Quasar Acquires SN27 and Rebrands It Orion, a Dedicated Data Subnet for Bittensor

Quasar (SN24) has confirmed the acquisition of Subnet 27 and relaunched it under a new identity called Orion, positioning it as a data-generation layer for the wider network.

Orion is built to generate, discover, and curate high-quality training data, first for the Quasar model series and then for any subnet or developer building on Bittensor.

The move addresses a bottleneck the Quasar team hit repeatedly while training on public datasets and limited synthetic supply, which capped how far the models could scale. It reframes an expensive data problem as decentralized infrastructure that the ecosystem can share.

The Data Problem Orion Solves

The Quasar team argues that architecture alone does not make a model competitive, since data ultimately shapes its knowledge, reasoning, and overall intelligence.

Their training so far relied on a mix of public datasets and a smaller portion of high-quality synthetic data from adaptive AI, which proved insufficient for reaching state-of-the-art performance at scale.

Generating a dataset of hundreds of billions of high-quality tokens can run past $100,000 depending on the models, filtering systems, and verification methods involved. Orion exists to bring that cost down while keeping the supply flowing.

How the Incentive Mechanism Works

Orion routes the data-generation workload through Bittensor miners rather than a centralized pipeline, which is where the cost savings come from.

According to the announcement, the mechanism can cut the price of a large dataset from roughly $100,000 to around $10,000, with individual miners each surfacing billions of tokens’ worth of high-quality data.

The original design for Orion and its incentive structure came out of discussions between the Quasar team and Const, alongside continued collaboration with Teutonic (SN3). The team also credited Mark Creaser and Siam Kidd as investors who helped secure the subnet slot.

The Product Layer and the Growth Loop

Orion is not being positioned as a subnet that feeds Quasar alone; it’s building a product on top of its incentive mechanism to sell training data to developers and companies beyond Bittensor.

The stated logic is a compounding loop where a stronger product drives more data demand, which strengthens the subnet, which in turn produces better Quasar models.

SILX AI, the company behind both Quasar & Orion, says it will bring on new team members dedicated to Orion while the core team continues developing both projects together.

Quasar builds the models, and Orion provides the fuel.

What This Means for the Ecosystem

A dedicated data subnet is a notable structural bet, since it treats training data as shared infrastructure rather than something each team sources privately.

If the economics hold, Orion could lower the barrier for smaller subnets and independent developers who cannot absorb six-figure data costs on their own.