nion
SN95Subnet in development, details not yet published
A subnet trading near its 60-day high without a single public document explaining why. No miners running, no GitHub, no website. Just a runic symbol, a name, and a team declaring: "Inference Better."
// Open sovereign AI inference
nion is Bittensor subnet 95, operated by the team behind Actual Computer (X: @actualinc). The subnet has no active miners and no publicly available documentation as of this writing. It is in pre-launch development.
The simple version: Based on the team's stated direction, nion is positioned as a decentralized alternative to centralized AI inference APIs. Rather than routing requests through OpenAI or AWS to run a model, applications could tap into a competitive network of model-hosting providers. The technical specifics have not been publicly published.
Centralized equivalent: OpenAI API, AWS Bedrock, Anthropic API
How it works:
The mechanism is undisclosed. Actual Computer describes their mission as "Inference Better | Open Sovereign AI" but has not released a whitepaper, technical specification, or open-source code. How miners and validators will operate remains unknown.
- The problem it solves: Centralized AI inference is controlled by a handful of providers with full authority over pricing, availability, and which models they permit. A decentralized alternative would open that market to permissionless participation.
- The opportunity: AI inference is already one of the highest-demand categories across Bittensor. Subnets like Targon and Chutes have demonstrated real developer usage. The demand is there.
- The Bittensor advantage: Decentralized inference creates a competitive market for compute where providers win or lose based on quality and latency, not brand lock-in. No single party can suspend access or reprice unilaterally.
- Traction signals: No miners are currently active. The market is pricing anticipated future value rather than current output. Price is up over 61% in 30 days, signaling staker conviction about what is being built.
Category: Inference and Compute | Centralized Competitor: OpenAI API, Anthropic API, AWS Bedrock
Inference is the most competitive subnet category on Bittensor. Targon (SN4) hosts open-weight model inference with one of the deepest liquidity pools in the network. Chutes (SN64) built the highest-throughput serverless inference layer in the ecosystem, processing over 160 billion tokens per day. nion enters this category without differentiating technical detail on record, but with a team that has already drawn meaningful staking interest from the community.
The name "nion" and its runic symbol ᚅ are deliberate choices. In Ogham script, nion represents the ash tree, associated with connectivity across planes. In the Elder Futhark runic alphabet, ᚅ represents necessity and constraint. Whether this signals a technical design principle or is simply distinctive stealth branding has not been clarified.
Mechanism:
There is no public documentation. The subnet has zero active miners. Actual Computer's only public signal is their tagline: "Inference Better | Open Sovereign AI." No miner code, validator scoring spec, or product surface is currently available to assess.
What the on-chain data does show is a market expressing confidence ahead of any product launch. Net TAO inflows reached +988 TAO over the past 7 days, placing nion among the top subnets for net flow during that period, including +774 TAO in a single day on April 24. The AMM pool holds approximately 17,300 TAO in liquidity, with root proportion at 27.5%, meaning nearly three-quarters of the pool reflects organic staking demand rather than protocol subsidy. Current emission share sits at approximately 2.9% of the 3,600 daily TAO network emissions, reflecting real Taoflow weight. A chain buy rate of 8.26% indicates active buying through the AMM on top of that.
The price is near its 60-day high at approximately 0.02900 TAO, up 21% over 7 days and 61% over 30 days. The 60-day floor was 0.01419 TAO, roughly half the current price. That move happened without a single miner going live. It reflects accumulation by stakers betting on the team's ability to deliver.
- Pre-launch execution: Zero active miners, no public code, no documentation. All market value rests on anticipated delivery. If timelines slip or the direction changes, there is no operational baseline supporting current pricing.
- Documentation gap: A team positioning around "Open Sovereign AI" with no published codebase or technical specification is a tension. The openness of the eventual product is unverifiable until documentation appears.
- Competition: Targon and Chutes are well-established inference subnets with entrenched validator networks, significant liquidity, and existing developer relationships. Differentiation will need to be substantive to carve out a position.
- Stake concentration: Gini coefficient among the top 100 staking addresses is 0.745. Concentrated staking positions can amplify price swings when large delegators move.