Why Do AI Crypto Networks Pay Participants More Than They Earn From Customers?

In most AI crypto networks, participants are paid more than the network earns from customers, because most rewards come from newly issued tokens rather than from buyers' money. At Bittensor, the largest such network, subnets' external revenue is estimated at $28–35 million a year, while annual TAO issuance is worth more than $300 million. Token holders cover the gap through dilution, which is why revenue and emissions must be compared separately.

What are emissions, and how do they differ from revenue?#

Emissions are new tokens a protocol mints on a preset schedule and hands to whoever does the work: miners, validators, subnet owners. No outside party pays for these tokens; they come from code. Revenue is money external customers pay for a service: model access, GPU rental, data processing.

The distinction matters. Emissions redistribute value inside the network: every new token shrinks everyone else's share. Revenue brings value in from outside. A network whose rewards are fully covered by revenue looks like an ordinary business. A network whose rewards are mostly emissions is effectively subsidizing its participants at token holders' expense, betting that customers will arrive later.

How much does Bittensor earn, and how much does it pay out in tokens?#

According to the Bittensor documentation, the network produces a block every 12 seconds. Since the first halving in December 2025, the reward has been 0.5 TAO per block, or about 3,600 TAO a day. That is roughly 1.31 million TAO a year. CoinGecko shows about 11.34 million TAO in circulation as of September 30, 2026, so annual issuance equals about 11.6% of current supply (our calculation).

The external revenue estimate comes from the SubConnect Bittensor Revenue Index published in late August. As reported by Crypto Briefing, 24 subnets earn $28–35 million a year from paying customers on an annualized basis. Fifteen of those estimates are rated high-confidence, backed by public dashboards and on-chain data. Compute and infrastructure subnets account for $23.1–27.3 million, or 78–82% of the total. The same outlet puts annual TAO emissions above $300 million, with external revenue covering only 9–12% of participant payouts.

Founder Jacob Steeves returned to subnet revenue in a keynote on September 28–29: per Crypto Briefing, two compute subnets, Lium (SN51) and Chutes (SN64), already earn more from customers than from emissions. For the rest, the reverse is true.

Why does the dollar value of emissions move with the token price while revenue does not?#

Emissions are fixed in tokens, not dollars. At the $320.33 TAO price KuCoin reported on September 27, 1.31 million TAO a year is worth about $421 million. At $299.2, CoinGecko's price on September 30, it is about $393 million, and $28–35 million of revenue covers roughly 7–9% of it (our calculation).

Two consequences follow. First, when the token rallies, the dollar gap between revenue and payouts widens even if the customer base has not shrunk. Second, participants who are paid in tokens but have dollar costs (GPU rental, power, servers) sell part of their rewards to cover them. The larger emissions are relative to real demand, the stronger the steady sell pressure the price has to absorb.

Why do subnets spend revenue buying back their own tokens?#

Since February 2025 Bittensor has run Dynamic TAO: each subnet has its own alpha token, and per the documentation a subnet mints up to 2 alpha tokens per block. Payouts split 18% to the subnet owner, 41% to miners and 41% to validators and their stakers.

According to Crypto Briefing, 14 of the 24 revenue-generating subnets route part of that revenue into buying back their alpha tokens. The logic mirrors stock buybacks: outside money buys the token on the market and partly offsets selling by emission recipients. But buybacks only work while revenue exists. If buyback volume is smaller than emissions, supply still grows, just more slowly. Per PANews, the index authors expect 20–25 subnets with buybacks by year-end; that is a forecast, not a fact.

How can you compare revenue and emissions for any AI token yourself?#

The method works for any network that pays participants in tokens, not just AI projects.

1. Find annual emissions in tokens: in the docs, the tokenomics, or block explorer data.
2. Multiply by the current price to get the dollar cost of payouts. Recalculate after every large price move.
3. Find external revenue and check its origin: third-party customers, public dashboards, on-chain payments. Money that subnets or projects pay each other is not external revenue.
4. Divide revenue by the dollar value of emissions. A ratio near 10% means about nine of every ten reward dollars are newly minted tokens.
5. Compare market cap with revenue. For Bittensor, a market cap of about $3.39 billion on CoinGecko against $28–35 million of revenue gives a multiple of roughly 97–121 (our calculation).

A high multiple does not by itself mean a token is overvalued; the market may be pricing in future growth. But it shows how much of the valuation rests on expectations rather than on customers' current money.

What are the limits of these estimates, and what are the risks for holders?#

The revenue figures are an independent researcher's estimates, not audited financials. Only 15 of 24 estimates are rated high-confidence; the rest rely on indirect data. Revenue is highly concentrated: per Crypto Briefing, three subnets (Lium, Targon and Chutes) generate almost half of the verified total. Losing one large customer would noticeably change the picture.

Annualized revenue is extrapolated from recent weeks or months, so a seasonal spike or one-off contract inflates the yearly figure too. The projection of more than $100 million by the end of 2026, cited by PANews, is the index authors' expectation.

For holders, the main risks are dilution (about 11.6% new supply a year on the current schedule), sell pressure from emission recipients, and a valuation that depends on future demand growth. Network governance is still centralized: per Crypto Briefing, Steeves plans to hand control to a decentralized model by December 2027, and emission and distribution rules may change before then. This article is not investment advice.