Original Title: $POD: The Buyback Engine Powering the Dolphin Inference Network
Original Source: Dolphin
Original Translation: Yuliya, PANews

Editor’s Note: Recently, the AI narrative within the Base ecosystem has experienced explosive growth, particularly with the privacy-focused generative AI platform Venice ($VVV) and its ecosystem projects. As a co-developer of Venice's core default model, the Dolphin Network and its token $POD have shown astonishing performance in May, with market cap skyrocketing from $12.2 million to $192 million, an increase of over 14x. This article details Dolphin Network's unique 'point-to-pool' economic model, token value capture mechanism, and innovative designs ensuring network security through staking and penalty mechanisms. Here’s a detailed breakdown and analysis of the mechanisms:

Peer-to-Pool economic model design.

The Dolphin network is designed as a 'Peer-to-Pool' system, aiming to repurpose everyone’s idle GPUs. Each AI model runs on a GPU provided by the network.

This differs from most AI DePIN networks. In other networks, buyers typically rent a node directly from the provider, establishing a one-to-one 'session'.

· On the supply side, nodes running the same model form a 'pool' to collectively handle task requests. The system allocates tasks randomly based on the availability of nodes, and there is no direct contact between the requester and the node provider. The only standard for nodes to earn rewards is based on the number of AI computation tasks they handle (i.e., inference tokens), with rewards paid in POD tokens from the protocol treasury.

· On the demand side, users utilizing the API purchase quotas directly from the protocol. The Dolphin network accepts various cryptocurrencies for payment, including $POD, $ETH, $BTC, $USDC, $XMR, and $ZEC.

100% of all revenue received by the protocol will be used to buy back POD tokens in the market—this directly offsets the issuance of tokens.

Buyers and sellers are separate, meaning that the POD rewards given to nodes can be more or less than what we earn from revenue.

To illustrate more intuitively, let's look at a specific case of running the Qwen3.6-35B model on the Dolphin network:

· Current cost for running datagen.dphn.ai: $0.50 for processing 1 million tokens.

· The cheapest competing price for similar products on OpenRouter: $1.00 for every 1 million tokens.

· The fee charged to users by Dolphin: $0.70.

· The cost paid to nodes by Dolphin: $0.50.

· Net buyback funds: $0.20 generated for every 1 million tokens.

In other words, the pricing of the Dolphin network is not only 30% lower than the cheapest centralized providers, but also generates a pure profit of $0.20 for every 1 million tokens to buy back POD in the market.

Why is this the best application scenario for DePIN?

This model is seen as a highly promising application direction in the DePIN space, mainly for the following reasons:

· Extremely high demand for AI inference: the market's desire for AI inference computing power is at a tipping point.

· A massive idle computing power pool: The supply of idle gaming GPUs capable of running local AI models is extremely large. This network model feels very similar to previous GPU mining (PoW), but since the output is genuinely commercially valuable AI computation, the potential for profit is much greater.

· Ignoring geographical location limits: Unlike many DePIN networks, the geographical location of AI inference is not important, thus avoiding coverage issues. Due to the high geographical flexibility of AI inference, a few hundred milliseconds of latency has minimal impact on user experience, enabling the Dolphin network to connect global consumers and computing resources, greatly enhancing the scalability and utilization of each node.

· The necessity of liquidity pooling for computation: This is the only way to unlock the largest GPU supply group (gamers and computer enthusiasts). It allows nodes to come online or offline at any time, without needing to ensure fixed online hours like P2P rented nodes. Previous GPU DePIN projects required a 1-to-1 binding between consumers and nodes, which simply doesn’t work for idle GPUs like gaming PCs or data center graphics cards, as the owners may want to reclaim their computers at any time. After all, no one wants to rent a GPU and suddenly lose connection because the owner took back their virtual machine.

Token mechanisms and value accumulation.

POD is the only valuable asset in the Dolphin ecosystem. 100% of the revenue generated by the network will automatically be used to buy back POD in the market. Furthermore, Dolphin has no externally-owned equity structure based on shareholders and will never introduce one in the future.

For POD holders, staking tokens in the xPOD treasury grants multiple exclusive benefits:

· Earn direct automatic compounding dividends from network token buybacks.

· Gain daily AI inference quotas, allowing free use of all models on the network.

