Chutes: Bittensor's Revenue Machine, Subsidised

The thesis

Chutes is the strongest revenue story Bittensor has. It’s also losing money on every customer who uses the product, and the team has said so themselves.

The first part is what you read in the headlines. Chutes was the first Bittensor subnet to cross $100 million in market cap. It runs serverless inferenceInferenceRunning a trained AI model to produce an answer. Inference is what happens when you type a prompt into ChatGPT and get a response. The model takes your input, computes a best guess, and returns it.Like asking an expert for their opinion. The training was the decades they spent becoming an expert. The inference is the 30 seconds it takes them to answer your specific question.Read more → for 50+ open-source models, ships an APIAPIApplication Programming Interface. A structured way for one piece of software to talk to another. In DeAI, APIs let applications request inference from a model without running the model themselves.Like a waiter in a restaurant. You don't walk into the kitchen and cook your own meal. You tell the waiter what you want, they tell the kitchen, the kitchen cooks it, and the waiter brings it back. The API is the waiter.Read more → that competes head-to-head with Together.ai and Replicate, and gets routed to by default on OpenRouter. Real product, real users, real revenue.

The second part is what doesn’t make the headlines. In February 2026, the team published a community announcement that quietly killed the free tier and explicitly named the problem: heavy users were extracting “56x to 324x” their subscription value at equivalent pay-as-you-go pricing. Translation: paid users were getting more compute than their fees covered, by a factor of 50 to 300, and the gap was being filled by TAO emissionsEmissionsNew tokens created and distributed by a blockchain protocol over time as rewards to validators, stakers, or miners. Emissions fund network security and participation at the cost of diluting existing holders.Like a company that pays employees partly in newly printed shares. Every year the total number of shares goes up, which means existing shareholders own a slightly smaller slice of the same company unless the company grows faster than the printing.Read more →.

This article looks at both sides. Chutes is the most important Bittensor subnet to understand because it’s the test case for whether the entire subnet modelModelA trained neural network that takes inputs (text, images, audio) and produces outputs (more text, classifications, generated content). In DeAI the model is the thing that actually does the work.Like a very experienced apprentice who has spent years watching thousands of masters make furniture. They can't explain how they know when a joint is right, but they can make a chair that looks and functions like a Chippendale. The training is invisible. The output is what matters.Read more → can produce a sustainable business or whether it produces subsidised cloud services that collapse when the subsidy stops. December 2026 is when we find out.

What it actually is

Chutes is a serverless inference platform. You hit an API endpoint, you get back a model response, you get billed per tokenTokenA digital unit of value or access rights tracked on a blockchain. Tokens can represent ownership in a project, a right to use a service, a share of future revenue, or simply a tradable asset with no underlying claim.Like a physical poker chip a casino issues. The chip itself has no value. What makes it worth something is what it lets you do at the casino, what the casino has promised, and how much other people will pay you for it.Read more →. The service runs on top of Bittensor subnet 64, which means the underlying compute comes from miners running Chutes’ validator-evaluated workloads in exchange for TAO emissions.

From a developer’s perspective it looks like any other inference API. From the network’s perspective it’s a market: miners compete on quality and uptime, validators score them, the highest performers earn the TAO emissions allocated to subnet 64. Roughly 14.39% of all TAO daily emissions flow to this subnet, which works out to approximately 518 TAO per day. At a TAO price of $320, that’s roughly $166K per day or $60M annualised in subsidy.

The model catalogue is broad. Chutes’ public model list includes Llama 3.2 and 3.3, DeepSeek V3.1, Qwen3 (4B through 235B), Mistral Small and Nemo, Kimi K2.5, GLM-5, Gemma 3, plus diffusion models, embeddings, speech, and 3D generation. The pricing is genuinely competitive with centralised alternatives:

These numbers are pulled live from Chutes’ API and Together.ai’s pricing page. The discount is real. The question is what funds it.

The OpenRouter contradiction

Chutes claims 160 billion tokens per day across all surfaces. Rayon Labs has been citing this number consistently since March. It’s also the foundation for the “approaching $10M ARR” claim that started circulating in early April.

The verifiable number tells a different story. Chutes is a provider on OpenRouter, the largest open inference router. OpenRouter publishes daily token throughput data per provider, which is one of the few ways to externally verify a Bittensor subnet’s actual usage. Chutes’ OpenRouter throughput peaked on 7 February 2026 at roughly 42 billion tokens per day. By late March it had declined to 8-12 billion per day. The most recent data point shows approximately 6.8 billion tokens per day across the 18 models Chutes serves on the platform.

Three possibilities, in order of charity:

1. OpenRouter is a small fraction of Chutes’ total volume. Direct API customers and the chutes.ai web interface dwarf the OpenRouter traffic. If that’s true, the team should publish a dashboard.
2. The peak was real, the decline is real, the 160B figure is stale. Chutes had a moment in February when one model (TNG/DeepSeek R1T2 Chimera) was getting heavy default routing on OpenRouter. That model has rolled off. The current 6.8B figure is the new baseline.
3. The 160B number was never accurate. It was an aspirational high-water mark that became a marketing line.

