Bittensor’s Revenue Era and the Gamma Bet Out of Montreal

A conference that sold a thesis, not a token

For most of its life, Bittensor has been argued about in the abstract: a decentralized market for machine intelligence, a token called TAO, and a hundred-odd subnets competing for a shared pot of emissions. In late September 2026 the argument moved into a room. Roughly a hundred independent teams filled New City Gas, a converted Victorian-era power plant in Montreal, for the two-day Exploit Summit on 28 and 29 September. The headline out of the event was not a price pump or a listing. It was a thesis: that Bittensor is leaving the era of farming token emissions and entering an era of selling to customers, and that the next job is to make the subnets buy from each other.

Two concrete things anchored that thesis. The first was a fresh dataset putting dollar figures on how much money the network actually pulls in from outside users. The second was a proposal, floated by co-founder Jacob Steeves, for a new on-chain credit called Gamma that would let one subnet pay another for compute, storage or inference. Neither is finished. The external revenue is still small next to what the protocol prints in rewards, and Gamma is a concept with no issuance schedule, supply cap or launch date attached. But together they reframe the question investors have asked since TAO first rallied: is there a business underneath the token, or just a very expensive incentive loop?

The timing matters. TAO changed hands near $310 at the time of writing, with a market capitalization around $3.5 billion and a rank near #34, according to CoinGecko, leaving it roughly 59% below the $757.60 all-time high set in March 2024. That is a recovery, not a breakout, and it sits against a Q4 backdrop in which a US government shutdown has slowed the regulatory machine to a crawl, a story we covered in our look at a two-person SEC racing the clock. A pending spot ETF, discussed below, is stuck in that freeze. So the Montreal pitch had to stand on its own merits rather than on a near-term catalyst, which is exactly why the revenue framing and the Gamma proposal deserve a close read.

What the Exploit Summit actually was

The event was run by Bittensor Commons, a community non-profit set up to push visibility and adoption of the network, and it was deliberately not a corporate keynote marathon. The program leaned adversarial: live demonstrations, workshops, head-to-head competitions and public debates about whether open, incentive-driven AI can be a genuine alternative to models built behind the walls of a few large companies. The organizers framed the stakes through Etienne Leroy, a director of the Opentensor Foundation.

The question is no longer whether machines can be made intelligent. It is who gets to build the systems that produce that intelligence, and on what terms.

Etienne Leroy, director, Opentensor Foundation

The speaker roster read like a who-is-who of the subnet economy rather than a lineup of outside sponsors: Steeves, now chief executive of the Affine subnet after stepping down from the foundation earlier in the year; Micaela Bazo of Metanova Labs; Will Squires and Steffen Cruz of Macrocosmos, the team behind the distributed-training work we profiled in Bittensor’s training bet; Rob Myers of Manifold Labs, which runs the Targon compute subnet; Marcus Graichen of Taostats; and Max Sebti of Score Technologies. One name stood out for a different reason: Jean-Thomas Ledore, a partner at PwC France, appeared on the agenda, and PwC France also turns up on the network’s client list. When your conference speaker is also your customer, the line between demo and sales call blurs, which is the kind of thing Montreal was built to surface rather than hide.

Bittensor’s revenue era, by the numbers

The dataset doing the heavy lifting in Montreal was the first Bittensor Revenue Index from SubConnect, published in late August 2026 and reported in detail by Crypto Briefing on 21 September. Its central claim is specific: 24 subnets generate an estimated $28 million to $35 million in annualized revenue from paying customers, money that comes from outside the network rather than from token emissions. Of those 24, the index flags 15 as high-confidence, meaning the revenue estimate is backed by public dashboards, on-chain data or company disclosures rather than a founder’s word.

That is a meaningful shift in how the network talks about itself. For two years the standard critique of Bittensor was that almost nobody paid for anything; subnets earned TAO by scoring well on benchmarks, then sold it. The index reframes maturity around customer adoption instead of leaderboard position. The clearest signal of that shift is behavioral: SubConnect reports that 14 of the 24 revenue-generating subnets now funnel external revenue into buybacks of their own alpha tokens, recycling customer dollars into their on-chain markets rather than simply dumping emissions. SubConnect’s own forward projection, cited across the trade press, is that ecosystem revenue could exceed $100 million and 35 to 40 subnets could be earning by the end of 2026. Treat that as a projection from an interested party, not a result; the measured figure is the $28 million to $35 million.

