What is Bittensor

A new way to organize digital work

Most explanations of Bittensor begin too late. They start with blockchains, tokens, miners, validators and a collection of strange terms that make the whole thing sound like it was designed specifically to frighten away normal people.

I think the easier place to start is with something much more familiar: a company.

Suppose you want to build a business that provides an AI service (or any other form of digital work). You need people to build the software, people to run the infrastructure, people to improve the product and probably people to supervise all those other people. Before long you have employment contracts, salaries, managers, recruiters, procurement, legal agreements and an HR department reminding everyone to complete some mandatory training before Friday.

None of this is necessarily bad. Companies are extraordinarily useful machines for organizing human work. But they are machines. A large amount of energy goes into maintaining the organization itself.

Bittensor experiments with a very different way of doing this.

Instead of one company having to find, hire and coordinate everyone who contributes to a digital service, Bittensor allows someone to create an open market around the work they need done. The rules of the market describe what kind of output is valuable. Anyone who can provide that output can compete. If they perform well, they can earn rewards. If someone on the other side wants to use what the market produces, they can buy or access the resulting service.

The strange idea at the heart of Bittensor is therefore quite simple:

Can markets coordinate digital work that we normally organize through companies?

These markets are being built for digital commodities: things such as compute, storage, data, predictions and AI models. Bittensor provides the infrastructure through which many of these markets can exist alongside each other.

From a company to a market

Imagine that you need thousands of GPUs to run AI models.

A normal company might raise money, sign contracts with data centers, buy hardware, hire infrastructure engineers and gradually build up supply. Bittensor gives you another option: create a market in which independent people and companies around the world can compete to provide that computing power.

You do not necessarily need to know who they are. You do not have to employ them. You do not have to negotiate a salary with every new participant or decide whether someone in Singapore deserves a promotion.

You mainly need to answer a harder and much more interesting question:

What does good work look like, and how can we measure it?

Once that can be defined, a market can start doing part of the organizational work for you.

If one participant finds cheaper GPUs, improves its software or builds a more efficient data center, it can become more competitive. Another participant may invent an entirely different approach. Someone else may notice that the existing system is poor and build a competing market around the same problem.

This is where Bittensor becomes more interesting than simply “AI on a blockchain”. The blockchain is useful infrastructure underneath the system, but it is not really the idea I find most important.

The interesting part is the attempt to turn competition itself into a machine for building technology.

Bitcoin showed that this can become extremely powerful

Bitcoin provides the clearest precedent.

Bitcoin did something very unusual: it created a global economic incentive for people to contribute computing power to a network without Bitcoin having to employ any of them. A miner can buy machines, find cheap electricity and join the competition. Nobody at Bitcoin headquarters interviews them first because, fortunately, Bitcoin headquarters does not exist.

The protocol defines the game and the reward. The market does the recruiting.

Over time, this created an enormous global industry of miners competing relentlessly to produce computation more cheaply and efficiently. That computation performs a narrow but important job: it helps secure Bitcoin and makes its transaction history extremely difficult to rewrite.

Bittensor takes inspiration from this mechanism but asks whether the same economic force can be pointed at a much wider range of useful digital work.

Instead of paying a global network primarily to produce hashes that secure one monetary system, why not create similar competitive markets for GPU power, AI inference, storage, predictions, data, cybersecurity, model development and other digital services?

That is the leap.

Bitcoin showed that a protocol can summon enormous quantities of resources from people it does not know, employ or control. Bittensor is trying to make that trick programmable.

Capitalism in its most ferocious form

There is something almost uncomfortably capitalist about this.

A traditional employee has a contract. They may have a fixed salary, a manager, a notice period and some protection from competition. Inside a Bittensor market, the relationship can be much harsher. You contribute something the market values and you get rewarded. Someone finds a significantly better way to do it and your position can disappear surprisingly quickly.

There is no reason for the market to care that you bought expensive hardware six months ago.

That sounds brutal because it is. But this competitive pressure is also one of the reasons I find Bittensor fascinating. It creates an environment where people have a constant financial incentive to make the product cheaper, faster or better. Bittensor’s own writing compares this with Bitcoin miners continuously searching for cheaper hardware and energy because the market punishes inefficiency.

Applied to useful digital services, that pressure could become a formidable engine of innovation.

Imagine that a market pays people for providing fast AI inference. One participant buys better GPUs. Another writes software that extracts more performance from older GPUs. Another finds unused data-center capacity in a part of the world where electricity is cheap. Another develops a completely different architecture and suddenly beats all of them.

Nobody had to schedule an innovation workshop.

The reward gave thousands of independent actors a reason to solve the problem.

This is also why Bittensor can potentially move very quickly. A normal company is limited by its employees, capital, management and ability to recruit. An open market can theoretically pull talent and resources from anywhere. Someone does not need to convince the subnet owner that they have a sufficiently impressive CV. If the system is genuinely open and their solution performs better, they can compete.

Because these markets are open and their mechanisms can be copied and improved, successful ideas can spread between them while weak approaches face continuous competitive pressure.

Of course, there is an important catch. Markets are only intelligent when the incentives are intelligent.

If you reward the wrong thing, people become extremely good at producing the wrong thing.

Designing a market that can reliably tell the difference between useful work and someone cleverly gaming the rules is one of the hardest parts of Bittensor. Open competition does not magically remove bad incentives, poor products, fraud or human stupidity. It can amplify those things too.

So Bittensor does not remove the need for good entrepreneurs. Someone still has to design the product, create the market, decide what should be rewarded and find real users. Teams still build companies around subnets. They still write software, talk to customers and make bad decisions from time to time.

What Bittensor changes is the possibility of outsourcing much of the production side to an open competitive market rather than building the entire organization yourself.

These markets are called subnets

Once the idea above makes sense, one of the most common Bittensor words becomes much less intimidating.

A subnet is basically one of these specialized markets. Think of a product department of a company. Or an R&D department of a company.

One subnet may organize a market around AI computing power. Another can reward people for storing data. Another might create a competition for weather predictions, cybersecurity discoveries, model training or some digital service that does not exist yet.

Each is free to define a different problem and a different way of determining what good performance looks like. This flexibility is important. Bittensor itself does not try to decide what the one correct form of useful computation is. It provides infrastructure on which entrepreneurs can create many different competitions for many different forms of digital work.

That is why describing Bittensor as one decentralized AI model misses most of the picture.

There can be many models. Many products. Many companies. Many markets.

On Bittensor, there are many companies building subnets in lots of different domains. You have companies using their subnet as a continuous adversarial game, meaning they invite ethical hackers to bypass their security systems. They then use these innovative methods to make their product even stronger (see RedTeam & Yanez).

You can find subnets that work on camera models or AI text/video/image detection methods like Score.

More about these subnets later in this beginner’s guide.

What connects them is the underlying Bittensor system and its shared economic asset, TAO. We will get to TAO on the next page; for now, it is enough to understand that Bittensor has an economic mechanism for rewarding participation and connecting these otherwise separate markets.