· Enjoy premium subscription status in Dolphin's web chat rooms, bots, and other ecosystem applications.

In the design of the tokenomics, Dolphin draws on the essence of numerous outstanding DeFi projects and deeply integrates the parts most aligned with distributed AI inference and training networks:

· Inspired by ETH mechanisms: node operators and validators are required to pay collateral, which will be directly deducted (forfeited) in the event of malicious behavior.

· Inspired by the CRV mechanism: providing reward acceleration features for node operators. Locking POD can double earnings, and based on the collateral earnings ratios of other platforms, a multiplier of 1.5 to 2 times is highly competitive in the market.

· Inspired by the xSUSHI/yCRV mechanism: introducing an auto-compounding staking treasury. Users do not need to manually claim rewards, which means that xPOD (the staked version of the Dolphin token) can directly serve as collateral for node operators.

· Inspired by stAAVE mechanisms: reasonable withdrawal cooling periods and withdrawal time windows are set to ensure the stability of network funds.

· Inspired by vlCVX/veCRV mechanisms: a 'bribery market' has been set up for daily unused xPOD computation quotas. Users can sell their unused computation quotas to earn higher staking returns.

Collateral binding, penalty fines, and reward doubling mechanisms.

In decentralized computing networks, cheating is undoubtedly the biggest threat faced. If left unchecked, node operators might secretly switch to smaller, stripped-down, or even fake AI models to still collect rewards. This would lead to a collapse in output quality, buyers of computing power would flee, and the entire ecosystem's flywheel would never get going.

To address this challenge, the Dolphin network has introduced a 'deductible collateral' mechanism, deeply binding the interests of node operators with the value of POD tokens. If malicious cheating behavior is verified, the node will be directly deducted an amount equivalent to 4 weeks of earnings in collateral. This makes cheating economically unviable.

By default, node operators earn POD that is in a 'bound state'. Once a node accumulates enough bound POD equivalent to 4 weeks of earnings, they can choose during weekly settlements whether to continue receiving bound POD or to receive liquid POD that can be traded at any time.

If you choose to receive liquid POD, the system will deduct a 20% fee. This money goes directly into the xPOD staking treasury, distributed to other stakers and honest node operators who have bound their collateral.

Nodes can further deposit xPOD into binding contracts, which not only boosts their earnings but also grants them eligibility to validate other nodes in the network.

The POD reward multiplier determines how much extra a node can earn beyond the base rewards. This mechanism is inspired by the liquidity provider (LP) acceleration mechanism of Curve Finance, but Dolphin has made specific modifications for the decentralized AI network, adding features like usage-based reward distribution, unified collateral calculation across accounts, and penalty fines for violations.

Simply put:

· Nodes earn base rewards by completing AI computations, verification tasks, and related protocol tasks.

· The system will apply a multiplier to the node rewards you earn based on the number of tokens bound in your account and your earnings ratio.

· When calculating the earnings ratio, the system looks at your average base rewards over the past few weeks and employs a 'fast rise, slow fall' smoothing algorithm: as you take on more computation tasks, your average earnings indicators will rise quickly; but when you are idle, they decrease very slowly.

· If your account has maintained collateral for more than 3 months, and your active collateral is at least 50,000 POD, you qualify to become a validator.

· If your bound collateral is equivalent to 6 months (26 weeks) of earnings, the system guarantees your rewards will be multiplied by at least 1.5 times.

· If your bound collateral exceeds 6 months of earnings, your reward multiplier can reach up to 2 times. The exact amount depends on your relative proportion to other over-bound participants and the absolute quantity exceeding the 6-month target.

All calculations are measured solely by the quantity of POD, and the rewards system does not involve any fiat price oracles. Collateral is calculated per account (wallet), and the resulting reward multiplier applies to all nodes under your account. If you add more nodes, your total account earnings will increase, so you need to proportionally increase your active collateral to maintain the original earnings multiplier.

Finally, the Dolphin network is set to release a paper tomorrow (Encrypted Live-Weight Proofs for Decentralized Inference). This paper will detail a lightweight validation system capable of verifying whether nodes run the correct models on various hardware, surpassing the standard TEE verification that can only be used on enterprise NVIDIA GPUs.

Original link.