Stake Requirements
Validators must meet a minimum stake threshold to register on a subnet and maintain an active slot. The exact threshold varies by subnet and fluctuates with competition — as more validators join a subnet, the minimum effective stake to remain registered increases because slots are awarded to the highest-staked participants.
On established subnets, the practical minimum to hold a validator slot is often several thousand TAO of combined own-stake plus delegation. Starting validators typically begin with their own capital and grow their delegation base over time. The important number to track is not just your own stake but your total validator-stake (own TAO plus all TAO delegated to you), since this is what determines your slot priority.
Unlike miners who pay a registration fee in TAO (burned), validators stake their TAO and can retrieve it if they deregister. Unstaking settles in the same transaction — there is no lockup or cooldown — but that does not make it liquid capital. Your stake sits in the subnet as alpha, so what you get back moves with the alpha price, and unwinding a validator-sized position sells it through the AMM at a price you push down as you go. Treat it as capital you can only exit gradually, and at a discount if you need it quickly.
Hardware
Validator hardware requirements depend on the subnet’s evaluation logic. At minimum, validators need a reliable server with sufficient CPU, RAM, and storage to run the validator software continuously. For subnets where the validation logic involves running a scoring model — such as a judge LLM or a computer vision model — a GPU is necessary and often a high-end one.
A practical baseline for most subnets: a dedicated server or cloud instance with 8+ CPU cores, 32 GB RAM, and NVMe SSD storage. For subnets requiring GPU-based scoring, budget for an NVIDIA A100 or equivalent. Network connectivity matters significantly more for validators than for miners — validators query multiple miners per epoch, so both upload and download bandwidth, plus low latency to miners, affect scoring quality and speed.
Redundancy is not optional for serious validators. A single machine with no failover is a liability. Production validator setups typically include a primary server, a hot standby, monitoring infrastructure, and automated alerting so that any downtime is detected and addressed within minutes.
Operating Costs
Monthly operating costs for a validator running on a single subnet typically range from a few hundred to several thousand dollars, depending on whether you use bare-metal servers or cloud infrastructure, and whether your subnet requires GPU-based scoring.
The main cost categories are: compute (server or cloud instance costs), bandwidth (validators with many miners to query generate significant outbound traffic), storage (logs, scoring history, model weights), and monitoring tooling. Validators running on multiple subnets multiply their compute costs but benefit from amortizing fixed infrastructure costs across more revenue streams.
Time is also a meaningful cost. Maintaining a production validator requires ongoing attention: keeping software updated, responding to subnet protocol changes, monitoring scoring quality, communicating with delegators, and handling incidents. Budget at least several hours per week per subnet, rising significantly if you operate custom scoring logic.
Revenue Model
Validator revenue comes from two streams. The first is emissions on your own staked TAO — the same yield any staker earns, proportional to your stake in the subnet pool. The second, and typically more significant, is the take rate: a percentage cut of emissions generated by TAO delegated to you by other stakers.
The take is a cut of emissions, not a yield on the delegated principal — which is the arithmetic people most often get wrong. Say delegators have staked 10,000 TAO with your hotkey, and over a year the emissions attributed to that delegation come to 500 TAO. At an 18% take you keep 90 TAO and the delegators receive 410. Halve the subnet’s emission rate and your revenue halves with it, whatever the delegated amount says.
Your take is capped rather than fixed: MaxDelegateTake is 11796/65535 — 18% — and has been capped there since runtime v320, and you may set less; the current ceiling sits on the chain parameters reference. At scale the take stream is usually far larger than the yield on your own stake, so the business is won by growing the delegation base rather than by the rate you charge on it.
This makes the validator business model more like a service business than a pure capital deployment: your return scales with the size and loyalty of your delegation base, which in turn scales with your reputation for uptime, honest scoring, and transparent communication. The next steps cover how to build that reputation.