All Subnets

iota

#9Training

Cooperative decentralized pre-training using data-parallel and model-parallel training across heterogeneous nodes.

Price

$0.031287-1.54%

24h

7d Change

+11.30%

Market Cap

$42.65M

Emission

4.45%

Miners

2

Validators

10

Competitor Mapping

iota

Decentralized

Market Cap

$42.65M

TAM

$30.00B

Model

Token-incentivized

Team

Decentralized miners

Google TPU Cloud + NVIDIA DGX

GOOGL

Market Cap

$3.50T

TAM

$30.00B

Model

Revenue-driven

Team

Enterprise

iota $42.65M12.2 bpsGoogle TPU Cloud + NVIDIA DGX $3.50T

Transforms isolated GPU contributors into a cooperating training cluster, competing with Google TPU pods and NVIDIA DGX. First cooperative decentralized training protocol.

Implied Valuation

If iota captures a share of Google TPU Cloud + NVIDIA DGX's $30.00B TAM

Bear - 0.1% TAM

$30.00M

implied valuation

-29.66%

vs $42.65M current

Base case

Base - 1% TAM

$300.00M

implied valuation

+603.39%

vs $42.65M current

Bull - 5% TAM

$1.50B

implied valuation

+3416.94%

vs $42.65M current

Scenarios illustrate potential scale relative to the traditional market. Actual outcomes depend on adoption, tokenomics, and competitive dynamics.

Thesis

Solves the key limitation of earlier pretraining subnets - miners cooperate instead of competing. Published on ArXiv with formal proofs. Operated by Macrocosmos, the largest subnet operator.

Team

Built by Macrocosmos (24-person team). Co-founded by Will Squires (CEO, MEng Civil Engineering, University of Warwick) and Steffen Cruz (CTO, PhD in Experimental Nuclear Physics, University of British Columbia, former CTO of Opentensor Foundation). Brian McCrindle is also a co-founder. Founded 2024. Team is doxxed.

  • Will Squires - Co-founder & CEO - MEng Civil Engineering, University of Warwick
  • Steffen Cruz - Co-founder & CTO - PhD Experimental Nuclear Physics, former CTO of Opentensor Foundation
  • Brian McCrindle - Co-founder - Background in proteins and pre-training

Funding

No specific VC round disclosed. DSV mentioned as institutional investor partner. Funded through TAO emissions from multiple subnets (SN1, SN9, SN13, SN25).

Traction

Pre-trained models from 700M to 14B parameters beating established baselines. Macrocosmos operates 4+ subnets (SN1, SN9, SN13, SN25). 24-person team - one of the largest dedicated Bittensor development teams. 3.2B+ rows of Reddit/X data open-sourced on HuggingFace (via SN13). Accepted to NeurIPS.

Recent News

  • Mar 2026 - Macrocosmos highlighted as 'most underrated team in TAO ecosystem' for accelerating decentralized training
  • 2026 - IOTA and Templar SN3 attacking same problem (decentralized training) from different angles
  • 2025 - SN9 work accepted at NeurIPS
  • 2026 - SDK launch and subnet dashboards completed

Risk

Cooperative training over commodity internet introduces latency and bandwidth constraints limiting practical model scale.