Jensen Huang Praises Bittensor: Distributed Training of 72 Billion Parameter Model "Remarkable," TAO Soars 20%

NVIDIA CEO Jensen Huang publicly praised the decentralized AI network Bittensor in an All-In interview, stating that its Subnet 3 successfully trained a large language model in a decentralized manner, which is “a pretty remarkable technological achievement.” Following this news, TAO’s price increased by about 20% within 24 hours of the video release, with trading volume surpassing $471 million.

(Background: Alpha Mining — Zhao Changpeng’s YZi Labs invests in TAO ecosystem “Tensorplex,” causing Bittensor subnetwork tokens to double in value)

(Additional context: What is the new king of AI concept coins, “Bittensor,” up to? A detailed overview of 34 subnetworks and future prospects)

In the latest episode of the All-In podcast, Chamath Palihapitiya brought up Bittensor first, and Jensen Huang responded with a rare positive comment.

He said, “Bittensor Subnet 3 successfully trained a 4 billion parameter Llama model, entirely in a decentralized way, with a group of people contributing their idle computing power to complete it together. They are able to run and manage the entire training process in a stateful manner. I think this is a pretty remarkable technological achievement.”

Huang’s quoted number of parameters is slightly off; in fact, Bittensor Subnet 3 completed the Covenant-72B model, with 72 billion parameters, making it the largest decentralized pre-trained LLM on this network to date, with an MMLU score of 67.1.

Decentralization and Proprietary Models Are Not a Choice

In the same podcast episode, Huang also shared a different perspective on the common “open source vs. closed source” debate, saying “These two things are not A or B; it’s A and B.” Decentralized infrastructure and proprietary models can coexist and are not mutually exclusive.

This statement was interpreted by the market as an endorsement of the development path for decentralized AI. TAO’s price surged about 20% within 24 hours of the video release, currently around $300, with a single-day trading volume of $471 million, hitting a recent high.

Recently, Bittensor has been very active. Zhao Changpeng’s YZi Labs has invested in the TAO ecosystem project Tensorplex, causing its associated subnetwork token to surge over 2 times. Huang’s public endorsement this time adds fuel to the entire TAO ecosystem and gives the outside world more concrete imagination about the feasibility of decentralized AI training.