Bittensor Subnets Are Already Generating $28–35 Million a Year

For years, the easiest way to measure activity on Bittensor was emissions. Now there is another metric worth watching: revenue from customers buying the subnets’ products or services.

SubConnect’s Bittensor Revenue Index: Volume I, published in August, estimates that 24 subnets are generating $28 million to $35 million in annualized external revenue.

That is revenue from things people and companies are buying, including GPU compute, inference, enterprise AI, media campaigns, and software. This number is not from TAO emissions or staking yield.

The revenue reported in this article is an estimate, not audited financial data. But it gives Bittensor something it has largely lacked, which is a baseline for measuring whether subnets are becoming businesses.

Where the revenue is coming from

The majority is concentrated in compute and infrastructure.

Category	Subnets	Estimated annual revenue
Compute & Infrastructure	9	$23.1–27.3M
Enterprise AI	6	$2.4–3.7M
Applied AI	10	$2.6–4.0M

Three subnets account for a large share of the total.

Subnet	Category	Estimated annual revenue	Margin	Revenue-funded alpha
Lium (SN51)	Compute & Infrastructure	$8–10M	Low	Yes
Targon (SN4)	Compute & Infrastructure	$5.5–6M	Low	Yes
Chutes (SN64)	Compute & Infrastructure	$4–5M	Low	Yes

Together, they represent close to half of the reported total.

The rest of the network is more fragmented.

Subnet	Category	Estimated annual revenue	Margin	Revenue-funded alpha
GM (SN28)	Compute & Infrastructure	$2.5–3M	Low	Not disclosed
Blockmachine (SN19)	Compute & Infrastructure	$1.3M	Low	Yes
Green Compute (SN110)	Compute & Infrastructure	$1.2M	Low	Yes (10% of revenue)
Score (SN44)	Enterprise AI	$1–2M	High	Planned
Synth (SN50)	Applied AI	$1–2M	High	Yes ($100K buy)
Vanta (SN8)	Applied AI	$750K–1M	Medium	Yes
AdTAO (SN21)	Enterprise AI	$500–700K	Medium	Yes (17% of profits)
Bitcast (SN93)	Applied AI	$500–550K	Low	Yes (64.5% to date)
Engy (SN53)	Compute & Infrastructure	$400–600K	Low	Yes (100% of revenue)
Yanez (SN54)	Enterprise AI	$300–400K	High	Yes (20%+ of receivables)
Almanac (SN41)	Applied AI	$275–325K	Low	Yes (100% of fees)
Hippius (SN75)	Compute & Infrastructure	$200K	Low	Revenue-share announced
Desearch (SN22)	Applied AI	$65–75K	High	Planned (50% of surplus)
ReadyAI (SN33)	Enterprise AI	$50–100K	High	Yes (75% of pipeline revenue)
Ditto (SN118)	Applied AI	$43K	Low	Yes (100% of revenue)
Leadpoet (SN71)	Enterprise AI	$30K	High	Planned

The full index includes 24 revenue-generating subnets, with several others declining to disclose figures.

Revenue is changing what “successful” means on Bittensor

A subnet can have impressive emissions, strong miner activity and a high-profile product, but those things do not necessarily mean it has a business.

Revenue introduces three harder questions:

1) Does anyone pay for it?
2) Can the subnet make money after its costs?
3) Does some of that money flow back into the subnet economy?

That last question is particularly important.

When the report says a subnet has “revenue-funded alpha,” it means the subnet is using money generated from customers to buy its own alpha token. Instead of relying entirely on emissions to create demand for the token, part of the business’s external revenue is being used to purchase alpha.

The exact mechanism differs by subnet. Some use 100% of their fees or revenue, while others allocate a percentage of profits, receivables, or revenue. Synth, for example, has made a fixed $100,000 alpha purchase, while Yanez directs 20% or more of receivables toward alpha. Almanac, Engy, and Ditto are among the subnets listed as using 100% of fees or revenue.

The idea is simple. A customer pays for a subnet’s product, the subnet generates revenue, and some of that revenue is used to buy alpha. External demand for the product can therefore create demand for the subnet’s token.

That creates a flywheel.

Customers → revenue → alpha purchases → stronger subnet economics.

It also explains why revenue alone is not enough. Compute currently generates the largest dollars, but SubConnect rates its margins as low. Software-oriented subnets generate less revenue but can retain more of it.

The strongest subnets may ultimately be those that can do all three: generate meaningful revenue, maintain healthy margins and turn part of that revenue back into their own economy.

The gap is still enormous

Bittensor currently produces roughly 1.31 million TAO a year in emissions following the 2025 halving. At the approximately $250 TAO price used in the report’s contemporaneous analysis, that represents roughly $329 million in annual token issuance.

Against $28–35 million in external revenue, subnet businesses currently cover only around a tenth of the network’s incentive economy.

That is the uncomfortable part of the story. Bittensor is generating real commercial revenue. It simply has not reached the point where that revenue can replace emissions as the dominant economic engine.

What comes next

Revenue is likely to become a more important measure of subnet success as Bittensor moves further away from an economy dominated by emissions.

The question will increasingly be whether a subnet can attract customers, retain them, and generate enough margin from that demand to sustain its operations and support its token economy.

The current $28–35 million figure shows that this transition has already begun, but it remains small relative to the value of annual emissions. For the network, the next milestone will be growing revenue to a point where commercial demand becomes a meaningful counterweight to emissions. SubConnect’s stated goal is for subnet revenue to approach 100% of emissions within two years.

SubConnect plans to repeat its revenue census every six months, giving the ecosystem a way to track that progression. Community projections have also pointed to $100 million in revenue and 35–40 revenue-generating subnets by the end of 2026, although those are expectations, not findings from the first index.

If subsequent editions show sustained growth in both the number of revenue-generating subnets and the dollars they generate, it would provide stronger evidence that Bittensor’s subnet economy is becoming increasingly supported by customers and less dependent on incentives.