The 麦通 MSX finally went TGE. This is a world-leading RWA trading platform. If you want to get in via Alpha, the expected cutoff is around 225–230 points.
A friend in my group asked me where the model with @OpenGradient is located. I answered offhandedly, “On-chain.” Then I realized something was off. I went back and checked the technical documentation: the model is not on-chain. On-chain there’s only a Blob ID. OpenGradient uses Walrus for decentralized storage, and the large files—everything like the AI model itself and the zkML proofs—are all stored as blobs off-chain. What the blockchain records is just the ID pointing to that blob, nothing more.
My first reaction was: isn’t that a compromise? But after thinking it through, I realized it’s not a compromise—it’s the only workable approach. A medium-sized AI model is already several GB, and a complete zkML proof file is also not small. If you really put all of that on-chain, the blockchain would get overwhelmed immediately: astronomical gas fees, and the network simply wouldn’t be able to run. Walrus’s design separates storage from verification. You don’t need to put the data on-chain; you only need to put the proof of “this data exists and hasn’t been tampered with” on-chain. The chain is the verification layer, not the storage layer.
This is a question many people haven’t thought through when they say “on-chain AI”: what exactly should be stored on-chain? OpenGradient’s answer is: store only the most minimal verifiable proof, and leave everything else to decentralized storage. This allows the model behind OpenGradient Chat to be any size, without being constrained by on-chain storage limits.
My judgment right now is very clear: until there’s publicly verifiable data about the stability of the Walrus storage layer, I’m treating this as the key link in the architecture integrity—not as a certainty. If you’re only doing short-term trading, this doesn’t concern you. But if you’re assessing whether OpenGradient can support real-scale AI workloads, then the reliability of the storage layer is the question nobody asks but should be asked. When do you think the market will start seriously pricing this “on-chain AI storage” issue?
Back to $OPG : the Walrus storage piece itself doesn’t directly consume OPG, but it’s the prerequisite for running large models across the inference network. Without reliable decentralized storage, the 4500+ models in the Model Hub are basically just display items. Whether the storage layer is stable determines whether the inference demand curve above can actually be fulfilled in practice. #opg $OPG
A friend in my group asked me where the model with @OpenGradient is located. I answered offhandedly, “On-chain.” Then I realized something was off. I went back and checked the technical documentation: the model is not on-chain. On-chain there’s only a Blob ID. OpenGradient uses Walrus for decentralized storage, and the large files—everything like the AI model itself and the zkML proofs—are all stored as blobs off-chain. What the blockchain records is just the ID pointing to that blob, nothing more.
My first reaction was: isn’t that a compromise? But after thinking it through, I realized it’s not a compromise—it’s the only workable approach. A medium-sized AI model is already several GB, and a complete zkML proof file is also not small. If you really put all of that on-chain, the blockchain would get overwhelmed immediately: astronomical gas fees, and the network simply wouldn’t be able to run. Walrus’s design separates storage from verification. You don’t need to put the data on-chain; you only need to put the proof of “this data exists and hasn’t been tampered with” on-chain. The chain is the verification layer, not the storage layer.
This is a question many people haven’t thought through when they say “on-chain AI”: what exactly should be stored on-chain? OpenGradient’s answer is: store only the most minimal verifiable proof, and leave everything else to decentralized storage. This allows the model behind OpenGradient Chat to be any size, without being constrained by on-chain storage limits.
My judgment right now is very clear: until there’s publicly verifiable data about the stability of the Walrus storage layer, I’m treating this as the key link in the architecture integrity—not as a certainty. If you’re only doing short-term trading, this doesn’t concern you. But if you’re assessing whether OpenGradient can support real-scale AI workloads, then the reliability of the storage layer is the question nobody asks but should be asked. When do you think the market will start seriously pricing this “on-chain AI storage” issue?
Back to $OPG : the Walrus storage piece itself doesn’t directly consume OPG, but it’s the prerequisite for running large models across the inference network. Without reliable decentralized storage, the 4500+ models in the Model Hub are basically just display items. Whether the storage layer is stable determines whether the inference demand curve above can actually be fulfilled in practice. #opg $OPG