Author: David & Goliath
Compiled by: TechFlow
Currently, the computing and training segments of the AI industry are mainly dominated by centralized Web2 giants. These companies occupy a dominant position thanks to their strong capital strength, the most advanced hardware equipment, and vast data resources. Although this situation may continue to exist in the development of the most powerful general machine learning (ML) models, Web3 networks may gradually become a more economical and accessible source of computing resources for mid-range or customized models.
Similarly, when the inference demand exceeds the capability of individual edge devices, some consumers may choose Web3 networks to obtain less censored and more diverse outputs. Rather than attempting to comprehensively disrupt the entire AI technology stack, Web3 participants should focus on these niche scenarios and fully leverage their unique advantages in censorship resistance, transparency, and social verifiability.
The hardware resources required to train the next generation of base models (such as GPT or BERT) are scarce and expensive, and the demand for the highest performance chips will continue to exceed supply. This resource scarcity leads to the concentration of hardware in the hands of a few well-funded leading enterprises, which use this hardware to train and commercialize the most optimized and complex base models.
However, the pace of hardware updates and replacements is extremely fast. So, how can these outdated mid-range or low-performance hardware be utilized?
These hardware are likely to be used to train simpler or more targeted models. By matching different categories of models with hardware of different performance, optimal resource allocation can be achieved. In this case, Web3 protocols can play a key role in coordinating access to diverse, low-cost computing resources. For example, consumers can use simple mid-range models trained on their personal data sets, and only choose high-end models trained and hosted by centralized enterprises when processing more complex tasks, while ensuring that user identities are hidden and prompt data is encrypted.
In addition to efficiency issues, concerns about biases and potential censorship in centralized models are also growing. The Web3 environment is known for its transparency and verifiability, and can provide training support for models that are overlooked or considered too sensitive by Web2. Although these models may not be competitive in terms of performance and innovation, they still have important value for certain groups in society. Therefore, Web3 protocols can carve out a unique market by providing more open, trustworthy, and censorship-resistant model training services.
Initially, the centralized and decentralized approaches can coexist, each serving different use cases. However, as Web3 continues to improve in developer experience and platform compatibility, and the network effects of open-source AI become more apparent, Web3 may ultimately compete in the core areas of centralized enterprises. Especially as consumers become more aware of the limitations of centralized models, the advantages of Web3 will become more pronounced.
In addition to training mid-range or domain-specific models, Web3 participants also have the advantage of providing more transparent and flexible inference solutions. Decentralized inference services can bring various benefits, such as zero downtime, modular combination of models, public model performance evaluation, and more diverse and uncensored outputs. These services can also effectively avoid the "vendor lock-in" problem that consumers face due to reliance on a few centralized providers. Similar to model training, the competitive advantage of the decentralized inference layer does not lie in the computing power itself, but in solving some long-standing issues, such as the lack of transparency in closed-source fine-tuning parameters, the lack of verifiability, and the high costs.
Dan Olshansky has proposed a promising vision, namely, through the AI inference routing network of POKT, to create more opportunities for AI researchers and engineers to put their research into practice and earn additional income through customized machine learning (ML) or artificial intelligence (AI) models. More importantly, this network can promote fairer competition in the inference service market by integrating inference results from different sources (including decentralized and centralized providers).
Although optimistic forecasts suggest that the entire AI technology stack may eventually migrate to the chain, from the current perspective, this goal still faces the huge challenge of data and computing resource centralization, as these resources provide significant competitive advantages to existing giants. However, decentralized coordination and computing networks have unique value in providing more personalized, economical, openly competitive, and censorship-resistant AI services. By focusing on the most critical niche markets for these values, Web3 can establish its own competitive barriers, ensuring that the most influential technologies of this era can evolve in multiple directions, benefiting a wider range of stakeholders, rather than being monopolized by a few traditional giants.
Finally, I would like to express my special thanks to the entire team of Placeholder Investment, as well as Kyle Samani from Multicoin Capital, Anand Iyer from Canonical VC, Keccak Wong from Nectar AI, Alpin Yukseloglu from Osmosis Labs, and Cameron Dennis from the NEAR Foundation, for their review and valuable feedback during the writing of this article.
