Depth Analysis》DeepSeek impacts on the protocols generated by Web3 AI upstream and downstream

DeepSeek bursts the last bubble on the agent track, DeFAI may give birth to a new life, and the industry's financing method will usher in a transformation. This article was written by Kevin, the Researcher at BlockBooster and reprinted by Foresight News. Binance Report: How is DeFAI reinventing the Decentralized Finance interactive experience? TLDR: The emergence of DeepSeek has shattered the Computing Power moat, and the optimization of Computing Power led by the Open Source model has become a new direction; DeepSeek Favourable Information The model layer and application layer in the upstream and downstream of the industry have a negative impact on the computing power protocol in the infrastructure. DeepSeek's Favourable Information inadvertently bursts the last bubble on the Agent track, and DeFAI is most likely to give birth to a new life; The zero-sum game of project financing is expected to come to an end, and new financing methods of community launch + small number of VCs may become the norm. The impact caused by DeepSeek will have a profound impact on the upstream and downstream of the AI industry this year, and DeepSeek has successfully enabled home consumer graphics cards to complete the large model training tasks that can only be undertaken by a large number of high-end GPUs. Computing Power, the first moat around AI, began to crumble, and when algorithmic efficiency ran at 68% per year, and hardware performance climbed linearly following Moore's Law, the valuation models ingrained in the past three years no longer applied, and the next chapter of AI will be opened by the Open Source model. Although Web3's AI protocol is completely different from Web2's, it is also inevitable to suffer the impact of DeepSeek, which will lead to entirely new use cases for Web3 AI upstream and downstream: infrastructure layer, interactive software layer, model layer, and application layer. Through the analysis of technical architecture, functional positioning and actual use cases, I divide the entire ecosystem into: infrastructure layer, interactive software layer, model layer, application layer, and sort out their dependencies: Infrastructure layer The infrastructure layer provides the underlying resources of Decentralization (Computing Power, Storage, L1), Among them, the Computing Power protocol includes: Render, Akash, io.net, etc.; Storage protocols include: Arweave, Filecoin, Storj, etc.; L1 has: NEAR, Olas, Fetch.ai, etc. Computing Power layer protocolsupport model training, inference and framework execution; The storage protocol stores training data, model arguments and on-chain interaction records; L1 optimizes data transfer efficiency through a dedicated node, droplatency. Interactive software layer The interactive software layer is a bridge between the online infrastructure and upper-layer applications, providing framework development tools, data services and privacy protection, among which the data annotation protocols are: Grass, Masa, Vana, etc.; The development framework protocol includes: Eliza, ARC, Swarms, etc. Privacy computing protocols include: Phala, etc. The data service layer provides fuel for model training, the development framework relies on the computing power and storage of the infrastructure layer, and the privacy computing layer protects the security of data in training/inference. Model layer The model layer is used for model development, training, and distribution, with the Open Source model training platform: Bittensor. The model layer relies on data from the Computing Power and intermediary software layers of the infrastructure layer. The model is deployed to on-chain through the development framework; The model marketplace feeds the training results to the Application Layer. Application Layer Application Layer is an AI product for end users, among which agents include: GOAT, AIXBT, etc.; DeFAI protocols include: Griffain, Buzz, etc. Application Layer calls the pre-trained model of the model layer; Privacy computing that relies on the intermediary software layer; Complex applications require real-time computing power at the infrastructure layer. DeepSeek may have a negative impact on DecentralizationComputing Power According to a sample survey, about 70% of Web3 AI projects actually call OpenAI or centralized cloud platforms, only 15% use Decentralization GPUs (such as the Bittensor subnet model), and the remaining 15% are hybrid architectures (sensitive data is processed locally, and general-purpose tasks are migrated to the cloud). The actual usage of the DecentralizationComputing Powerprotocol was much lower than expected and did not match its actual market capitalization. There are three reasons for the low usage: Web2 developers use the original toolchain when migrating to Web3; The Decentralization GPU platform has yet to realize its price advantage; Some projects circumvent data compliance scrutiny in the name of "Decentralization", and the actual computing power still relies on centralized cloud. AWS/GCP has a 90%+ market share of AI Computing Power, compared to Akash's equivalent Computing Power of only 0.2% of AWS. The moats of centralized cloud platforms include: cluster management, RDMA high-speed network, and elastic scaling; The Decentralization cloud platform has an improved version of web3 of the above technology, but the shortcomings that cannot be perfected are, latency problems: distributed node communication latency is 6 times that of centralized cloud; Toolchain fragmentation: PyTorch/TensorFlow does not natively support Decentralization scheduling. DeepSeek reduces computing power consumption by 50% through sparse training, and dynamic model pruning enables consumer-grade GPUs to train tens of billions of argument models. Demand for high-end GPUs in the short term has been significantly lowered, and the market potential of Edge Computing has been revalued. As shown in the chart above, before the advent of DeepSeek, most protocols and applications in the industry used platforms such as AWS, and only a few use cases were deployed in Decentralization GPU networks, which saw the latter's price advantage over consumer-grade computing power and did not follow the impact of latency. This situation may worsen with the advent of DeepSeek. DeepSeek has released the limitations of long-tail developers, low-cost and efficient inference models will be popularized at an unprecedented speed, in fact, the above-mentioned centralized cloud platform and many countries have begun to deploy DeepSeek, and the large drop in inference costs will spawn a large number of front-end applications, which have a huge demand for consumer-grade GPUs. In the face of the upcoming huge market, centralized cloud platforms will launch a new round of user competition, not only with the top platforms, but also with countless small centralized cloud platforms. The most direct way to compete is to reduce prices, and it is foreseeable that the price of 4090 on centralized platforms will usher in a reduction, which is a disaster for Web3's Computing Power platform. When price is not the only moat for the latter, and the computing power platform in the industry is forced to lower prices, the result is that io.net, Render, and Akash cannot afford it. The price war will destroy the latter's last remaining valuation ceiling, and the death spiral from declining earnings and user churn could take DecentralizationComputing Powerprotocol in a new direction. The specific significance of DeepSeek to the upstream and downstream protocols of the industry As shown in the figure, I think DeepSeek will have different impacts on the infrastructure layer, model layer and application layer, in terms of positive effects: Application layer will benefit from a large drop in inference costs, and more applications can use low cost to ensure that agent applications are online for a long time and complete tasks in real time; ...