Bittensor’s governance cracked this week when a major subnet operator, Covenant AI, pulled its three subnets, accused co-founder Jacob Steves of sabotage, and triggered a 15% price drop. On This Week in Startups, Jason Calacanis called it a “trip and fall” but acknowledged the core tension: anonymous, global talent networks depend on founder-run infrastructure. The protocol now needs staking mechanisms to prevent future “rug pulls.”
“If you randomly set one weight to nine billion, you don't get a better model; you get junk.”
- Vitalik Buterin, The a16z Show
The Bittensor drama is a microcosm of a larger failure. According to The AI Daily Brief, companies are dumping 93% of their AI budgets into infrastructure and models, leaving just 7% for training the humans expected to use them. A KPMG study found 75% of CEOs admit their AI strategy is “more for show,” while 44% of Gen Z workers actively sabotage corporate AI tools they find useless.
This human failure coincides with a Wall Street reassessment. Fears that AI would destroy incumbent SaaS giants have faded. AWS CEO Matt Garman argued on The AI Daily Brief that deep domain knowledge gives existing firms an edge, not a death sentence. The cybersecurity sector, in particular, is seen as a winner as AI expands the attack surface.
The real bottleneck to decentralization isn't software but physics. On The a16z Show, Guillaume Verdon argued that as long as state-of-the-art AI requires hundred-kilowatt data centers, power will stay centralized. His company, Extropic, is betting on superconducting hardware to achieve a 10,000x efficiency gain. Vitalik Buterin agreed, advocating for “verifiable hardware” with cryptographic attestation to prevent surveillance.
“Acceleration without intentionality breaks the delicate structures that make human life valuable.”
- Vitalik Buterin, The a16z Show
The path forward requires new hardware to enable personal sovereignty and new governance to manage human incentives. Bittensor’s stumble is the first test of whether decentralized AI can build systems robust enough to survive its own founders.
