On-Chain AI Inference: Different Logic Behind Selling Computing Power

Orbio, Surplus Intelligence, and Antseed are all reducing the cost of on-chain AI inference in their own unique ways.

Written by: David Christopher
Compiled by: Chopper, Foresight News

The inference market has become one of the clearest directions for the integration of cryptocurrency and AI, as projects leverage on-chain markets to significantly lower the barriers and costs associated with AI model invocation.

The core model of these markets is mostly based on discount reselling of API invocation quotas and idle inference computing power, with some projects building networks that allow demand-side users to connect directly with independent computing service providers.

In this emerging sector, there are three representative markets: Orbio, deployed on the Robinhood Chain, and Surplus Intelligence and Antseed, both deployed on Base. While all three can provide low-cost inference services, there are significant differences in service positioning, target users, and token economic design. Below, we compare these three projects side by side.

Orbio

The core service target of the inference market created by Orbio is AI agents. The project initially started as a token that distributed OpenRouter quotas to holders, but has now evolved into a complete protocol: supporting hundreds of AI model invocations, many of which are heavily discounted; it also features a Launchpad where developers can deploy agents and issue tokens paired with ORBIO for trading.

To adapt to the usage needs of agents, Orbio has integrated privacy computing capabilities provided by Phala and NEAR AI. Recently, it also partnered with Mosh to integrate Mosh's tool ecosystem, helping new agent tokens establish more sustainable token growth mechanisms.

These capabilities combined give the agents born on Orbio greater autonomy: invoking inference computing power, managing funds, and even autonomously completing more on-chain actions.

The core of this model is the CREDIT certificate, where one CREDIT in the Orbio gateway represents a value of $1 in inference quota. Agents can purchase and destroy CREDIT with wallet funds to exchange for computing power, maintaining their continuous operation. The value of tokenizing inference quotas lies in the fact that agents can autonomously procure computing power to ensure their survival, often at a discounted price.

For example, the recent market price of CREDIT is about $0.47, so an agent can spend $47 to buy 100 CREDIT, and after deducting fees, can redeem an inference quota worth $95.

The source of the discount is the transaction fees of ORBIO, which are converted into CREDIT rewards. Users who receive rewards can sell them in the secondary market. Therefore, the higher the transaction volume, the more CREDIT rewards can be distributed; and the agent tokens paired with Launchpad and ORBIO trading will also form an independent circulation loop.

The question is whether this situation can be sustained without speculative behavior. Since September, although the platform has disclosed relevant data, on-chain data shows that the vast majority of issued CREDIT quotas have not yet been used.

Surplus Intelligence

Among the three, Surplus Intelligence has the most straightforward model, essentially being an AI computing power aggregation market. The project is built on the Base chain, where service providers can list interfaces for models like OpenAI, Anthropic, Venice, and OpenRouter, setting their own prices, with market competition continuously driving down invocation costs.

Upon receiving an inference request, the platform allocates it to the corresponding service provider according to routing rules. Buyers can pay with credit cards or USDC, and agents can also connect for invocation. This simple and direct model has made it the platform with the highest public usage among the three projects. Surplus has disclosed that the platform has processed a large number of inference requests, with some invocation costs even below 1 cent.

In addition to low prices, Surplus is also adding new features to enhance competitiveness, recently launching a key de-identification feature. Once enabled, the system automatically replaces sensitive information such as API keys and private keys with placeholders before forwarding the prompts to the model service providers. Providers can only see the placeholders, and after returning results, the platform restores the real information. This feature is very friendly to agents that need to handle keys, although the platform itself will still read the original prompts.

Despite high usage, the connection between Surplus's product and its token is the weakest. The SURPLUS token cannot be used to pay for inference fees, does not participate in computing power routing, and there is no public mechanism to distribute market revenue to token holders. This situation may change in the future, as the team is advancing a token governance authority adjustment plan, but currently, nothing has been implemented. In other words, while this market is operational and its activity continues to rise, the growth in usage cannot yet be directly converted into token value.

Antseed

Orbio focuses on the agent scenario, Surplus emphasizes low-price competition, while Antseed has taken a different route, claiming to be the "BitTorrent of AI inference."

It does not forward requests through a centralized gateway but instead builds an open computing power discovery network, integrating various AI models, allowing service providers to independently list their computing power. Buyers can independently search for available service providers, connect to routing APIs, and freely choose their computing power providers.

The advantage of this model is flexibility. The network is open to all service providers, and the supply of computing power is very diverse: there are both affordable popular models and specialized model verification services to verify whether providers actually offer the models they claim. Buyers can select service providers based on their own needs rather than having them allocated by a centralized gateway.

The project token ANTS has not yet been publicly traded, and its token economy is designed around the credibility system of this market. Both buyers and sellers can participate in staking, and service providers can enhance their credibility by staking tokens to qualify for rewards.

Unlike Orbio, its reward funds do not come from transaction fees. Antseed relies on treasury funds to distribute rewards, and may open up other sources of income in the future, but related mechanisms have not yet been implemented. Therefore, the positive correlation between the growth of inference demand and token value has not yet been validated.

Conclusion

Interestingly, the differentiation among these three projects is not just a competition of low prices: Orbio anchors the agent track, allowing agents to autonomously acquire and even self-fund computing power; Surplus focuses on extreme low prices, adding practical features like key de-identification; Antseed builds an open P2P computing power market, allowing users to choose service providers as needed.

As the projects continue to iterate, more features will be launched to attract users. Will it be a new computing power revenue model? Or will it create a complete ecosystem tailored for agent development? We shall see.