Bittensor's On-Chain DeepSeek Moment: Can TAO's Subnet Architecture Survive Its Own Centralization Crisis?
When Bittensor's Templar subnet finished training Covenant-72B in March 2026 — a 72-billion-parameter language model built without a single data center — it felt like decentralized AI had finally delivered on its founding promise. TAO surged past $340. Grayscale filed to convert its Bittensor Trust into a spot ETF. Then, barely two weeks later, Covenant AI's founder called the whole project "decentralization theatre" and walked out, crashing the token 23% in hours.
The whiplash encapsulates everything happening inside Bittensor right now: a network that is simultaneously producing real AI capabilities and struggling with the governance contradictions of building open infrastructure around a single visionary founder.