Bittensor’s AI Compute Subnets Collectively Reach $20M ARR
Subnets that actually generate revenue are pulling in the lion’s share of the Bittensor network’s emissions. Today, three of the top 5 subnets—Chutes, Targon, Lium—collectively account for nearly 20% of the network's entire emission allocation, worth just over $2.25M/day. Having only switched on monetization about three months ago, they are already running at a combined $20M annualized revenue run-rate.
The common theme connecting these subnets is that they are all oriented around compute. Specifically, aggregating GPUs to provide AI-focused compute.
The dTAO marketplace hasn’t agreed on the right metrics to value these GPU and inference subnets relative to one another yet. As the table above shows, Targon, Chutes, and Lium are each priced differently based on their GPU, and even ARR, multiples to current Adjusted Market Caps. Each subnet is pioneering its own path with different target customers and strategies for scaling their compute capabilities. Although it’s apparent compute subnets deserve a significant portion of Bittensor’s emissions, the market’s lack of consensus on the token’s valuation means we can expect a reshuffling as each team’s revenue roadmap comes into clearer focus.
Chutes and Targon
Chutes and Targon cater to app developers. They make it dead simple to integrate AI (e.g., LLMs or image models) capabilities into an app. Chutes leverages its GPU stockpile to serve more than 120B tokens per day through OpenRouter and direct customer deals, driving $4.3M in ARR.
Targon carved out a different path, initially incentivizing H200 GPU suppliers for their trusted execution environment (TEE) enabled hardware. The breakthrough was the Targon Virtual Machine (TVM), a software runtime that seamlessly leverages NVIDIA’s nvTrust SDK to guarantee that compute jobs are executed honestly and confidentially, even on untrusted infrastructure.
This gave Targon an advantage that let them land enterprise customers requiring confidential compute. Public data shows Targon processed 639B tokens in the month of July. And according to the team, as of September, the subnet is monetizing at a $10.4M ARR.
Chutes is now developing its own confidential compute capabilities, which should strengthen its ability to win enterprise accounts and expand revenue alongside its existing retail developer base.
Both subnets have quickly become the inference platforms of choice for product-focused teams on Bittensor like Ridges, Dippy, and Score. Eventually, these compute platforms will likely expand and begin to offer a broader set of services that companies like Together.ai are offering: fine-tuning, GPU cluster rentals, and dedicated endpoints.
Lium
Lium gives AI engineers and researchers low-level access to GPUs, where they can rent machines by the hour and use them for everything from running small-scale training experiments to fine-tuning open-source models.
Lium’s supply is concentrated around high-end GPUs more suitable for model training. Nearly 85% of Lium’s revenue comes from renters of B200s, H200s, and A100s. Lium has rapidly accumulated one of the deepest pools of the B200 GPUs, the strongest Nvidia GPU. It now hosts over 4x as many B200s as Vast.ai and sustains more than 50% higher rental rates.
Since July this year, Lium has already scaled to a $5.3M annualized revenue run-rate. Its natural customer base started out as miners on compute-hungry, model-development subnets like Score, Templar, and IOTA, which require the most powerful hardware on the network.
Secret Sauce
Chutes and Targon offer some of the lowest dollar per token prices for open-source models, and, for certain GPUs (e.g., H100s), Lium has the lowest hourly rates across all GPU rental markets. The key to achieving this is by tapping into a specific set of miners: datacenters and compute owners.
These entities often have their GPUs sitting idle while they look for customers to sign long-term, bulk GPU rental agreements. Idle GPUs earn no revenue, and GPUs are depreciating faster than ever, meaning these compute owners are eager to generate revenue against these assets by any means possible.
Compute-based subnets are a natural fit for compute owners; they can plug their GPUs into these networks without signing any contracts, earn a continuous stream of rewards, and pull back their GPUs at any time. This gives them the flexibility to continue operating their core business (i.e., longer-term rental agreements) while earning revenue if their GPUs aren’t actively being leveraged.
Everyone wants an AI Company
Chutes, Targon, and Lium are all AI companies. They transform one of the most in demand, baseline commodities (i.e., GPU compute) into venture-backable businesses by layering on value-additive software and services.
The core scaling constraint they all face is the size of their reward budget to compute suppliers. Generating revenue and implementing some form of token buyback program are, to date, how these teams are resolving this scaling constraint. Driving revenues back into the subnet (via buybacks) puts upward price pressure on the token, helping sustain or grow the value of the compute supplier reward pool, and eventually (with enough revenues) making the token supply deflationary. Buybacks accrue value back to the token while also serving as a means of reinvesting in the core business.
The buyback programs associated with these tokens link them to the fast-growing businesses building on top of the subnets, not the underlying GPU commodity. And as the world makes clear that the AI compute train isn’t slowing down, everyone wants a stake in an AI company.
This content is provided for informational purposes only and does not constitute investment advice or a recommendation to buy or sell any security. Unsupervised Capital holds a position in TAO and may hold positions in the subnet tokens or other digital assets discussed herein and may buy, sell, or change positions at any time. Past performance is not indicative of future results. Digital assets involve substantial risk, including potential total loss of capital. Consult your own advisers regarding any investment decisions.
