Problem: Bittensor aims to address two distinct issues with its decentralized computing system. First, there is an issue with Bitcoin, which a lot of people point out that it is wasteful in the computing it uses. Mining Bitcoin takes a lot of energy in setups similar to datacenters just to create an artificial scarcity. While this is one of the advantages of Bitcoin, many critics point out that it is extremely wasteful and doesn’t create any real value for people besides the subjective value of that artificial scarcity. The second problem is the problem of centralized compute limitations. With all of the projects in AI, it is very difficult for hyperscalers to bear the entire brunt of the up-front invested capital. Hyperscaler stocks have gotten hit because of the capex pledges upwards of $200 billion. The massive capex growth rates that hyperscalers have sustained recently can only go so much further before their buildout is capped.
Solution: Bittensor is seeking to fix both of those problems simultaneously by using ideas from Bitcoin's proof-of-work system to develop new AI technology, with the TAO token as an incentive for developers and people who provide compute to the projects on Bittensor. Bittensor projects are called subnets, and there are 128 subnets on the platform. Subnets can be bought with TAO; some have gone for upwards of $1 million for developers to begin a project on the subnet. Being an owner means being a rule setter, allowing the owners to set incentive structures and project guidelines while receiving some of the subnets’ earnings. Then, miners begin a highly competitive process of writing and optimizing code and providing the computational power to run it in order to receive their token rewards. Also, validators check the integrity of the miners and the owner’s rule system. Furthermore, a lot of subnets are open-source, which adds to development further. Overall, Bittensor seeks to shift AI development from centralized systems to decentralized systems, which allows for accelerated code production and a way to crowd-source compute power.
Founders: Bittensor was founded in 2019 by Jacob Steeves and Ala Shaabana. Both worked in machine learning, and Ala is still at Bittensor, while Jacob has since left to found another startup, Affine.
Implications: This project is super exciting for not only decentralized development, but also value systems with crypto. I believe that decentralizing the computing investment for AI is a huge value add because it takes capex pressure off hyperscalers. I also think it is meaningful for hyperscalers to not only have to compete amongst themselves but also against an entire ecosystem of developers. Some subnets, like Chutes and Ridges, pose real competition to the hyperscalers. Chutes provides AWS-like services for 85% less than AWS does, and Ridges has developed a coding platform similar to Claude Code. This also speaks to the open-source progress that has been so prevalent recently. Labs, especially Chinese labs, MoonShot, have been developing leading AI models all for free. If people decide that they value owning their AI agents and work on these open-source projects might become even more popular than the frontier labs’ models. More importantly, open-source’s nature of inviting as many collaborators into one project seems to accelerate the development of new ideas greatly, a great example of this is OpenClaw and how it is the leading agent right now. Overall, I think Bittensor is a very interesting project to look out for, and I would expect some blue-chip subsets to make meaningful progress in competing with centralized AI companies.