I’m not in a position to say which of these is right. Neither is anyone outside Rayon Labs. What I can say is that the “explosive growth” narrative from March doesn’t match the publicly verifiable February-to-April trajectory, and any article that cites the 160B number without flagging this gap is misleading the reader.

The subsidy economics

This is the part that matters more than any single throughput number.

Pine Analytics published a bear case on Bittensor on 24 March 2026. Their estimate of Chutes’ actual external customer revenue is $1.3-2.4 million annualised. Against that, the subnet receives roughly 518 TAO per day in emissions, which at current TAO prices is approximately $60 million annualised. The subsidy ratio is 22 to 40 times. Customer payments fund a small fraction of operations. TAO inflationInflationThe annual rate at which new tokens are created and added to the circulating supply. Most networks use inflation to pay validators, stakers, and infrastructure providers from freshly minted tokens rather than real revenue.Like a landlord who raises the rent every year. If your salary goes up at the same rate, you break even. If it doesn't, you get poorer without noticing, because the number on your payslip hasn't changed but the ground under it has shifted.Read more → funds the rest.

The pricing math follows directly. Pine calculates that if Chutes had to cover its costs from customer revenue alone (without emission subsidy), the break-even price would be approximately $1.41 per million tokens. Together.ai charges $0.88 per million tokens for Llama 70B. Unsubsidised Chutes would be 1.6 times more expensive than the centralised alternative, not cheaper. The cost advantage that makes Chutes attractive depends entirely on the emission subsidy continuing at its current rate.

The team confirmed this directly in February. The community announcement eliminated the free tier (200 requests per day for roughly a year) and pulled four frontier models (GLM-5, Kimi K2.5, Qwen 3.5, MiniMax M2.5) out of the Base subscription tier. The stated reason: “heavy users were extracting 56x to 324x their subscription value at equivalent pay-as-you-go costs.” That’s not Pine Analytics talking. That’s Chutes’ own blog post.

Most projects don’t admit this. Chutes did. Credit where it’s due. The honesty doesn’t fix the economics, but it does mean the team is aware of the problem and trying to address it.

The December 2026 deadline

Bittensor’s emission schedule halves on a fixed cadence. The first halvingHalvingA protocol event that cuts the rate of new token emissions by half. Halvings are scheduled in advance, happen automatically at fixed intervals, and are a core mechanism for enforcing declining token supply growth over time.Like a savings account where the interest rate is contractually cut in half every four years. You still earn interest, but the rate drops on a known schedule, and the issuer can't change it without breaking the contract.Read more → happened on 14 December 2025, dropping daily emissions from 7,200 to 3,600 TAO. The next halving happens approximately 14 December 2026, dropping daily emissions from 3,600 to 1,800 TAO.

What that means for Chutes specifically:

The halving cuts Chutes’ subsidy in half. That’s the easy part. The hard part is that for Chutes to reach actual sustainability (revenue equal to or greater than emission value), customer revenue needs to grow from roughly $2 million to roughly $30 million in eight months. That’s a 15x increase from the verified base. Even from the upper-bound self-reported $10M ARR claim, it’s still a 3x increase.

It can be done. Together.ai grew faster than that during their early traction phase. The question is whether decentralised serverless inference has a structural reason to be cheaper than centralised serverless inference at scale, or whether the subsidy was the only thing creating the price difference. If it’s the latter, Chutes’ competitive position in December 2026 is materially worse than its position today.

The team and the entity question

Chutes was founded by Jon Durbin. Before Bittensor he was known as the author of Airoboros, a self-instruct fine-tuningFine-tuningThe process of taking a pre-trained model and training it further on a smaller, more specialised dataset to adapt it to a specific task, domain, or style. Fine-tuning is much cheaper than training from scratch.Like hiring an experienced general practitioner doctor and giving them six months of focused training in a sub-speciality. You don't have to teach them medicine from scratch. You just narrow their expertise to the area you actually need.Read more → framework that’s been widely used in the open-source LLMLLMLarge Language Model. A neural network trained on vast amounts of text to predict the next word in a sequence. Modern LLMs (GPT, Claude, Llama, Qwen, DeepSeek) generate human-quality text and are the foundation of most modern AI products.Like an autocomplete that read every book ever written. It has no memory of individual texts but it has absorbed the patterns of language so deeply that it can generate paragraphs that sound human. The skill is statistical, not conscious.Read more → trainingTrainingThe one-time process of teaching a neural network to perform a task by showing it massive amounts of example data and adjusting its internal weights until the outputs are good. Training builds the model; inference uses it.Like the years an apprentice spends learning a trade. You don't see any of the actual work, just thousands of repeated mistakes gradually becoming competence. By the end, the apprentice can do the job. The training was invisible, but the skill is now permanent.Read more → scene. Real technical credentials, established public reputation, no anonymous founder problem at the individual level.

Rayon Labs is a different question. Rayon Labs is the umbrella name for the team that runs Chutes (SN64), Gradients (SN56), and Nineteen (SN19). Combined, those three subnets capture roughly 23.7% of all TAO daily emissions. Nearly a quarter of Bittensor’s output flows to a single team.

