Bittensor defines a marketplace for digital commodities. It doesn’t contemplate a specific digital commodity in its design but aims to be general and create a protocol that is sufficiently broad to accommodate many different types of computational resources, all relying on the TAO token for incentivization.

It has 32 different subnets, covering different types of computational or virtual resources, ranging from text prompting, to translation, to audio, to vision, to transcription, and many more. (see: https://taostats.io/subnets/).

Like PoW, miners are paid to create computational resources. However, in Bitcoin, miners just have to create a valid nonce proving a sufficient amount of arbitrary work done. In Bittensor, miners produce responses to queries (typically machine learning or AI related), so the “work” done is theoretically useful.

Tokenomics: it has a fixed supply of 21m, with 6.3m issued. The FDV is $14b and the free float market cap is $4.2b. It’s traded on a variety of exchanges like KuCoin and Gate, so not the tier 1s so far. You can also buy wTAO on uniswap and other defi venues. It’s trading near its ATH. 88% of free float supply is staked or delegated, and only 11.3% is freely circulating. It follows a similar halving cycle to bitcoin. Miners need to acquire TAO in order to participate in the network.

Network revenue: neither Messari (https://messari.io/project/bittensor) nor Dune (https://dune.com/spencerswanson/intersection-of-web3-and-ai-decentralized-compute) , nor Tokenterminal (https://tokenterminal.com/terminal/projects/bittensor) appear to have metrics on actual revenue or utilization figures. The official explorer taostats does not have a dashboard on this either. It seems that these must be de minimis. This is a significant concern. In theory, the token valuation could come from demand to stake TAO in order to be a miner, but you’d want to see meaningful revenue accreting to miners to justify TAOs valuation on that basis. Unless I’m missing something (and I looked thoroughly) I don’t see any network-wide revenue. Joseph Jacks says it’s generating >$150m in revenue (https://twitter.com/JosephJacks_/status/1751076985286369495) but I think that’s false. I’m guessing he’s referring to staking rewards, which is really just redistribution between existing holders.

Conceptually: I understand why an open marketplace for AI inference would make sense. However, the challenge would be demonstrating why this is better than a bilateral relationship with an AI cloud provider like Core Weave, Lambda, or any of the hyperscalers or large AI companies. Even if they do end up participating in Bittensor, this seems like an additional layer of intermediation (versus just buying inference from them directly).

Conclusion: it seems like TAO is being valued based on AI hype and excitement, which is considerable based on NVIDIA performance. It is one of the more “mature” AI blockchain projects, and it seems to be one of the perceived winners in the category. It doesn’t seem like it’s being valued based on network revenue or any more “fundamental” financial metrics. Conceptually, I have a hard time justifying it. I don’t see why it is a significant improvement over the existing, highly efficient market for AI inference. Based on the extremely advanced valuation (and, in my opinion, AI hype reaching a topping stage), I am not particularly constructive on the token in the medium term. I would want to see more transparent financial metrics before I could devise a fuller opinion and certainly evidence of meaningful network usage. However, since the valuation is mainly hype driven as far as I can tell, it can run up much further in the near term, because no one is actually doing a DCF on this thing – just buying into the excitement and the general blockchain/AI thesis.

Summary:

Narrative: 10/10
Tokenomics: 7/10
Ability to generate hype: 9/10
Network revenue: 0/10
Evidence of usage: 0/10

Do with this information what you will.