Bittensor Subnets in 2026: A Reshuffled Map
Bittensor subnets in 2026 look different than they did six months ago. The network's emissions map - which subnets earn TAO rewards, and how much - has been reshuffled by two forces: the high-profile exit of Covenant AI, and the Emissions Refactor that went live on May 13, 2026.
For anyone holding TAO or evaluating the network, the subnet layer is where the real story lives. TAO the token is a claim on the productivity of the subnets beneath it. So the question is not just "where is the price" but "which subnets are actually winning, and why." This article maps the post-Covenant landscape.
The Covenant AI Exit: A Quick Recap
Covenant AI operated three subnets: Templar (SN3), Basilica (SN39), and Grail (SN81). In a public and acrimonious departure, the team - led by Sam Dare - sold roughly $10 million in TAO, withdrew its infrastructure from all three subnets, and accused the network of what Dare called "decentralization theatre."
The exit mattered for two reasons. First, the immediate market shock: TAO sold off sharply on the news. Second, the structural question it raised - if one team can pull infrastructure from three subnets at once and destabilize the token, how decentralized is the network really?
But there is a less-discussed consequence that matters more for the subnet map: when a major operator exits, the emissions it was capturing do not vanish. They get redistributed. Covenant's departure freed up reward flow that other subnets could compete for - and the Emissions Refactor a week later concentrated that flow toward measured performance. The result is a clearer hierarchy of winners.
How dTAO Decides Who Wins
To understand which subnets are winning, you have to understand dTAO - the dynamic TAO mechanism that governs emissions.
Under dTAO, each subnet has its own Alpha token. Alpha tokens trade against TAO in automated market maker (AMM) liquidity pools. The price of a subnet's Alpha token, set by that pool, reflects the market's assessment of the subnet's value - and that price feeds directly into how much TAO emission the subnet receives.
In plain terms: dTAO turns subnet quality into a market. Participants stake into the subnets they believe are productive, which lifts the subnet's Alpha price, which increases its share of emissions. A subnet producing real, valuable AI work attracts stake, commands a higher Alpha valuation, and earns more rewards. A subnet producing little attracts no stake and bleeds emissions.
This is the mechanism doing the work in 2026. It is not a foundation committee picking winners - it is liquidity flowing toward perceived productivity. The Emissions Refactor of May 13 tightened the link, restructuring rewards to concentrate on top-performing subnets and reduce the share leaking to underperformers.
The Subnets Winning in 2026
With the mechanics established, here is where the stake and emissions have been flowing.
SN64 - Chutes
Chutes, built by Rayon Labs, has emerged as one of the strongest performers. Its focus is serverless inference and GPU compute - letting developers run AI models on demand without managing infrastructure. The pitch is concrete and the demand is real: serverless inference is a clear, monetizable use case rather than a research abstraction. Chutes has consistently attracted stake, and under the post-refactor emissions logic that translates directly into a strong reward share.
SN4 - Targon
Targon, operated by Manifold Labs, may be the strongest commercial story on the network. It focuses on enterprise compute and has the kind of traction that is rare in crypto: a roughly $10.5 million Series A, more than 4 million users reached through the consumer-facing Dippy app, and a projected annual revenue figure in the range of $10.4 million. Real users and real revenue are exactly what the dTAO market is designed to reward. Targon is a case study in a subnet that looks less like a crypto experiment and more like a venture-funded compute business that happens to settle on Bittensor.
SN3 - Templar
Templar is the open question. It was a Covenant subnet, and Covenant's exit leaves its operational future uncertain. Templar still exists at the token and protocol level - a subnet does not vanish when an operator leaves - but its trajectory now depends on whether new participants step in to run and improve it. Templar is the clearest example of the post-Covenant risk: a subnet whose value now hinges on succession rather than momentum.
SN51 - Lium
Lium occupies the hardware-heavy end of the network. It runs short-term GPU rental backed by a fleet reported at more than 500 NVIDIA H100s, using a Proof of Compute mechanism to verify that the hardware doing the work is real. In a market where AI compute is the genuine bottleneck, a subnet supplying verifiable H100 capacity has a tangible product. Lium is a reminder that not every winning subnet is a model - some are infrastructure.
The Pattern: Revenue and Real Compute Win
Step back from the individual subnets and a pattern is clear. The subnets winning in 2026 - Chutes, Targon, Lium - share a trait: they sell something concrete. Serverless inference, enterprise compute with paying users, verifiable GPU rental. They are not winning on narrative; they are winning because the dTAO market can see real demand and routes stake accordingly.
The Emissions Refactor reinforced this. By concentrating rewards on top performers, it widened the gap between subnets with a product and subnets with a pitch. For TAO holders, this is arguably healthy: a network that rewards measurable output is a more defensible investment than one that sprays emissions evenly across 128 subnets regardless of results.
The 256-Subnet Expansion: Doubling the Field
Looking ahead, the network is moving toward a 256-subnet capacity - doubling the current 128. More slots means more room for new AI applications to launch and compete.
The expansion is double-edged. On the positive side, it lowers the barrier for the next Chutes or Targon to find a slot and prove itself, increasing the surface area for genuine innovation. On the cautionary side, doubling the field without doubling the demand for AI services risks spreading emissions thinner and creating more underperforming subnets - exactly what the Emissions Refactor was designed to discourage. The two changes are in tension, and how that tension resolves is one of the more interesting things to watch on the network through 2026.
How AIOKA Tracks TAO
At AIOKA, TAO is one of seven markets our councils analyze. We do not trade individual subnets - but the health of the subnet layer is precisely what determines whether TAO the token is worth holding, so the council's read on TAO is informed by the kind of analysis above.
The TAO council monitors the Kraken liquidity floor and a set of risk flags before issuing any verdict. That liquidity-aware framing matters for an asset like TAO, where a subnet-level shock - another Covenant-style exit - can drain depth fast. A system that ignores liquidity would read a post-shock dip as a clean discount; AIOKA's council treats thinning depth as a risk signal that calls for patience.
TAO trades in paper mode within AIOKA while the council builds a verified track record. Every verdict is logged, and you can review the full history across all seven markets at aioka.io/track-record.
The Verdict: A Healthier, Harder Network
After Covenant AI, the Bittensor subnet map in 2026 is both healthier and harder. Healthier, because the winning subnets - Chutes, Targon, Lium - are winning on real compute and real revenue, and the Emissions Refactor rewards exactly that. Harder, because the Covenant exit exposed a genuine fragility, and the 256-subnet expansion will test whether the network can scale its slot count without diluting its quality.
For anyone evaluating TAO, the subnet layer is the place to look. The token's long-term case is only as strong as the productivity of the subnets beneath it - and in 2026, that productivity is finally becoming measurable.
To see how AIOKA's TAO council weighs liquidity and risk across this landscape, alongside six other markets, visit aioka.io.
*This article is for informational purposes only and does not constitute financial advice. Past performance does not guarantee future results. Always do your own research before making any investment decisions.*