Analyzing the Bittensor ecosystem network effect and the decision of AI/ML developers to stay or leave

Author: Seth Bloomberg, Messari Researcher Source: @bloomberg_seth Translation: Shan Ouba, Golden Finance

Since everyone has been talking about Bittensor recently, I want to share some of my thoughts and concerns about it.

For a base protocol to succeed, network effects must be developed. This is the key to attracting and retaining users and developers.For smart contract platforms like Ethereum, network effects come from on-chain liquidity and reliable execution capabilities. For Bittensor, the current sources of its network effects are:

1. AI/ML developers familiar with network mining (i.e. developers who develop AI models and services);

2. A deep and liquid TAO market that can absorb inflationary selling pressure. Less mature networks are generally more impacted by token selloffs.

Since point 2 is broadly applicable to most crypto networks, I’ll focus primarily on point 1. To elaborate, the network effect or flywheel effect created by point 1 is: new subnets benefit from experienced Bittensor AI/ML developers as they continue to produce high-quality output → AI/ML engineers benefit from continued TAO inflation rewards. Hopefully, rising TAO rewards, along with Bittensor’s AI/ML talent, will continue to attract more subnet builders, which in turn will attract more AI/ML talent, and so on.

Whether you’re bullish or bearish on TAO, it’s worth exploring how this network effect might break down. That’s where we’ll get into that next.

Scenario 1: TAO Inflation < Operating Costs

AI/ML talent is attracted to the network because the TAO inflation rewards they receive exceed their operating costs on Bittensor. The math is simple: there is no need to raise funds, acquire customers, or generate revenue at scale. You simply integrate your model into Bittensor's subnet and receive TAO inflation rewards. The underlying assumption is that based on the current (or expected future) value of TAO rewards, you will have some positive profit relative to your (USD-denominated) expenses. If the above math doesn't work, then this is the simplest case for AI/ML talent to leave Bittensor for other opportunities.

Scenario 2: Opportunity Cost Calculation

Suppose you are one of those cool AI/ML developers mining on Bittensor. Or maybe you are a small team of two or three AI/ML developers building products on Bittensor. At some point you have to ask yourself: “What is the opportunity cost of continuing to operate here as a miner?” You know the benefit of staying on Bittensor is purely TAO inflation rewards and TAO price appreciation. Once you have fixed one variable, you then calculate the other: is the benefit of leaving Bittensor and starting a new protocol greater? I know… this may be treasonous to some people, but you would be unfair to yourself if you don’t consider this.

Bittensor has recognized the need to provide additional benefits to subnets and their miners. Dynamic TAO (in the next tweet I will link to a post by Sami that explains Dynamic TAO in detail) seems to be the solution that is recognized.

It is in testnet stage now and may enter production environment by the end of this year or early next year. With Dynamic TAO, each subnet will effectively have a Uniswap V2-style pool of funds, and its subnet token will be paired/priced with TAO. The “dynamic” part of the upgrade here comes from the new dynamic TAO inflation model: the more TAO staked in the subnet pool, the more TAO inflation rewards are allocated to the subnet, and the higher the price of the subnet token. This sounds like a good deal. AI developers can get the potential benefits of the subnet token while retaining the benefits of being in the Bittensor ecosystem (i.e. TAO inflation rewards).

However, pegging the subnet token to the TAO places an artificial cap on the valuation of a single subnet token. Ethereum and its L2, or Helium and its subDAOs, have similar token dynamics/relationships. This is where the opportunity cost calculation is key. If you are one of those awesome AI/ML developers or capable subnet owners, why limit yourself as described above? Why not build your own protocol or network? I bet there are probably a group of VCs who are tired of seeing boring AI and crypto projects, and they may be eager to fund you, and may give a higher valuation than a subnet with a strict/rigid token design. Even better, you’ve accumulated a ton of TAO through mining, done some math, figured you could self-fund the launch of your own protocol, and try to attract some applications/users/revenue before considering outside capital (which is rare in today’s market).

This is the battleground that Bittensor is in the long term in my opinion — designing its mechanisms, incentives, and ecosystem to retain the best AI/ML talent.

I think a recent tweet from @mrink0 also hints at this (if not, I apologize for quoting it out of context; correct me if I’m wrong). To me, people like Nous leaving Bittensor to build their own network suggests that for some Bittensor builders, the incentives to stay on Bittensor are currently not comparable to building/launching your own network.