POD connects open AI models to a network that pays people to run them. Its token value is a function of customers buying compute and operators bonding tokens. Most recently, the Upbit listing sent POD from $0.312 to $0.889 in three daily closes, followed by a 31% fall from that peak. Venice already links AI use to token purchases; Dolphin's opportunity is to extend its working network into paid customer services.

Choose the AI exposure before choosing the token
An AI token can finance hardware, reward model builders, buy access to a service, or settle payments between agents. Choosing between them means choosing what kind of AI demand to own. Akash rents compute, Render pays for rendering and AI jobs, and Nosana matches inference jobs to GPU owners. Aethir and io.net coordinate hardware supply, while Bittensor rewards competing networks for producing useful outputs. Their tokens give suppliers a reason to join, through bootstrapping inflationary emissions. The demand side of the equation, however, is a much harder and slower process.

Venice and Dolphin offer AI models and access to inference, the work of producing those AI answers, while Virtuals earns transaction taxes from agent-token launches and trading. Payment standards such as x402 by Coinbase or MPP by Tempo (Machine Payments Protocol) let agents pay for services without requiring a new AI token. Each project's economics depend on who pays, what they receive, and why that payment needs the token. That’s why most of the time a useful AI product and an attractive token are separate questions.

The peer set follows those roles, with VVV offering the closest access-and-inference comparison and AKT, RENDER, ATH, IO, and NOS covering supplier economics. TAO represents model incentives, VIRTUAL agent commerce, and GRASS data supply. On Oct 9, POD's circulating market value was $49.1M, against $293.9M using all 500M tokens. Its reported circulating share is 16.7%, while liquidity in the form of tokens available for trading is much thinner.

Customer spending has different destinations, even among projects that rent out compute. Akash's lease payments fund suppliers, and its March upgrade removed the earlier take rate. AKT burned for credits can return as freshly minted supplier payments. Nosana converts card and USDC purchases into NOS, which suppliers receive; it lists a 10% service charge, while its deployed job contract sets the fee for stakers to zero. Charging customers doesn't automatically give holders income though. The crypto and AI sector ultimately consists of many different businesses and payment schedules that aren't comparable profits. When customers spend less than suppliers earn, someone must fund the difference, usually through inflationary emissions while demand builds out.
What VVV gets right, and what holders actually own
VVV rose 13.9 times between Dec 31 2025 and Oct 8, outperforming AKT, RENDER, TAO, VIRTUAL, and BTC over that window. Venice is the closest precedent for POD's access-and-inference proposition. The comparison starts with what holders receive. Venice's users buy a service, while VVV holders get access, rewards, and exposure to token purchases. Company shareholders own equity, with different rights; another AI token needs its own customer demand and holder benefits to earn the same comparison.

Venice launched VVV with 14M tokens of scheduled annual issuance, then reduced that rate to 2M by Oct 1, an 86% cut in new supply. It also added daily inference credits through DIEM and purchases funded by customer spending. Those changes happened alongside VVV's repricing, connecting improved token economics with the rally. Lower issuance and added purchases improved the balance between new supply and token demand.

Burns still offset only a small part of the new supply. Between Sep 9 and Oct 8, 195,328 VVV was minted vs 34,638 burned, leaving 160,690 net new tokens. New-subscription and API-credit purchases supplied 18,514 of those burns, while a discretionary burn supplied 16,124. The customer-linked channels therefore accounted for 53.5% of all burns. The discretionary portion, however, buys “good-will,” showing the intentionality to reward token holders. Venice's purchase policy assigns 5% of API-credit sales to burns, meaning more sales add token demand even while issuance exceeds burns.

One staked DIEM supplies $1 of API credits each day, and unused credits don't roll over. Minting DIEM locks staked VVV, with 80% of the associated emissions going to the minter and 20% to Venice. Credit holders use inference while Venice pays for the compute. Those AI credits measure service use, while demand for the allowance depends on its price, reliability, and how much inference its holders consume.

For reference, on Oct 9 minting one DIEM required about 438 staked VVV, worth $9,695, with that deposit recoverable by returning DIEM. Meanwhile, DIEM's reported secondary-market price was about $1,619 that day, without the minter's collateral claim. Buying DIEM needs a smaller upfront payment, but that payment buys access; minting instead locks recoverable VVV and retains token-price risk. Users can still buy the allowance without adding a new VVV lock.
It’s worth noting that Venice also has shareholders, and in July Voorhees separated equity's control over cash flows from VVV's access and burn benefits. His company's revenue isn't VVV-holder revenue. Similarly, Dolphin's Foundation has no shareholders and promises all inference receipts for POD purchases. Its treasury uses a 4-of-7 Safe, with six team members and the Foundation as signers. Treasury signers govern the allocations, and the split of purchased POD between reserves and staking yield remains undecided.

Dolphin gives users control over their models
Eric Hartford built Dolphin around this goal: users should choose how their models behave, as his 2023 essay on user-selected alignment explains. In closed AI models, a system prompt sets the main instructions and tone; open weight models let the owner retain and run a particular version, as well as fine-tuning to change learned behavior. Hartford previously worked at AWS, eBay, and Microsoft, and received an a16z open-source grant in 2023. That grant supported his open-source work; it wasn't a venture investment in POD.
Dolphin and Venice co-developed and released Dolphin Mistral 24B Venice Edition, which Venice distributes alongside other developers' models. The public weights can run without POD, and the collaboration grants no share of Venice revenue to POD holders. Dolphin's opportunity is to make controlled models easier to run through a shared network, with its coupon API already offering DeepSeek V4.1 Flash. Dolphin has announced broader paid access and custom-model subscriptions, but has not announced a launch date.

