems, while cloud service providers lower deployment thresholds through elastic computing power services.
• Resource allocation adjustment: Enterprises seek a balance between hardware infrastructure investment and efficient Algorithm development.
• The Rise of Open Source Communities: Open source models allow for the sharing of algorithm innovations and computing power optimizations, accelerating technological iteration and diffusion.

Technical Innovations of DeepSeek

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

Model Architecture Optimization

DeepSeek adopts a combination architecture of Transformer + MOE (Mixture of Experts) and introduces a Multi-Head Latent Attention mechanism (MLA). This architecture resembles a super team where the Transformer handles regular tasks, while MOE acts like a group of experts within the team, each with their own area of expertise. When encountering specific problems, the most proficient expert handles the task, significantly improving the model's efficiency and accuracy. The MLA mechanism allows the model to flexibly focus on different important details when processing information, further enhancing the model's performance.

Innovative Training Methods

DeepSeek has proposed the FP8 mixed precision training framework. This framework acts like an intelligent resource allocator, dynamically selecting the appropriate computing power based on the needs of different stages during the training process. When high-precision calculations are required, it uses a higher precision to ensure the model's accuracy; while when lower precision is acceptable, it reduces the precision to save computing resources, improve training speed, and reduce memory usage.

Improvement in Inference Efficiency

During the inference phase, DeepSeek introduces the Multi-token Prediction (MTP) technology. Traditional inference methods predict one Token at a time, step by step. The MTP technology can predict multiple Tokens at once, significantly speeding up the inference process and reducing the cost of inference.

Reinforcement Learning Algorithm Breakthrough

DeepSeek's new reinforcement learning algorithm GRPO (Generalized Reward-Penalized Optimization) optimizes the model training process. Reinforcement learning is like providing the model with a coach, who guides the model to learn better behaviors through rewards and penalties. Traditional reinforcement learning algorithms may consume a lot of Computing Power in this process, while DeepSeek's new algorithm is more efficient, able to reduce unnecessary computations while ensuring improvements in model performance, thus achieving a balance between performance and cost.

These innovations are not isolated technical points, but rather form a complete technical system that reduces computing power requirements throughout the entire chain from training to inference. Regular consumer-grade graphics cards can now run powerful AI models, significantly lowering the barrier to entry for AI applications and enabling more developers and enterprises to participate in AI innovation.

Impact on Chip Manufacturers

Many people believe that DeepSeek bypassed the software layer of GPU manufacturers, thereby freeing itself from dependence on them. In fact, DeepSeek directly optimizes algorithms through the GPU manufacturers' PTX (Parallel Thread Execution) layer. PTX is an intermediate representation language that lies between high-level GPU code and actual GPU instructions. By operating at this level, DeepSeek can achieve more precise performance tuning.

The impact on chip manufacturers is twofold. On one hand, DeepSeek is actually more deeply tied to hardware and the ecosystem, and the lowering of the entry barrier for AI applications may expand the overall market size; on the other hand, the algorithm optimization of DeepSeek may change the market demand structure for high-end chips. Some AI models that originally required high-end GPUs to run can now run efficiently on mid-range or even consumer-grade graphics cards.

The Significance of AI Industry for China

The algorithm optimization of DeepSeek provides a technical breakthrough path for China's AI industry. Against the backdrop of constraints on high-end chips, the idea of "software compensating for hardware" alleviates the dependence on top imported chips.

Upstream, efficient algorithms reduce the pressure on computing power demand, allowing computing power service providers to extend hardware usage cycles and improve return on investment through software optimization. Downstream, the optimized open-source models lower the barriers to AI application development. Many small and medium-sized enterprises can develop competitive applications based on the DeepSeek model without the need for large amounts of computing power resources, which will lead to the emergence of more AI solutions in vertical fields.

The Profound Impact of Web3+AI

Decentralized AI Infrastructure

The optimization of DeepSeek's algorithm provides new momentum for Web3 AI infrastructure. The innovative architecture, efficient algorithms, and lower computing power requirements make decentralized AI inference possible. The MoE architecture is naturally suitable for distributed deployment, where different nodes can hold different expert networks without the need for a single node to store the complete model. This significantly reduces the storage and computing requirements of a single node, thereby improving the flexibility and efficiency of the model.

The FP8 training framework further reduces the demand for high-end computing resources, allowing more computing power to be added to the node network. This not only lowers the threshold for participating in decentralized AI computation but also enhances the overall computing capability and efficiency of the network.

Multi-Agent System

• Intelligent Trading Strategy Optimization: By the collaborative operation of agents such as real-time market data analysis agent, short-term price fluctuation prediction agent, on-chain trading execution agent, and trading result supervision agent, it helps users achieve higher returns.

• Automated execution of smart contracts: smart contract monitoring agents, smart contract execution agents, execution result supervision agents, etc. work together to achieve more complex business logic automation.

• Personalized portfolio management: AI helps users in real-time to find the best staking or liquidity provision opportunities based on their risk preferences, investment goals, and financial situation.

DeepSeek is innovating through algorithms to find breakthroughs under the constraint of Computing Power, paving a differentiated development path for China's AI industry. Lowering application thresholds, promoting the integration of Web3 and AI, reducing dependence on high-end chips, and empowering financial innovation are reshaping the digital economy landscape. The future of AI development is no longer just a competition of Computing Power, but a competition of the collaborative optimization of Computing Power and algorithms. On this new track, innovators like DeepSeek are redefining the rules of the game with Chinese wisdom.