Going from there to $70 million works out to roughly 100% growth over two months.
A quick primer for anyone new to the project. Bittensor is organized into subnets, which are specialized marketplaces that each focus on a particular AI task or service.
Each subnet has its own token, known as an “alpha” token, sitting alongside the network’s main token, TAO.
Where the money is coming from
According to SubConnect, the bulk of the growth is coming from subnets that sell compute and inference.
Three subnets lead the pack: Lium (SN51), Targon (SN4), and Engy (SN53). Bitcast, Almanac, Score, and Vanta are also contributing to the total.
Compute and infrastructure subnets account for approximately 78–82% of total verified revenue in the latest report.
Buybacks enter the chat
Some subnets are doing something notable with their outside income. They are using it to buy back their own alpha tokens.
When SubConnect launched its initial Revenue Index, fourteen revenue-generating subnets were already funding buybacks this way. That number has since increased.
The emissions problem nobody can ignore
Bittensor’s network emissions are estimated at more than $300 million a year.
Against that, external revenue covers only 9-23% of the emissions figure. The network is still paying out far more in tokens than it brings in from customers.
SubConnect itself flags this. It acknowledges the challenge of keeping growth going while also improving profit margins and reducing dependence on emissions.
A caveat on the numbers
It is worth being clear about how these figures are built. They are estimates drawn from off-chain payment records.
They are not audited on-chain totals. So the revenue is real enough to report, but not yet verified to the standard a public company filing would require.
What this means
SubConnect still holds to an ambitious target. It projects revenue exceeding $100 million by year-end 2026, based on the current growth trajectory.
The more important metric to watch may not be headline revenue at all. It is the ratio between external revenue and emissions. Against that benchmark, external revenue currently covers only 9-23% of the emissions figure.
The concentration in compute subnets is another thing to monitor. A network where roughly four-fifths of verified revenue comes from one category is exposed if pricing for GPU capacity and inference gets more competitive.