POD coordinates independent providers through payments and bonds that the network can slash for failing its requirements. Operators receive tokens for completed work and bond them as collateral, while validators check that they serve the required models. Stakers hold xPOD, a share of the staking vault, and receive fees when operators claim liquid rewards. Customers need the service, and suppliers need its payment and bonding system; downloading and running Dolphin's open models requires neither a chain nor POD.
The staking vault received 20,358.11 POD in liquid-claim fees from Sep 6 through Oct 6, paid from operator rewards. Over that interval, the POD behind each xPOD share grew by just 0.00793%, before changes in POD's price. Stakers receive POD fees, and therefore their return also depends on the token price. These payments don't measure customer cash receipts.

Dolphin also funds synthetic training data work from its treasury. Its jobs totaled 8.11T AI text tokens from Jun 26 to Oct 8. Large September jobs raised the daily average to 22.65B against a median of 9.55B, while Oct 1 to Oct 8 averaged 14.80B with a 15.37B median. The average fell while the median rose. Hardware, workloads, and scheduling affect these totals, which again don't measure customer spending.

Altogether, Dolphin combines model control with a network designed for model work, competing for developers who want configurable models without managing hardware. Contrast that vs other names. Akash provides general compute, Nosana coordinates inference jobs, and Venice packages access to many models. For Dolphin, repeat paid use at a sustainable supplier cost is the basis for an operating business that ultimately balances customer payments with operator costs to determine whether the network can fund that work without reserves.
A customer buys POD; an operator can sell it
Under Dolphin's announced payment design, all inference receipts paid in another currency buy POD, while customers paying POD skip that market purchase. The network then rewards operators, and some tokens return to market. Operators initially receive bonded rewards, with four weeks of expected earnings in collateral permitting liquid claims. Each liquid claim pays a 20% fee to xPOD stakers, whether the operator later sells or holds the tokens, and longer bonds support reward boosts of up to 1.3 times.

Dolphin's May 11 design example charged $0.70 and paid operators $0.50 per million AI text tokens, though those historical rates aren't the current DeepSeek tariff. A $100 payment outside POD buys $100 of POD and funds $71.43 of gross operator rewards at those rates. If operators bond 25%, $17.86 stays bonded, $10.71 reaches stakers, and $42.86 is liquid. Assuming operators sell all liquid rewards, $57.14 of the original purchase remains unmatched by those sales. The illustration counts token purchases. Other costs and sellers remain outside it, preventing a profit estimate.
Rewards per dollar received and the share of new rewards bonded determine how much buying survives operator sales. At rewards equal to 150% of receipts, with no bonding, operator sales exceed customer purchases by $20 per $100, and treasury reserves fund the shortfall. In this fixed-dollar illustration, higher token prices reduce the tokens needed for a bond; existing holdings can meet it. Net buying follows N = R − W × (1 − b) × 0.8 − X, where R is external-currency receipts, W gross rewards, b the bonded share, and X other sales.

This way, a POD holder gains exposure to operator collateral, claim-fee redistribution, and planned demand from paid inference. At the historical reward ratio and 25% bonding, $10M of yearly receipts, all paid outside POD, would leave $5.71M after assumed operator sales. That equals 1.9% of the Oct 9 diluted value of $293.9M. For now, POD purchases recycle tokens rather than burn them.
Separate the listing price from the operating case
From its May 11 LBank debut through Oct 8, POD's daily closing market value ranged from $8.38M to $74.27M. The median was $16.44M, and 124 of 151 closes were below $25M. A fixed 500M supply limits minting while treasury rewards can still release existing tokens. All 117.6M seed tokens, 23.5% of supply, unlocked on Jun 20, 2025. Recently Upbit expanded market access and lifted prices without establishing new paid inference, adding up to the token’s speculative premium.

Thirteen new Upbit KRW listings from July through September had complete 30-day records by Oct 8, with a median day-30 close 35% below the listing-day close. The middle half ranged from 48% below to 11% below, and ten finished below that baseline. POD remained 9% above its own listing-day KRW close on day six.
On Oct 9, Aerodrome's POD/USDC pool had $154K in reported 24-hour volume, versus $3.11M on Upbit's KRW pair, while Kyber estimated 35,685 USDC for 62,500 POD, 0.91% below the one-POD rate before gas. About 4.04M xPOD entered the exit queue from Oct 2 to Oct 7, with withdrawal windows beginning Dec 31 through Jan 5 (holders can withdraw without selling).
Overall, Dolphin's open models, working network, and Venice distribution give it a credible route to AI users. The VVV precedent defines the economic opportunity: repeat customers buy inference, those receipts buy tokens, and suppliers earn enough to cover their costs. Paid receipts, operator retention, and purchases net of supplier sales determine how much AI use becomes POD demand. As of Oct 9, the listing rally has established stronger market access, while customer receipts and supplier costs remain the missing quantities needed to value the operating case.
Disclaimers
This report on Dolphin was not paid for or commissioned. Alea has no commercial relationship with Dolphin or POD. Members of the Alea Research team, including those directly involved in the analysis above, may have positions in the tokens discussed. This memo reflects sources available on Oct 9, 2026, with each figure stating its own data cutoff.
This content is provided for educational purposes only and does not constitute financial or investment advice. You should do your own research and only invest what you can afford to lose. Alea Research is a research platform and not an investment or financial advisor. Historical returns and listing outcomes do not predict future results. The token-flow examples use stated assumptions; they are neither forecasts nor price targets. Crypto assets carry market, liquidity, smart-contract, and execution risks.


