On-Chain AI Inference: Both Sell Compute, but the Logic Is Worlds Apart

Written by: David Christopher

Translated by: Chopper, Foresight News

Inference markets have become one of the clearest practical directions for combining crypto and AI, with projects using on-chain markets to significantly lower the barriers and costs of calling AI models.

The core model of these markets is mostly reselling discounted API call credits and idle inference compute, while some projects build networks that connect demand directly with independent compute service providers.

This emerging sector has three representative markets: Orbio, deployed on Robinhood Chain, and Surplus Intelligence and Antseed, deployed on Base. Although all three offer low-cost inference services, they differ markedly in service positioning, target users, and tokenomics. Below, we compare the three projects side by side.

Orbio

The inference market built by Orbio primarily serves AI agents. The project began as a token that distributed OpenRouter credits to holders and has since developed into a full protocol: it supports calls to hundreds of AI models, many at steep discounts, and includes a Launchpad where developers can deploy agents and issue tokens paired with ORBIO.

To better serve agent use cases, Orbio has integrated privacy computing features provided by Phala and NEAR AI. It recently partnered with Mosh to connect to Mosh's tool ecosystem, helping new agent tokens build more sustainable token growth mechanisms.

Together, these capabilities give agents launched on Orbio greater autonomy: calling inference compute, managing funds, and even completing more on-chain actions independently.

The core of this model is the CREDIT token, where one CREDIT represents $1 worth of inference quota on the Orbio gateway. Agents can use wallet funds to buy and burn CREDIT in exchange for compute, sustaining their ongoing operation. The value of tokenizing inference quota is that agents can autonomously procure compute to sustain themselves, often at a discount.

For example: CREDIT's recent market price is about $0.47, so an agent spending $47 to buy 100 CREDIT can redeem $95 worth of inference quota after fees.

The discount comes from ORBIO trading fees, which are converted into CREDIT rewards for users to sell on secondary markets. The higher the trading volume, the more CREDIT rewards can be issued; agent tokens paired with ORBIO on Launchpad also form independent circulation loops.

The question is whether this can remain sustainable without speculation. Although the platform has disclosed relevant data since September, on-chain data shows that most CREDIT quota issued to date has not been used.

Surplus Intelligence

Among the three, Surplus Intelligence has the most straightforward model: it is essentially an AI compute aggregation marketplace. Built on Base, it lets providers list model APIs such as OpenAI, Anthropic, Venice, and OpenRouter, set their own prices, and competition continuously drives call prices down.

After receiving inference requests, the platform routes them to the corresponding providers according to routing rules. Buyers can pay with credit cards or USDC, and agents can also connect and make calls. Its simple, direct model has made it the most publicly used platform among the three. Surplus has disclosed that the platform has processed a large volume of inference requests, with many calls costing less than 1 cent.

Beyond low prices, Surplus is adding features to improve competitiveness, recently launching a key redaction feature. When enabled, the system automatically replaces sensitive information such as API keys and private keys with placeholders before forwarding prompts to model providers. Providers only see placeholders, and the platform restores the real information after results are returned. This feature is friendly to agents that handle keys, though the platform itself still reads the original prompts.

Despite high usage, Surplus has the weakest link between its product and token. The SURPLUS token cannot be used to pay for inference fees, is not involved in compute routing or scheduling, and there is no public mechanism to distribute market revenue to token holders. This may change in the future, as the team is advancing a plan to adjust token governance permissions, but nothing has been implemented yet. In other words, although this market has been proven viable and activity continues to rise, growth in usage cannot yet be directly translated into token value.

Antseed

Orbio focuses on agent scenarios, Surplus competes on low prices, while Antseed has taken a different path, calling itself 'the BitTorrent of AI inference'.

Instead of routing requests through a centralized gateway, it builds an open compute-discovery network that connects to various AI models, allowing service providers to independently list compute capacity. Buyers can independently search for available service providers, connect to routing APIs, and freely choose compute providers.

The advantage of this model is flexibility. The network is open to all service providers, and compute supply is highly diversified: it includes both affordable popular models and dedicated model verification services used to confirm whether providers actually deliver the models they claim. Buyers can select providers based on their own needs rather than being assigned by a centralized gateway.

The project token ANTS is not yet publicly traded, and its tokenomics are designed around the market's reputation system. Both buyers and sellers can participate in staking; service providers that stake tokens can improve their reputation and become eligible for rewards.

Unlike Orbio, its reward funding does not come from transaction fees. Antseed distributes rewards from treasury funds; it may open up other revenue sources in the future, but the relevant mechanisms have not yet been implemented. Therefore, the positive link between growing inference demand and token value has not yet been verified.

Summary

Interestingly, the differentiation among these three projects is not just about competing on low prices: Orbio is anchored in the agent sector, enabling agents to autonomously acquire and even self-fund compute; Surplus emphasizes ultra-low pricing, combined with practical features such as key masking; Antseed builds an open P2P compute marketplace, allowing users to select providers on demand.

As the projects continue to iterate, more features will be launched to attract users. Will it be a new compute revenue model, or a full ecosystem tailored to agent development? We will wait and see.

Source: TechFlow