Const’s Fight Against Intelligence Fascism Through Permissionless AI

In 2015, a DARPA contractor and Bitcoin whale paid a Simon Fraser University student one Bitcoin to write a blog post about neuromorphic chips.

That student was Jacob ‘Const’ Steeves, Bittensor’s co-founder, and the encounter set in motion the specific fusion of Bitcoin mining and machine learning that would eventually become the network he now runs.

In his appearance on bitcoin.com interviews, he covered the full origin story, the architecture of the protocol, the Sam Dare rug pull and what came after, and a political warning about what he calls “intelligence fascism.”

Highlights From the Interview

The interview moved through Const’s early exposure to neuromorphic computing, the mining-plus-AI insight, the subnet architecture, the rug pull recovery, and the political stakes of what Bittensor is trying to prevent.

1. Bitcoin is roughly 9,000 times more efficient at producing FLOPS than any company on Earth: The Bitcoin network sits at about 1,000 times the size of every supercomputer combined, built for roughly $50 billion if you sell every mined Bitcoin at issuance price.

That efficiency, born from permissionless competition, became the specific primitive Bittensor set out to redirect at AI instead of hashes.

2. The origin story runs through neuromorphic hardware and memristors: After studying AI at Simon Fraser, the Const cold-called the head of the DARPA SyNAPSE program (which had funded IBM’s True North chip up to a billion dollars) and got hired to work on liquid-hardware computing.

The technology hinged on a circuit component called a memristor, which behaves like a synapse expanding when current flows through it.

3. The Cohere story shaped the whole thesis: Before launching Bittensor, the co-founder tried to convince the Cohere team (Aidan Gomez and colleagues) to build Bittensor with him.

They chose the classic AI company path instead. His argument then was that nation-state fiat money would inevitably outspend a Canadian AI startup, so any real alternative needed a permissionless edge OpenAI could not replicate.

4. Bittensor’s architecture is a two-level mining tree: Subnets function as encapsulated Bitcoins, each with a defined reward function measuring whatever the designer wants: GPU count, inference speed, storage capacity, model performance.

TAO Mining 101: What Miners Actually Do on Bittensor

The Bittensor blockchain sits above the subnets, and subnets themselves compete to mine $TAO emissions based on the market price of their token.

5. The best subnets are those where the mechanism is the product: Compute, inference, and storage are commodity markets where permissionless competition beats corporate procurement.

Subnets that build a product first and slap a mechanism onto it after have not worked well historically. The mechanism must sit at the core.

6. Sam Dare’s rug pull was the second major one in Bittensor history: Const took personal responsibility on air, explaining that he had backed Dare to run one of the most important subnets.

Subnet Owners Respond to the Covenant Exit

This gave him the initial funding and the mechanism design, and (in retrospect) he should never have handed over the keys. All Dare did was sell his own shares, which anyone in the world can technically do.

7. The Conviction upgrade closed the rug-pull loophole: Subnet owners can now lock their funds to signal long-term commitment, which prevents the sell-everything-in-one-block pattern that hurt token holders in previous incidents.

Alpha token holders also gained voting rights over the subnet itself, tightening the link between the project and its community.

8. $TAO value accrual runs through two channels currently, with a third coming: Roughly $100 million per year in project dividends flow back to $TAO holders against approximately $300 million in annual inflation, keeping Bittensor in a net inflationary phase.

Every subnet purchase also requires buying $TAO first, capturing demand flow through the base token. Utility token integration (paying for compute and inference directly in $TAO) is projected to arrive within six to eight months.

9. On VVV-style token-plus-equity splits (fundamentally opposed): Const was direct about why Bittensor has no exterior equity for the founders. Adding equity alongside the token creates two privileged classes in the community, opens a backdoor for venture capital to control the system without going through the same rails as everyone else, and confuses where the true value of the project lives.

10. Intelligence fascism is the specific outcome he is warning about: Historically, tyrants could not push populations too far because people still held physical power against the emperor. Mechanized warfare and AI productivity are removing that counterweight, allowing economies to function with far fewer workers.

This could bring states and AI giants closer together, leaving individuals with little power to push back. Bittensor solves that.

The Warning Under the Architecture

Bittensor is often called a decentralized AI network, but its political stakes give the technical choices much greater weight. If Const is right, the gap between permissionless and closed AI could determine whether humans retain any economic counterweight at all.

The interview ended with a challenge to OpenAI and Anthropic, arguing that Bittensor products already compete on cost, speed, and bandwidth in AI markets like inference and compute. What remains is aligning those primitives into state-of-the-art, which Const sees as a question of time rather than possibility, and the next twelve months will test that claim.