Where the money actually comes from

Dig into the index and the revenue is lopsided. Compute and infrastructure subnets, the ones renting GPUs and serving model inference, account for an estimated $23.1 million to $27.3 million, or 78% to 82% of the total. Applied AI and enterprise tools make up the rest. The top three earners by paying-customer revenue are Lium (SN51), a GPU rental marketplace, at $8 million to $10 million; Targon (SN4), a confidential-compute cloud run by Manifold Labs, at $5.5 million to $6 million; and Chutes (SN64), the serverless inference subnet from Rayon Labs we dissected in our Rayon Labs profile, at $4 million to $5 million. Those three alone contribute close to half of all verified revenue.

The named customers matter as much as the numbers, because they are the thing a benchmark score can never be: proof that somebody outside the crypto bubble wrote a check. Bittensor’s client list, as surfaced around the summit, reportedly includes PwC France, Dropbox and an unnamed NYSE-listed real estate investment trust. None of those are household crypto names, which is the point; they are ordinary enterprises buying compute or inference because it is cheaper or more private, not because they hold TAO.

The subsidy gap nobody closed yet

Here is the number that keeps the revenue era honest. Against that $28 million to $35 million of external income, the protocol still prints token emissions estimated at more than $300 million a year at recent prices. In other words, customer revenue covers somewhere between 9% and 12% of what the network pays out in rewards. The emissions still dwarf the business by roughly eight to ten times. A maturing company does not usually describe a situation where sales fund a tenth of the payroll as a new era, yet in crypto terms the ratio is a genuine improvement over a year ago, when the honest answer to who pays was close to nobody.

The emissions math is worth stating plainly because it is the denominator in every subsidy argument. After the first halving in December 2025, the protocol issues roughly 3,600 TAO per day, split 18% to the subnet owner, 41% to miners and 41% to validators and their delegators, and paid out in each subnet’s alpha token rather than in TAO itself. Around 70% of the circulating supply is staked, earning something close to 10% a year, which is the same incentive structure that makes a network like Ethereum attractive to hold rather than spend, a dynamic we unpacked in our guide to Ethereum solo staking. The difference is that Ethereum’s issuance is tiny; Bittensor’s is enormous relative to its revenue, and that is precisely the gap the Montreal roadmap is trying to close.

How real is the revenue, really?

Even the $28 million to $35 million deserves a skeptic’s read, because the estimates behind it are wide and, in places, thinly sourced. The independent analysts at Own Your Mind traced Lium’s revenue to a single figure of about $432,000 a month, published in a newsletter in late April 2026; Lium and its parent Datura had put out no dollar figure of their own, and there is no on-chain rental-settlement dashboard to check it against. Annualized, that $432,000 a month is roughly $5.2 million, below SubConnect’s $8 million to $10 million range, and it left the subnet with a subsidy ratio of about 3.5 to 4.9 times, meaning emissions handed it three to five dollars for every dollar of measured revenue.

Chutes is the subnet everyone points to when they claim Bittensor has paying customers, and it is also the one where self-reported and verifiable numbers diverge most. Rayon Labs has described throughput as high as 160 billion tokens a day; the verifiable figure from OpenRouter peaked near 42 billion tokens a day in February 2026 and settled to 8 to 12 billion by spring, implying roughly $1.1 million to $5.6 million of annual revenue rather than the headline. Pine Analytics put Chutes’ subsidy ratio at 22 to 40 times, with a break-even price around $1.41 per million tokens against the roughly $0.88 per million charged by centralized rivals. Targon’s $10.4 million figure, meanwhile, is self-reported, unaudited and unaccompanied by a live dashboard. The table below lines up the index estimates against what can actually be measured.

Bittensor against the centralized clouds

The revenue only means something measured against who the subnets actually compete with, and that is not other crypto projects. A GPU-rental or inference subnet competes with Amazon Web Services, CoreWeave and a cluster of decentralized-compute peers like Render, Akash and io.net. The pitch is familiar: cheaper, more private, harder to censor. But the Chutes break-even math complicates the cheaper half of that claim. Unsubsidized, Chutes would need to charge roughly $1.41 per million tokens to cover its costs, against the roughly $0.88 per million that centralized rivals charge. The low prices that win customers are therefore partly a product of the emissions subsidy, not a durable cost advantage baked into the architecture.

That is the uncomfortable core of the revenue era. Some of the demand exists because TAO holders are quietly footing part of the bill through dilution. If Gamma and the Emission Gate succeed in tying rewards to genuine usage, the subsidy narrows and prices drift up toward the true cost of the compute, which is the healthy outcome even if it thins the order book. If they fail, the revenue was never standalone to begin with. Either way, the honest test is simple: can any subnet sell at a price that covers its own costs without the token printing the difference? As of October 2026, very few can, which is exactly why the Montreal roadmap spends so much energy on turning emissions into something other than a subsidy.

