DeepSeek V3 Update Leads a New Paradigm in AI

Last night, DeepSeek released the V3 version update on a certain platform—DeepSeek-V3-0324, with model parameters reaching 68.5 billion, showing significant improvements in coding capabilities, UI design, and reasoning abilities.

At the recent 2025 GTC conference, the CEO of a tech company highly praised DeepSeek. He emphasized that the market's previous belief that DeepSeek's efficient model would reduce the understanding of chip demand was incorrect, and that future computing demands will only increase, not decrease.

DeepSeek, as a representative product of algorithm breakthroughs, has a relationship with chip supply that is worth discussing. Let's first analyze the significance of computing power and algorithms for the development of the AI industry.

The Symbiotic Evolution of Computing Power and Algorithms

In the field of AI, the enhancement of computing power provides a foundation for running more complex algorithms, enabling models to handle larger amounts of data and learn more complex patterns; meanwhile, the optimization of algorithms can utilize computing power more efficiently, improving the utilization efficiency of computational resources.

The symbiotic relationship between computing power and algorithms is reshaping the AI industry landscape:

1. Differentiation of technical routes: Some companies pursue the construction of super-large computing power clusters, while others focus on optimizing algorithm efficiency, forming different technical factions.

2. Industry Chain Restructuring: A certain chip company has become the leader in AI computing power through its ecosystem, while cloud service providers lower deployment barriers through elastic computing services.

3. Resource allocation adjustment: Enterprises seek a balance between investment in hardware infrastructure and the development of efficient algorithms.

4. Rise of Open Source Communities: Open source models such as DeepSeek and LLaMA allow for the sharing of algorithm innovations and computing power optimization results, accelerating technology iteration and diffusion.

Technical Innovations of DeepSeek

The success of DeepSeek is closely related to its technological innovations. Here is a brief explanation of its main innovations:

model architecture optimization

DeepSeek employs a combination architecture of Transformer + MOE (Mixture of Experts) and introduces a Multi-Head Latent Attention (MLA) mechanism. This architecture functions like an efficient team of experts, capable of deploying the most suitable expert for different tasks, significantly enhancing the model's efficiency and accuracy.

Innovative Training Methods

DeepSeek has proposed an FP8 mixed-precision training framework. This framework can dynamically select the appropriate computation precision based on the needs of different stages during training, improving training speed and reducing memory usage while ensuring model accuracy.

inference efficiency improvement

During the inference stage, DeepSeek introduced Multi-token Prediction (MTP) technology. This technology allows for the prediction of multiple tokens at once, significantly accelerating the inference speed while reducing the inference cost.

Breakthrough in Reinforcement Learning Algorithms

DeepSeek's new reinforcement learning algorithm GRPO (Generalized Reward-Penalized Optimization) optimizes the model training process. This algorithm achieves a balance between performance and cost while ensuring an improvement in model performance and reducing unnecessary computations.

These innovations have formed a complete technical system that comprehensively reduces computing power requirements from training to inference. Now, ordinary consumer-grade graphics cards can run powerful AI models, significantly lowering the barrier to AI applications and enabling more developers and enterprises to participate in AI innovation.

Impact on Chip Suppliers

There are opinions that DeepSeek has bypassed the software layer of a certain chip company, thus freeing itself from dependence on it. In reality, DeepSeek optimizes algorithms directly through the company's underlying instruction set. By operating at this level, DeepSeek can achieve more precise performance tuning.

The impact on chip suppliers is twofold. On one hand, DeepSeek is more deeply integrated with its hardware and ecosystem, and the lowering of the threshold for AI applications may expand the overall market size; on the other hand, DeepSeek's algorithm optimization may change the market demand structure for high-end chips, as some AI models that originally required high-end GPUs to run can now efficiently operate on mid-range or even consumer-grade graphics cards.

The Significance of AI Industry in China

DeepSeek's algorithm optimization provides a technical breakthrough path for the Chinese AI industry. In the context of high-end chip limitations, the idea of "software compensating for hardware" alleviates the dependence on top imported chips.

Upstream, efficient algorithms have reduced the pressure on computing power demand, allowing computing service providers to extend hardware usage cycles through software optimization and improve return on investment. Downstream, the optimized open-source models have lowered 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 a large amount of computing resources, which will give rise to more AI solutions in vertical fields.

The Profound Impact of Web3+AI

Decentralized AI Infrastructure

DeepSeek's algorithm optimization provides new momentum for Web3 AI infrastructure. The innovative architecture, efficient algorithms, and lower computational power requirements make decentralized AI inference possible. The MoE architecture is inherently suitable for distributed deployment, allowing different nodes to hold different expert networks without requiring a single node to store the complete model, significantly reducing the storage and computation requirements of a single node, thereby enhancing the flexibility and efficiency of the model.

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

Multi-Agent Systems

1. Intelligent Trading Strategy Optimization: By analyzing real-time market data, predicting short-term price fluctuations, executing on-chain transactions, and supervising trading results through the collaborative operation of multiple agents, it helps users achieve higher returns.

2. Automated execution of smart contracts: Multiple agents collaborate on monitoring, executing, and supervising the results of smart contracts, enabling the automation of more complex business logic.

3. 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 seeks breakthroughs through algorithmic innovation under computational constraints, paving a differentiated development path for China's AI industry. Lowering application barriers, promoting the integration of Web3 and AI, reducing reliance on high-end chips, and empowering financial innovation—these impacts are reshaping the digital economy landscape. The future of AI development is no longer just a competition of computing power, but a competition of collaborative optimization between computing power and algorithms. On this new track, innovators like DeepSeek are redefining the rules of the game with Chinese wisdom.