Dolphin and its POD token have become a useful test case for the crypto-AI market. In a report, Alea Research framed the project as an attempt to balance two forces: token emissions paid to GPU operators on the supply side, and value accrual to POD from customer demand on the other.
The backdrop is a broader shift in how investors are looking at AI-linked crypto assets. Alea Research said VVV’s gains in both price and product adoption have pushed liquid investors to revisit the sector and consider whether these tokens can capture value without leaning on the older DePIN-style model of heavy inflation. Early-stage supply incentives still matter for bootstrapping a network. But holders now want demand from actual product usage to offset a larger share of issuance.
Dolphin’s recent trading action has made that question more concrete. POD doubled after its Upbit listing, then pulled back sharply from a local high. The report treats that move as a chance to reassess the asset and weigh speculative premium against the mechanics of revenue-driven accumulation.
How POD is meant to work
POD is the token issued for Dolphin’s AI inference network. Customers pay for AI responses, and Dolphin routes those requests to GPU operators running the selected open model. The key detail is what happens to the payment flow: non-POD customer payments are directed into POD purchases, while customers who pay directly in POD skip the market-buy step.
Separately, Dolphin’s treasury pays GPU operators in POD for running the models. In other words, the buy side and the reward side are not the same transaction. Alea Research argues that this separation is what makes net token demand depend on whether operators hold or sell the POD they receive.
POD reached $0.9545 on Oct. 4 and closed at $0.6114 on Oct. 8, down 35.9% from that day’s high. Upbit listed POD on Oct. 2, and Bithumb listed it on Oct. 6.
Unlike many older AI tokens that never moved past roadmap promises, Dolphin already has a live product. It opened API access on Oct. 11, 2026, offering DeepSeek V4.1 Flash. The first 1,000 registered users received $30 in free credits, and coupon pricing was listed at $0.01 per million AI text tokens for input and $0.07 for output.
Inference network, staking, and operator constraints
Dolphin’s inference network runs user-controlled open-weight models that anyone can download and operate without needing POD. The report notes that Dolphin and Venice jointly developed and released Dolphin Mistral 24B Venice Edition in April 2025. Venice distributes that model, but POD holders do not receive a share of Venice revenue.
That model currently powers a Flipper closed beta that includes coding credits and a hosted sandbox environment.
Alongside inference, Dolphin funds treasury-backed synthetic training data tasks. Those tasks run in a network where GPU operators participate by earning and staking POD. Validators check their work, and xPOD stakers collect fees.
Before claiming liquid rewards, operators must stake four weeks of expected income and pay a 20% fee. That fee increases the POD reserves held for xPOD stakers. If an operator fakes work, the stake can be slashed. Withdrawing an existing stake requires a three-month cooldown, followed by a one-week exit window.
Venice’s VVV model: staking for inference capacity
To show how Dolphin’s design could evolve, the report compares it with Venice and VVV.
Venice is effectively selling access to open models. Users stake VVV to mint DIEM, and each DIEM gives the holder $1 of inference capacity per day. That turns inference into a staked, prepaid capacity model rather than a pay-per-request model.
Venice does not charge per token used. Instead, any user or AI agent can stake VVV and claim a proportional share of the network’s daily inference capacity. That capacity is measured in Diem, where 1 Diem is roughly equal to $1 of usage. If a user stakes 1% of all VVV, that user can draw about 1% of daily capacity.
Stakers earn yield from newly issued VVV, now set at 2 million VVV per year. Allocation shifts with utilization: near 0% and 100% utilization, the split leans toward stakers; near the 50% target, it leans toward the Venice treasury to fund expansion. At 0% or 100% utilization, about 80% goes to stakers. Near the 50% target, about 80% goes to Venice.
As Venice adds compute, total Diem rises, and the inference share represented by each staked VVV expands with it. The system also allows users to stake VVV to mint sVVV, then lock sVVV to mint tradable DIEM tokens. Those DIEM tokens are themselves staked to receive the daily $1-per-DIEM allowance, turning capacity into an ownable and transferable asset.
To counter inflation, Venice runs a revenue-funded programmatic buyback-and-burn system. Pro, Pro+, and Max subscriptions cost $2, $5, and $10, respectively. In addition, every $100 of API credits purchased triggers a $5 burn. The report says about 33.91 million VVV has been burned so far, roughly 20% of supply, though that figure includes discretionary burns, with the broader effect of pushing VVV toward net deflation.
