A Bittensor subnet is a separate marketplace inside the network, built to produce one kind of digital output, and it is the unit an allocator should evaluate, not the network as a whole. Five questions decide whether a given subnet deserves a place in a diligence file: what it is paid to produce, how that output is scored, who does the scoring, how concentrated the rewards and stake are, and how you would get out.

What a subnet is

Bittensor describes itself as a network where many independent subnets each run their own task, such as serving a model, storing data or running a prediction contest. Participants who produce the output, and participants who score it, are paid in the network’s token according to how their work is rated. Because every subnet defines its own task and scoring rules, two subnets can have almost nothing in common. A general view of “Bittensor” therefore says little about any one of them.

Five questions to ask about a subnet

1. What is it paid to produce, and who would buy that output?

Rewards that come from the network’s own issuance are a subsidy, not revenue. Ask whether anyone outside the subnet pays for what it produces, and whether that demand would survive if the subsidy shrank. A clear answer names a customer or a use. A vague answer is a finding.

2. How is the output scored?

The scoring rule is the product. Read it as you would read a contract: can it be gamed, does it measure what the task claims to measure, and can its authors change it without notice? A rule that rewards looking busy produces busy work.

3. Who validates, and who can see them?

Validators decide who is rewarded. Check how many there are, whether a few hold most of the stake, and whether the subnet’s own team is among them. The question is the same one a risk committee asks of any protocol: who controls it.

4. How concentrated are rewards and stake?

If a handful of participants earn most of the rewards, the subnet depends on them staying. Ask for the distribution, not the average, and for how it has moved over the period you can observe.

5. How would you exit?

Find out whether the position can be sold in the size you intend, at what cost, and whether the subnet has a token of its own that trades against the network token. Thin markets turn a paper gain into a price you cannot get. Write the exit path down before the entry, with the settlement and custody steps your own policy requires.

What to do with the answers

Treat the answers as inputs to a position limit, not a yes or no on the whole network. A subnet with a named buyer, a scoring rule you can read and a spread of validators can be sized; one without them can be watched. For the wider case on decentralised AI, see what an allocator should ask before engaging.

About the summit

Proof of Talk is an invitation-only executive summit on capital, digital assets and AI. Abu Dhabi, on 3–4 December 2026 at the Louvre Abu Dhabi, is capped at 2,000 people; Paris, on 2–3 June 2027 at the Musée des Arts Décoratifs, Louvre Palace, is capped at 2,500. Admission is by application and review, with no public ticket sale. Read what Proof of Talk is or the Abu Dhabi edition page.

Bittensor and decentralised AI are on the Abu Dhabi agenda themes; our Bittensor page sets out how they fit the programme. To apply for Abu Dhabi, request access. A person reviews each request and it is not a guarantee of a place.

How this guide was made

Proof of Talk’s editorial team wrote it on 21 November 2026 as a general explainer for decision-makers. It is not investment, accounting, legal or tax advice, and it quotes no market figure. We organise the summit it mentions, so it is not a neutral comparison, and it makes no claim about any other event or company.