@const_reborn@FacetSir The barrier of entry to mine on all the major subnets are way too high now, which can be seen by the lack of mining UIDs. What made bittensor great initially (major participation) isn’t there now on the majority of subnets. Those running subnets are at fault but it’s the design.
@DrocksAlex2 Noticed this too after being away from the network. Bittensors largest subnet chutes has 30 miners
2nd Templar has 2 miners
3rd Targon has 6 miners
That’s 3 largest subnets 768 UIDs and only 38 miners
Nvidia is worth $4.4 trillion.
OpenAI just raised $110 billion in a single round. More than all of crypto VC since 2022. Combined.
And the #1 AI blockchain is sitting at $2.3 billion. Nobody is paying attention.
Bittensor just ripped 30% in a week. Not on hype. On a real technical milestone.
A team on Subnet 3 just trained a 72 billion parameter language model. Fully decentralized. No central cluster. Anyone with GPUs could join.
That's never been done before. Anywhere.
Here's what most people are missing about Bittensor:
• 21 million max supply. Same as Bitcoin. 4-year halving cycle. No pre-mine.
• First halving already happened December 2025. Daily emissions cut 50%.
• Grayscale filed for a TAO ETF (ticker: GTAO) on NYSE Arca. Bitwise is right behind them.
• Founder stepped down February 2026. Fully headless protocol now. No CEO.
• 129+ active subnets producing real AI products. Not roadmap promises.
• Subnet market cap hit $822 million. Record 27% of TAO's total value.
The subnets are what make this different from every other AI token.
Chutes (Subnet 64) is the #1 inference provider on OpenRouter. Beating centralized players. 85% cheaper than AWS. First subnet to $100M market cap.
Ridges (Subnet 62) builds autonomous coding agents. Outperformed Claude 4 on benchmarks. Backed by Stillcore Capital. $10K+ daily prize pool for miners.
Bitcast (Subnet 93) is building a decentralized creator network. Brands pay for results. Creators earn TAO. No contracts. No middlemen. Up 50% this week alone.
While most of crypto chases memecoins, capital is quietly rotating into AI.
AI tokens led the entire market last week. +7% sector-wide. Bitcoin did 2.5%.
The ETF filing is the part nobody is talking about.
Think about what the Bitcoin ETF did. $42K to $73K in three months. TAO is at $2.3 billion market cap right now. Imagine institutional money pouring into that.
AI is not a trend. It's the infrastructure of the next 20 years. And the largest AI blockchain is still early.
Another step forward for $TAO and Grayscale.
90 days ago, Grayscale Bittensor Trust became an @SECGov Reporting Company, which means today its private placement holding period has been reduced from 1 year to 6 months.
Grayscale Bittensor Trust is open to eligible accredited investors seeking exposure to $TAO here:
https://t.co/E0MIPlS03j
🚨 Bittensor and $TAO history made.
@tplr_ai SN3 just completed the largest decentralized LLM pre-training run in history.
Covenant-72B.
72 billion parameters.
1.1 trillion tokens.
Trained over the open internet.
No data center.
No centralized cluster.
No whitelist.
Anyone with GPUs could join or leave freely.
Let that satisfactorily sink in.
To do this centralized, you need thousands of GPUs connected by 400 Gb/s InfiniBand. Millions of dollars in infrastructure. Permission from whoever owns the cluster.
Templar did it on 500 Mb/s commodity internet connections. 70+ independent peers across the globe. California, London, Frankfurt, São Paulo, Virginia.
The numbers tell everything.
146x gradient compression vs standard methods. 1.2 minute sync time vs INTELLECT-1's 8.3 minutes. 94.5% compute utilization vs 82.1%. A model 7x larger trained 7x faster in communication rounds.
And if you Doubt Bittensor, the results will speak?
MMLU: 67.1. That beats LLaMA-2-70B (65.6). Beats LLM360 K2 (65.5). Both trained in centralized data centers with unlimited bandwidth.
This was permissionless. No application. No vetting. No whitelist. Previous attempts at decentralized training (INTELLECT-1 at 10B, Psyche at 40B) still required approved participants.
Covenant solved the trust problem with Gauntlet, a blockchain-native validator that scores every submitted update in real time. Free-riders get caught. Noise gets filtered. Honest compute gets rewarded.
This ran on Bittensor Subnet 3. The model is open. Apache 2.0 license. Weights on HuggingFace. Paper on arXiv.
Here's what most people will scroll past.
The constraint on decentralized AI training was never physics. It was coordination. Templar just proved that coordination is solved.
If a 72B model can be trained competitively over the open internet with no central authority, the argument that you need Big Tech infrastructure to build frontier AI is over.
This is what Bittensor was built for.
$TAO
Is this the beginning of bull market for Bittensor?
> TAO showing resilience in choppy market
> Subnets total value at ~30% of Bittensor
> Decentralized Compute/Inference PMF
> Instos interest (TAO = DeAI index/AI BTC)
> First Halving has yet to be priced in