I was chatting with a friend who works in AI these days, and he mentioned that training a large model can account for 70% of the total cost in GPU expenses. I suddenly thought, could Boundless's prover network also get a share of this pie?
On the surface, both ZK proof generation and AI training require a large amount of GPU computation. However, upon deeper understanding, it becomes clear that the computational models of the two are quite different. AI training requires continuous, high-bandwidth GPU cluster collaboration, while ZK proof generation resembles independent batch processing tasks.
However, there is an interesting intersection point: the verification of AI reasoning. Many people are now concerned about the authenticity of AI-generated content. If ZK proofs could be used to verify that the outputs of AI models indeed come from specific models and inputs, it would be a huge market.
Boundless is actually already exploring this direction. Their zkML project enables machine learning models to run in zkVM, producing verifiable inference results. Although the performance is not yet good enough, the direction is correct.
Imagine this scenario: you generate a piece of text using GPT, but how do you prove that it was indeed generated by GPT and not written by a human? If there is ZK proof, you can demonstrate the authenticity of the output without exposing the model weights. This is very important for copyright protection and authenticity verification of AI-generated content.
But the challenges are significant. AI models often have tens of billions of parameters, and the overhead of running them in zkVM is enormous. Current technology may only be able to handle some small models. However, with advances in hardware and algorithms, this problem will eventually be solved.
Another idea is to use Boundless's network as a supplement to AI computation. For example, auditing intermediate results of AI training and model updates. These do not require continuous computation and are perfectly suited for Boundless's model.
In terms of market size, the AI computing market is much larger than the ZK proof market. If Boundless can tap into even 1% of the AI market, the demand for ZKC will surge. This could be a bigger opportunity than pure blockchain applications.
Competitors are also taking action. Projects like Gensyn and Akash are working on decentralized AI computation. However, they primarily provide raw computing power, while Boundless's advantage lies in verifiability. The two can actually complement each other.
In the short term, I am not very optimistic. The integration of AI and ZK is still in its early stages, and both the technology and the market are immature. However, in the long term, this is an inevitable trend. As AI becomes more powerful, verifying the authenticity and correctness of its outputs will become increasingly important.
For investors, this presents an additional upside. Even if blockchain applications do not develop as expected, the AI market may still bring growth. Having an additional growth engine is always beneficial.