Venice also cut annual issuance from 14 million at launch to 2 million VVV starting Oct. 1. The report says September burns offset 17.6% of minted supply, including customer-funded burns and discretionary burns that took place before the Oct. 1 issuance cut. Alea Research describes the setup as a two-step path: reduce emissions first, then scale buybacks as product adoption grows.
How much customer spending becomes POD demand
The report then returns to Dolphin and asks a direct question: how much customer spending actually turns into token demand?
Dolphin is described as a point-to-pool system that reuses idle GPUs. Unlike DePIN models that connect compute buyers to rented nodes from providers, each model in Dolphin runs on a GPU pool. Nodes are assigned tasks randomly based on availability and receive POD rewards from the treasury according to the number of inference tokens they process.
Alea Research uses Dolphin’s May 11 pricing for Qwen 3.6 35B as an example. The model was priced at $0.70 per million AI text tokens, with $0.50 allocated to operators. At that rate, treasury-funded operator rewards equal $71.43 for every $100 of customer purchases, because the reward-to-revenue ratio is 0.50 / 0.70 = 0.7143.
After the 20% liquid claim fee, operators receive $57.14 and xPOD receives $14.29. If operators sell what they receive, net POD buying falls to $42.86, calculated as 100 minus 57.14. If fee tokens are then redeemed and sold as well, the figure drops again to $28.57, or 42.86 minus 14.29.
The report expresses the same result as 100 − (100 × 0.50 / 0.70) = $28.57, or 28.6%, assuming the POD price is unchanged and operators have already funded their bond. Even in that worst-case scenario, every $100 of customer spending still creates $28.57 of net POD demand. If operators keep 25% of rewards, net demand rises to $46.43.
Alea Research says this works because Dolphin sends 100% of non-POD customer payments into POD buybacks while paying GPU operators separately in POD from the treasury. Since the two sides are split, the treasury can distribute more POD than buybacks absorb, and the net effect depends on whether operators sell.
Token allocation, relative performance, and peer comparisons
On token allocation, Dolphin assigned 87.5 million xPOD and 37.5 million POD to the team, with a cliff in May 2028. Each stream releases 1% at the cliff, and the remaining 99% unlocks over 10 years.
Total supply is listed at 500,000,000 POD, broken down as follows:
- Treasury: 214,634,419 POD, or 42.93%, reserved for node emissions and future financing.
- Team: 125,000,000 POD, or 25.00%, for current team allocation and future hiring, vesting through 2038.
- Seed round: 117,600,000 POD, or 23.52%, sold in June 2024 for $886,000 and fully unlocked from June 2025.
- Uniswap v4 pool: 42,765,581 POD, or 8.55%, for on-chain liquidity.
From May 11 to Oct. 8, POD rose 77.8%, while VVV gained 32.3%. Using each token’s U.S. dollar price on May 11 as the starting point, POD posted higher returns on 92 of 151 daily observations.
The report also compares Dolphin with Akash and Render. Akash sells cloud capacity, while Render uses customer task payments to buy and burn tokens. Over the same period, AKT fell 13.8% and RENDER fell 6.5%. Alea Research says their BME, or burn-minus-mint equilibrium, makes the token a mandatory settlement medium because each task goes through burn and mint. Dolphin separates buyers and sellers through buybacks on one side and treasury emissions on the other, while Venice bypasses per-transaction token flow entirely through a one-time staking model with daily capacity withdrawals.
Listings and trading concentration
The final section tracks POD’s exchange rollout and where volume has concentrated.
After the Upbit listing triggered a gain of more than 100%, MEXC announced POD spot trading on Oct. 2, and HTX followed on Oct. 4. MEXC also announced a perpetual futures market on Oct. 2, and Gate said it would launch perpetuals the same day.
According to the report, Upbit accounted for 52.1% of reported spot volume across 11 spot pairs, concentrating more than half of trading activity on a single venue.
The report does not present Dolphin as a settled success. Instead, it treats the project as a live structure under market test: whether user spending can keep translating into token demand, and whether treasury emissions can stay balanced against operator selling, still depends on product adoption and demand outside pure speculation.
