🚨 WHEN ANTHROPIC'S CO-FOUNDER VALIDATES YOUR THESIS, MAYBE PAY ATTENTION TO
$TAO DIRECT MENTION OF SUBNET.
Jack Clark, Anthropic co-founder, former OpenAI Policy Director, one of the most respected voices in AI, published analysis on decentralized AI training that should be front-page news.
Instead? Crickets
WHO IS JACK CLARK?
- Co-founded Anthropic (the company behind Claude)
- Former OpenAI Policy Director
- Writes Import AI newsletter analyzing cutting-edge AI research
- When he speaks, frontier labs listen
This isn't some crypto influencer. This is a founder of a $350B Valuation AI company.
WHAT DID HE SAY?
Jack analyzed Epoch AI's research on decentralized training and here are the numbers:
Decentralized training growth: 20X per year
Centralized (frontier) growth: 5X per year
Read that again. Decentralized is growing 4X FASTER than OpenAI/Anthropic/Google.
Current:
- Decentralized runs are 1000X smaller than frontier models
- But growing 4X faster per year means...
Math check:
At current rates, decentralized training reaches frontier-competitive scale in few years
@covenant_ai (Bittensor subnet operator)
@tplr_ai highlighted as the key innovations:
DiLoCo:
500x bandwidth reduction
Quantization 10× (fewer bits per message)
Sparsification 10× (Send most important ignore rest)
That’s not 520×.
= improvements compound MULTIPLICATIVELY
500 × 10 × 10 = 50,000× less
Epoch AI's conclusion: Bandwidth won't limit the scale of decentralized training anytime soon.
BITTENSOR CONNECTION
Here's what most people are missing:
Templar, the world's largest active decentralized training network (9×10¹⁷ FLOP/s), is run by Covenant AI.
Covenant AI = Bittensor subnet
Choose Choose the Covenant Index
@covenant_ai @basilic_ai @grail_ai from
@TrustedStake
Jack Clark just validated the exact infrastructure this Bittensor Company is building 🎯
WHY THIS MATTER?
Anthropic is a frontier lab building centralized AI. Their co-founder is publicly acknowledging that:
1. Decentralized training is real and working
2. The gap to frontier is closing at 4X per year
3. Technical barriers (bandwidth) are being solved
4. Specific Bittensor infrastructure (Templar) is leading this
He's citing the competition as credible.
THE MARKET IS ASLEEP
OpenAI valuation: $500B
Anthropic valuation: $350B
Google DeepMind: Hundreds of billions in parent company value
Bittensor market cap: $2.8B
Thats valued at 1% of Anthropic.
Make it make sense.
MATH IS SIMPLE
If decentralized training:
- Grows 20X annually (vs 5X centralized)
- Has solved bandwidth constraints
- Already has 600,000X scale increase since 2020
- Is being validated by Anthropic's co-founder
Then,
$TAO at ,268 is pricing in ZERO probability this thesis plays out.
NO COVERAGE WTFN?
Here's my theory:
1. Crypto stigma media sees
$TAO and ignores the AI breakthrough
2. Complexity decentralized training requires understanding both AI and crypto
3. Vested interests centralized AI labs have $100B invested in their approach
4. Timing most don't realize how fast this is moving
But the math doesn't care about narratives.
WHAT JACK DIDN'T SAY (BUT IMPLIED)
From Import AI 439:
"Decentralized training gives a way for academic, government, independent, or non-tech-industry actors to pool their compute to change this dynamic."
Translation: Bittensor democratizes access to frontier AI development.
Not 5 companies deciding the future.
195,000 wallets participating in the AI revolution.
Imagine:
- Apple's co-founder validated a competing smartphone OS
That's what just happened.
Jack Clark built Anthropic to compete with OpenAI in centralized AI.
This should be front-page news.
I've been saying decentralized AI will get the credit it deserves.
When history writes the story of decentralized AI, they'll point to Import AI 439 (January 5, 2026) as the moment a frontier lab co-founder confirmed what we already knew.
🎯
$TAO
Sources 🔗
importai.substack.com/p/impo…