Step 4 of 14
How Mining Works
Now that you know what Bittensor, subnets, miners, and validators are, this page explains the full process of mining — step by step. How does someone actually go from "I want to mine" to "I'm earning TAO every day"? And more importantly, what do you need to understand so you can help make it happen?
You won't be doing this alone. This page explains the full picture so you understand how everything connects. Your specific role in each step will become clear as you go through the tools training (Nexus, Claude Code, GSD).
1. The Mining Lifecycle
Mining isn't just "turn on a computer and earn money." It's a multi-step process that involves research, setup, monitoring, and adjustment. Here's the complete journey:
- Find a subnet worth mining — You use Nexus to browse subnets, compare emission data, check competition levels, and identify the ones that look profitable. This is research, not coding.
- Study how the subnet works — You use Claude Code to read the subnet's source code. Every subnet has a scoring function (how it judges miners). Understanding this is your biggest advantage.
- Set up the miner software on a server — This means renting a GPU or CPU machine (depending on what the subnet requires), installing the mining software, and configuring it correctly.
- The miner starts doing work — Once running, the miner automatically receives tasks from validators and responds to them. It answers queries, processes data, generates outputs — whatever that specific subnet requires.
- Validators score your work — The validators on that subnet evaluate every miner's output and assign scores. Your score is relative — it's compared to every other miner on the same subnet.
- Based on scores, you earn TAO — Every tempo (~12 seconds), new TAO is distributed. Higher scores mean a bigger share. This happens automatically, 24/7.
- Monitor performance and adjust — You check emission data regularly. If earnings are dropping, you investigate: Is the competition increasing? Did the scoring function change? Is the hardware too slow? Then you adapt.
Analogy: Think of mining like applying for jobs at different companies (subnets). Each company has different requirements, pays differently, and judges you by different criteria. Your job is to find the best company, understand their expectations, and perform well. If one company stops paying well, you look for a better one.
2. Hardware: GPUs and CPUs
Not all subnets need the same type of computer. The two main categories are:
| Hardware | What it is | Used for | Cost |
|---|---|---|---|
| GPU (Graphics Processing Unit) | A powerful chip originally designed for video games, now used for AI calculations | Heavy AI tasks: language models, image generation, data analysis | Expensive — $200-2000+/month to rent |
| CPU (Central Processing Unit) | The standard "brain" of any computer — good at general tasks | Lighter tasks: web scraping, data collection, simple calculations | Cheaper — $20-100/month to rent |
Key points about hardware:
- The team rents GPUs from services like Vast.ai and Targon instead of buying them. Buying a GPU costs thousands of euros upfront, while renting lets you start and stop as needed.
- Cost matters a lot. If you spend $500/month renting a GPU but only earn $300 in TAO from that subnet, you're losing $200 every month. This is why your analysis of profitability is critical before deploying.
- Different subnets, different requirements. Before entering any subnet, you need to know: Does it need a GPU or CPU? How powerful? This information comes from reading the subnet's code and documentation.
Analogy: GPUs are like race cars — fast, expensive, built for heavy work. CPUs are like regular cars — cheaper, perfectly good for many tasks, but they can't win a race against a GPU. Choosing the right "vehicle" for each subnet is part of the strategy.
3. Registration
Before you can mine on any subnet, you have to register. This is like paying an entrance fee to enter a competition. Here's how it works:
- Registration costs TAO. Every subnet charges a fee to register a new miner. This fee is paid once (but you pay again if you get kicked out and want to re-enter).
- Each subnet has limited slots — usually 256 UIDs (unique identifiers). Think of these as 256 seats in a room. Once all seats are full, the room is at capacity.
- If all slots are full, the worst-performing miner gets kicked out when a new miner registers. So if you register and the subnet is full, the miner with the lowest score loses their seat — and you take it.
- Registration fees vary widely. Some subnets cost as little as 0.1 TAO (~$30) to register. Others can cost 10+ TAO (~$3,000+). The fee is set by the network based on demand.
This is why research matters. If you pay 5 TAO ($1,500) to register on a subnet and then discover it's not profitable, that money is gone. You can't get a refund. Good analysis before registering saves real money.
4. Scoring and Rewards
This is the most important concept in mining. Understanding how you're scored determines how much you earn.
- Every subnet has a scoring function. This is a piece of code that validators use to evaluate miners' outputs. It's different for every subnet — a text generation subnet scores differently than a data scraping subnet.
