Decentralized 72B model boosts AI token narrative

TLDR

A new 72?billion?parameter model trained on a decentralized AI network is giving fresh credibility to the AI + crypto narrative.

1. Bittensor (TAO) subnet Templar unveiled Covenant?72B, a 72B?parameter LLM trained fully on Bittensors decentralized network, with no central compute cluster.
2. The milestone has coincided with a sharp rally in TAO and renewed leadership from major AI tokens such as Render (RENDER) and Artificial Superintelligence Alliance (FET).
3. Sustainability now hinges on real usage: on?chain demand for decentralized models, multiple subnets shipping tangible products, and competition from non?crypto open models.

Deep Dive

1. What The 72B Decentralized Model Is

Templar, a Bittensor (TAO) subnet, announced Covenant?72B, a 72?billion?parameter large language model reportedly trained entirely on Bittensors decentralized miner network using around 1.1 trillion tokens.

The team claims any GPU holder could join training over standard consumer internet, with no centralized cluster or gated access, using a method called SparseLoCo to do local optimization and compress updates before sending them to the network. They argue the resulting model competes with centralized offerings like Metas LLaMA?2?70B as a proof of concept for internet?scale, fully decentralized training.

2. How It Boosts AI Tokens

Coverage of Covenant?72B coincided with a strong move in TAO and a week where AI?linked tokens led sector gains, with names like Render (RENDER), DeXe (DEXE), and Artificial Superintelligence Alliance (FET) posting sizable advances.

In Bittensors design, demand for subnets like Templar feeds back into TAO, because users need TAO to access subnet tokens and participate in the ecosystem. A viral technical milestone plus speculative flows into AI themes created a reflexive loop: narrative strength drove token demand, which in turn pulled more attention to the decentralized AI story.

3. What To Watch Next

Several parallel trends are reinforcing the narrative: Tethers QVAC project is pushing fully local, device?run AI; Nvidia and others are releasing ever?larger open models; and AI agent infrastructure is integrating directly with crypto rails.

For TAO and peers, key signals are:

1. whether Covenant?72B (and successors) see real inference usage and fees,
2. whether other Bittensor subnets ship similarly visible breakthroughs, and
3. how decentralized models compare to centralized and non?crypto open?source competitors on cost and quality.

Speculative froth is a risk: if usage and revenue lag far behind token prices, the sector is vulnerable to sharp mean reversion once the initial hype fades.

Conclusion

A 72B?parameter model trained fully on a decentralized crypto network gives the AI?token narrative a rare, tangible technical win, helping TAO and other AI coins lead recent performance. Whether this becomes a durable trend depends on converting that one?off milestone into ongoing model usage, multiple shipping products across networks, and clear advantages versus traditional AI stacks.