I Open at Close 🪄🎩
@deaf
1/2🧵 Bittensor subnets spotlight: - Who sells the shovels in the age of AI acceleration? - Those who make GPUs. GPUs are fundamental in AI development; built to perform many calculations at the same time. This parallel processing power makes them ideal for training AI models, which involves analyzing vast amounts of data simultaneously. Think of a GPU like a massive team of librarians, each scanning different books at once to find patterns as opposed to one person flipping through pages alone. GPU market is dominated by Nvidia with a 90% share, and their enterprise flagship products are pretty costly; e.g., an H100 GPU costs 30 bandos. With the rise of Crypto x AI there are many projects aiming to make GPUs more accessible and even better in terms of performance. Hard but possible, as proved by DeepSeek. I'll highlight Bittensor subnets as it's the most established Crypto x AI network to date. According to Magellan there are 4 subnets focused on decentralized compute 👇
1 reply
0 recast
1 reaction
I Open at Close 🪄🎩
@deaf
1. Subnet 12 by Compute Horde 2. Subnet 27 by Neural Internet (the attached image also belongs to them; i love retro and trees, what can I say). 3. Subnet 51 by Celium Compute 4. Subnet 64 by Rayon Labs which is the most popular one. I want to do a deeper dive to understand their differences and how to evaluate properly decentralized AI projects focused on compute. For now I just wanna say it's very important to have successful projects powered by decentralized and open-source AI, both in terms of hardware (compute) and software (AI agents and apps). Otherwise, we are heading into an era of having the most powerful tech ever in a few pairs of hands that are definitely not looking to hug us. I hope you found this write-up intriguing cheers <3
0 reply
0 recast
0 reaction