- The scoring function is what you analyze with Claude Code. When you read a subnet's source code, the scoring function is the first thing you look for. It tells you exactly what validators care about: speed? accuracy? completeness? creativity?
- Scores are relative, not absolute. You don't need a "perfect" score. You need a better score than other miners. If there are 100 miners and you rank in the top 10, you earn a big share. If you're in the bottom 10, you earn almost nothing.
- Higher score = more TAO emissions. The TAO flowing into a subnet every 12 seconds gets split among all miners based on their scores. A miner with double the score earns roughly double the TAO.
Your biggest advantage: Most miners just download the default mining software and run it without changes. If you understand the scoring function deeply — what validators are actually looking for — you can optimize the miner to score higher than everyone using the default setup. This is where your analysis with Claude Code creates real value.
5. Monitoring: Are You Earning?
Deploying a miner isn't the end — it's the beginning of an ongoing process. You need to constantly check: Is this miner actually making money?
- After deploying, you check emission data to see how much TAO your miner is actually earning. This data is public — every miner's earnings are visible on the blockchain.
- Nexus shows emission per miner for any subnet. You can see exactly how your miner ranks compared to others, and whether your share of emissions is going up or down.
- If your emissions are low compared to average, something needs to change. Low emissions mean your miner is underperforming relative to competitors.
Common problems when emissions are low:
- Hardware too slow — Your GPU or CPU can't process tasks fast enough, so validators give you low speed scores.
- Wrong configuration — The miner software isn't set up optimally for that specific subnet's requirements.
- Scoring function changed — The subnet updated its code, and now validators judge differently. What worked last week might not work today.
- Too much competition — More miners joined the subnet, and your relative ranking dropped even though your actual performance didn't change.
Mining is not passive income. It requires regular monitoring and adjustment. Subnets evolve, competitors come and go, and scoring rules change. The miners who earn the most are the ones who pay attention and adapt quickly.
6. The Economics
Before entering any subnet, you need to evaluate whether it's actually worth mining. Here's a quick reference table for the key factors:
| Factor | Good Sign | Bad Sign |
|---|---|---|
| Emission per miner | High and stable | Low or dropping |
| Gini coefficient | Below 0.3 (rewards are shared fairly) | Above 0.7 (one or two miners take everything) |
| Miner count | Moderate (50-200 miners) | Too crowded (500+) or too few (< 10, might be dead) |
| Hardware cost | Lower than your TAO earnings | Higher than your TAO earnings |
| Registration cost | Low (< 1 TAO) | High (> 5 TAO, risky if subnet isn't proven) |
The Gini coefficient might be new to you. It's a number between 0 and 1 that measures how equally rewards are distributed:
- 0.0 = perfectly equal. Every miner earns the same amount.
- 1.0 = perfectly unequal. One miner takes everything, everyone else gets nothing.
- Below 0.3 = good. Rewards are spread fairly among miners. You have a realistic chance of earning well.
- Above 0.7 = dangerous. A few top miners dominate. Unless you can beat them, you'll earn almost nothing.
Simple rule: If
TAO earned per month > hardware cost per month + registration cost, the subnet is profitable. If not, move on. Nexus gives you the data to calculate this before you spend any money.
7. Why This Matters to You
You might be wondering: "Why do I need to know all this if I'm not the one running the mining software?" Because everything you do feeds into this process:
- Right now, you analyze subnets and make recommendations. You use Nexus to find opportunities and Claude Code to understand how subnets work. Your recommendations directly decide where the team's money and hardware go.
- Eventually, you'll go from "this subnet looks good" to actually setting it up. The more you understand the full lifecycle, the more confidently you can move from research to execution.
- The skills you're building are exactly what you need. Nexus analysis teaches you to read the market. Claude Code teaches you to understand the scoring functions. GSD teaches you to manage complex projects. These three skills combined are what make a successful mining strategist.
- You don't need to know how to code the miner. Claude Code handles the technical analysis for you. What you need to know is what to look for (scoring functions, hardware requirements, competition data) and how to make decisions (is this profitable? is it too risky? should we enter now or wait?).
Think of it this way: A good football coach doesn't need to be the fastest runner on the field. They need to understand the game deeply enough to make the right strategic calls. You're the coach. The miners are the players. The subnets are the matches. Your job is to pick the right matches and develop the right strategies.