DeAI: In the era of "barbaric growth" of AI, why is Web3 needed to govern it

null

Author: K, Web3Caff Research Fellow

In the development trajectory of artificial intelligence, the past two years have experienced a profound structural shift. Model capabilities continue to break through, inference efficiency keeps improving, and global capital and state machinery are flocking in. However, behind the fervor and capital-focused wave of centralization, DeAI (Decentralized AI Training and Inference Architecture) is becoming another path to the future, directly addressing two major hidden dangers in today’s AI development: blind trust mechanisms and expansion vulnerabilities.

The prosperity of centralized AI is built on massive physical infrastructure, from supercomputing clusters to closed black boxes for model inference, from packaged SaaS products to internal enterprise API calls. But just as the internet evolved from closed to open, from Web2 platforms to Web3 protocols, AI development will inevitably face two fundamental issues: First, how can users verify that the model inference results have not been tampered with and are authentic? Second, when training and inference cross geographical, device, cultural, and legal boundaries, can centralized architectures still maintain cost and performance advantages?

DeAI networks propose a solution radically different from the centralized paradigm. They center around “Verifiable Compute,” using cryptography and consensus mechanisms to ensure that each model operation has a traceable and provable execution path. This not only addresses users’ “blind trust” in models but also provides a universal trust foundation for cross-border collaboration. Pioneers like Prime Intellect and Inference Labs have already achieved partial verifiable inference in remote GPU clusters, opening new possibilities for distributed training and autonomous AI services. [70]

From an economic perspective, the rise of DeAI is also closely related to the shift in AI industry RoG (Return-on-GPU, i.e., revenue generated per hour of GPU compute). The design of GPT-4.1 no longer simply pursues larger models and more compute power but emphasizes fine-tuning and inference resource allocation, such as reusing existing context during generation and reducing unnecessary recomputation, thereby lowering invalid outputs and token consumption, and directing more compute power toward truly valuable inference processes. [68] This marks a shift in industry focus from “how many GPUs can be burned” to “how much value can be gained per hour.” This efficiency-oriented approach provides an excellent breakthrough point for decentralized AI networks.

The high fixed costs and efficiency bottlenecks of centralized GPU clusters in large-scale deployment will be hard to match by a permissionless, heterogeneous GPU network contributed by users worldwide. If such a network possesses “verifiability,” it can not only compete with the cost structures of centralized infrastructure providers like AWS and Azure but also inherently has transparency and trustworthiness advantages.

Moreover, the impact of DeAI extends far beyond technology; it will reshape the ownership and participation structure of AI development. In the current closed training ecosystems dominated by giants like OpenAI and Anthropic, most developers can only act as “model users” and cannot participate in training profits or inference decision-making. In DeAI networks, every contributor—whether providing computing power, data, or developing agent applications—can participate in governance and share rewards through protocols. This is not only an innovation in economic mechanisms but also a step forward in AI development ethics.

Of course, DeAI is still in early exploration. It has yet to establish performance levels that can replace centralized models, nor has it broken through network stability and verification efficiency bottlenecks. But the future of AI will not be a single path; it will run in parallel. Centralized platforms will continue to dominate the enterprise market, pursuing ultimate RoG optimization and productization; while DeAI networks will grow in edge scenarios and emerging markets, gradually evolving into a vibrant open model ecosystem. Just as the internet represents freedom of information, DeAI signifies autonomous intelligence. Its importance lies not only in technological advantages but also in offering a different world of possibilities—one where trust in specific intermediaries is unnecessary, yet trust in intelligence itself remains.

This content is excerpted from the research report: “Web3 2025 Annual 40,000-word Report (Part Two): The Intersection of Finance, Computing, and Internet Order, Industry Shift About to Begin? A Panoramic Breakdown of Structural Changes, Value Potential, Risk Boundaries, and Future Outlook,” published by Web3Caff Research.

This report (available for free reading) was authored by Web3Caff Research Fellow K, systematically analyzing the core logic of Web3 development stage changes around 2025, with a focus on why application exploration and system collaboration are gradually becoming new focal points amid ongoing evolution of underlying infrastructure and regulatory capabilities. Key points include:

• Stage evolution background: The internal reasons behind the end of infrastructure construction and the shift in industry focus;
• Key mechanism changes: The gradual clarification of rule frameworks and on-chain mechanisms, and their impact on system operation;
• Main application directions: Exploration paths around payment settlement, real-world scenario mapping, and programmable collaboration;
• Future development directions: The outlook for Web3 evolution beyond 2026.