DePIN Hunting Journey: AI Computing Power as Bait, the Road is Long and Challenging

Author: Hedy Bi, OKG Research Institute

The Hong Kong Web3 Carnival has come to an end, but the pulse of Web3 freedom continues to beat and permeate other industries. Compared to the previous cycle, the logic behind the current bull market has shifted from "native innovation narrative" to "mainstream recognition, capital-driven model." The observed development stage of Web3 has also evolved from "closed and niche absolute freedom" to "relative freedom under true inclusivity."

In this context, failing to step out of the box for analysis means that waiting for native innovation narratives is no longer suitable for the current development of Web3. Since the entire Web3 has embraced compliance, it has refocused on the financial sector under the continuous push from the Hong Kong government. Mainstream financial institutions are also accelerating their participation in Web3 through RWA and spot ETFs.

At this conference, in addition to mainstream financial institutions entering Web3, we also saw an opportunity to connect Web2 and Web3—the DePIN track. Especially with the advancement of AI large models, the sub-track of resource allocation in DePIN has once again become a hot topic.

Source: OKG Research

Computing Power as Bait, but AI Large Model Training is Not the Best Landing Scenario for DePIN

"Blockchain builds trust through technology, while AI is an industry that requires trust immensely." Haseed Qureshi, managing partner of Dragonfly Capital, stated at the conference.

DePIN is not a new track; it was proposed several years ago. The explosion of AI large models has led to extensive discussions in the industry regarding computing power and data, with estimates showing that the cost of large model computation increases by 31 times each year. The global GPU shortage has placed companies like NVIDIA at the top of the market demand food chain, giving them significant pricing power. The debate over whether to monopolize or decentralize has become a reason for renewed discussions in the Web3 DePIN track.

Although AI large model training is the cause, Rome wasn't built in a day; AI large model training is not currently the best landing scenario for DePIN. The requirements for computing power in AI large model production mainly revolve around two aspects: inference and training. In the training phase, a complex neural network model is trained by feeding it a large amount of data. In the inference phase, the trained model is used to draw various conclusions from a large amount of data.

Source: NVIDIA

The combination of decentralization and computing power sees a decreasing difficulty level from training to fine-tuning training and then to inference. In DePIN, more projects in the industry are focused on inference rather than training. The primary reason most enterprises use NVIDIA GPU clusters for AI training is their powerful parallel computing capabilities and memory bandwidth. Compared to the inference phase, the requirements for parallel computing power and bandwidth are significantly lower. Moreover, large model training places greater emphasis on stability because if the training is interrupted, it needs to be restarted. Building a decentralized computing power application on Ethereum for GPT would incur gas fees as high as $10 billion for a single matrix multiplication operation and take up to a month.

Additionally, I analyzed several popular projects from this conference, revealing a situation where supply exceeds demand, meaning the global supply of computing power surpasses the demand from AI developers for model training or inference tasks. This does not mean that demand does not exist; OpenAI's founder, Sam Altman, proposed raising $7 trillion to build an advanced chip factory ten times the current size of TSMC for chip production and model training. Research from Stanford University also indicates that for any language model, once the scale of training parameters exceeds a critical threshold, its performance (such as accuracy) improves dramatically. This is in stark contrast to the principle of "great effort yields miraculous results," and it also means that in reality, the concept of decentralized computing power still has many challenges to address.

The "Historical Origins" of the DePIN Track Can Be Traced Back to the "Sharing Economy"

The concept of DePIN is not difficult to understand and can even be traced back to Web2. Looking back at the internet industry, for at least 15 years, Web2 players have been immersed in aggregating individuals' tangible assets to create a "sharing economy." If intangible assets (such as idle servers) are directly redistributed to demand-side users through peer-to-peer (P2P) or peer-to-business (P2B) models, decentralized technology like blockchain can optimize production relationships through incentive mechanisms. This is what DePIN aims to achieve.

Therefore, there is a high enthusiasm on the supply side within the DePIN track. In fact, Web2 has been laying the groundwork for "redistribution" for a long time; this time, intermediaries are directly removed. Currently, there are nearly a thousand DePIN projects, especially in the Solana ecosystem. According to Messari, the Solana ecosystem is leading in DePIN infrastructure due to its high integration and performance of the public chain. In terms of regional distribution, it is expected that several of the top 10 DePIN projects will come from Asia between 2024 and 2025.

Source: Messari

Web3 and AI have many intersections, with computing power being the first point of focus as the universal currency of the future digital world. However, the most reasonable landing scenario for decentralized computing power is not the easiest to implement.

At the intersection of Web3 and AI, in addition to continuously overcoming technical challenges, there are many other branches worth exploring, such as empowering creators with ownership through AI agents and small AI model computing power scenarios, which may be more feasible. The success of business models and breakthroughs in technology will always find a balance, and DePIN is accelerating this process, with DePIN's "hunting journey" likely to yield fruitful results.