I went looking for the corporate entity. There isn’t one I could find. No Delaware C-corp registered to anyone named Jon Durbin or under “Rayon Labs.” No Cayman entity. No UK Companies House filing. No Crunchbase or PitchBook profile. No disclosed funding round. The only published company description is “independent teams collaborating,” which isn’t really an entity. The only public GitHub org has one repo with two stars, while the actual Chutes platform code lives at github.com/chutesai/chutes, a separate org. The published Chutes repository contains the client SDKSDKSoftware Development Kit. A collection of code libraries, documentation, and tools that lets developers integrate a service into their applications without writing everything from scratch. SDKs are how projects become easy to build with.Like a plug-and-play kit for building furniture. You don't have to mill your own wood, forge your own screws, or design the joinery from scratch. The kit gives you pre-cut parts and instructions so you can assemble the thing in an afternoon.Read more → only. The miner, validatorValidatorA computer that runs the full blockchain protocol, verifies transactions, and proposes new blocks. Validators are the workers that keep a Proof of Stake network running, and they earn rewards for doing the work correctly.Like a notary public who witnesses and stamps legal documents. Validators witness transactions, check they follow the rules, and stamp them into the permanent record. A notary who commits fraud loses their license. Validators work the same way, except the license is staked tokens that get slashed on misbehaviour.Read more →, and platform code are private.

This isn’t necessarily a problem in isolation. Plenty of legitimate crypto projects operate as informal teams without traditional corporate structure. But the combination matters: 23.7% emission concentration, no disclosed legal entity, pseudonymous co-founders (Namoray is publicly known only by handle), and a closed-source core inference platform. For sovereignty-first readers, those are four risks compounding.

The bull case is that Jon Durbin’s track record speaks for itself, the Airoboros work is well-regarded in the open-source MLMLMachine Learning. The branch of AI where systems learn patterns from data instead of being explicitly programmed with rules. Modern AI (LLMs, image generation, recommendation systems) is almost entirely machine learning.Like teaching a child to recognise dogs by showing them thousands of pictures of dogs, instead of writing down a precise rulebook for what makes a dog. The child learns the pattern from examples rather than from instructions.Read more → community, and the team has consistently shipped product. The bear case is that 23.7% of Bittensor’s economic output flows to a team with no published entity, no disclosed governance, and no formal accountability mechanism. Both are true at the same time.

What I’d want to see

Five things would change my read on Chutes materially:

1. A public revenue dashboard. Anything that gets external customer revenue from “trust the team’s number” to “verifiable on a URL” closes the credibility gap with Pine Analytics’ independent estimate.
2. The OpenRouter trajectory reversing. Sustained throughput growth on the one platform we can verify would mean the decline narrative is wrong.
3. Open-sourcing more of the platform. The client SDK being MIT is fine. The miner and validator code being closed is fine. The inference orchestration layer staying closed forever is not consistent with “decentralised inference” framing.
4. A formal Rayon Labs entity. Even a simple disclosure of the legal structure, jurisdiction, and accountability framework. The current state is closer to “trust us” than “verify us.”
5. Customer concentration disclosure. If 80% of Chutes revenue comes from one OpenRouter integration or one large customer, that’s a different risk profile than 80% coming from thousands of individual developers.

None of these are unreasonable asks. Together.ai publishes most of them. So does OpenRouter. So does almost every centralised inference provider Chutes competes with. The fact that none of them are available is itself information.

The honest assessment

Chutes is the most interesting thing happening on Bittensor right now. It’s the only subnet with a product that genuine developers use, real revenue from external customers, and a team that’s been honest about the problems.

It’s also losing money on every customer at current prices, the volume narrative doesn’t match the publicly verifiable data, and it sits on top of a $60 million annualised TAO subsidy that gets cut in half in December. The team will need to grow customer revenue roughly 15x from the verified base to reach sustainability before the next halving compresses the subsidy further.

I think Chutes will survive the December 2026 halving. The team is competent and they’ve shown they understand the problem. What I’m less sure about is whether they survive the third halving in December 2027 if revenue growth doesn’t materially accelerate. The subsidy compounds the wrong way for them: each halving makes the math harder, not easier, and there’s no version of decentralised serverless inference where the structural cost is meaningfully below centralised serverless inference. The advantages of decentralisation in this category are sovereignty and censorship-resistance, not unit cost.

For Bittensor holders, Chutes is the bet that one subnet can build a sustainable business inside the network. If it works, the entire subnet model gets validated. If it doesn’t, the network’s strongest revenue story turns out to have been a subsidy story all along. December 2026 is when we know.

For the broader Bittensor revenue picture, see Bittensor subnets: where the revenue actually is. For how to actually buy and stakeStakingLocking up a cryptocurrency to help secure a blockchain network, usually in exchange for rewards. The locked tokens act as a security deposit that can be taken away if the staker misbehaves.Like putting down a large rental deposit for an apartment. You get the money back if you behave, you earn interest while it's locked, and the landlord takes it if you trash the place.Read more → into subnet 64, see our Bittensor subnet staking guide. For the underlying token economics, see dTAO subnet economics.