The full-stack intelligence network

Steeves’ keynote tied the revenue data to a bigger structural idea, one Crypto Briefing summarized as a full-stack intelligence network. The vision maps the ecosystem into six specialized layers: data, compute, storage, pre-training, inference and AI agent functionality. The point is not that each layer exists in isolation, which is already roughly true, but that the layers transact with one another. An agent subnet would buy inference from an inference subnet, which would rent GPUs from a compute subnet, which would pull training data from a data subnet, each paying the one below it. In Steeves’ framing the subnets do not merely coexist; they do business.

If that sounds like a cloud-computing supply chain, that is the intention. The difference is that today’s subnets mostly do not buy from each other at all; they compete for the same shared emissions and sell, if they sell, to the outside world. The AI-agent layer is the most speculative and the most interesting, because autonomous agents that hold value and pay for services are exactly the primitive the rest of crypto has been circling, with all the custody and safety questions we raised in our piece on trusting an agent with a wallet. A network where agents natively pay inference bills is a cleaner version of that thesis than most, because the payment rail and the compute are part of the same system.

Enter Gamma: turning rivals into customers

The missing piece in that supply chain is a unit of account, and that is what Gamma is meant to be. As Crypto Briefing described the proposal, Bittensor’s subnets have spent years competing for the same pot of TAO emissions, and Gamma wants to flip that dynamic by turning rivals into customers. Mechanically, a subnet would convert a portion of its TAO emissions into Gamma credits, and those credits could then be spent on compute, inference, storage or other services provided by fellow subnets or by outside providers. An inference subnet that needs storage would pay a storage subnet in Gamma rather than going to the open market or building its own.

The conceptual shift underneath the mechanics is the part worth sitting with. Today, TAO and alpha emissions function as rewards that recipients mostly liquidate, which is the source of the chronic sell pressure that has weighed on the token. Gamma would reframe those emissions as an operating budget: capital that gets reinvested inside the network instead of sold. The stated goal is not to end competition but to layer a collaboration mechanism on top of it, so the whole becomes more capable than the sum of its parts. An applied subnet settling, say, prediction-market outcomes, a category going mainstream as we described in our look at institutional prediction markets, could buy the inference it needs from a specialist rather than reinventing it, and pay in a credit that keeps the money inside Bittensor.

What Gamma would do, and what is still blank

For all the attention it drew, Gamma is a sketch, not a specification. The proposal sits at the conceptual stage, and the open questions are the important ones: the exact token mechanics, how the credit system would be governed, and how Gamma would interact with the alpha-token markets that already define each subnet’s economy. One floated idea is that subnets could burn alpha tokens to acquire Gamma credits, which would tie the new layer directly into existing markets, but Crypto Briefing was explicit that those mechanics remain to be formalized.

The conference itself underscored how early this is. The Gamma session on the agenda, an OpenDev deep dive led by Steeves, ran a brisk 20 minutes, and as TokenPost noted, the agenda offered no details on token issuance, supply, redemption mechanics or a launch date. That is not unusual for Bittensor. Dynamic TAO, the system that gave every subnet its own alpha token and let the market set emission shares, took well over a year to move from concept to live network after it was first described, finally launching in February 2025. Gamma is a further layer of complexity on top of that architecture, which makes any timeline genuinely uncertain. Anyone treating it as a near-term catalyst is front-running a whitepaper that does not exist yet.

The case for and against a closed-loop economy

The bull case for Gamma is coherent. If subnets spend credits with one another instead of dumping emissions, money circulates inside the network, sell pressure falls, and emissions start to look like a reinvested operating budget rather than a faucet. That logic already drove an earlier 2026 upgrade, Root Reborn, which replaced a permanent block-by-block alpha sell stream with per-validator escrow realized only when a staker claims. Gamma extends the same instinct from validators to subnets. Vertical integration is the prize: a stack where a customer can buy data, training, inference and agents from one coordinated network, priced in one internal unit, is a far stronger pitch than a leaderboard of isolated models.

The bear case has two prongs, and both are serious. The first is complexity fatigue. Gamma would be the fourth or fifth major economic rewrite in roughly 18 months, after dTAO, the Emission Gate, Root Reborn and Conviction voting, and not everyone underwriting the token is thrilled about the pace. Mark Creaser, chief executive of DSV Fund, told tao.media in August that “a casino at least keeps the same rules all night; Bittensor changes the table mid-hand and calls it an upgrade,” adding that “nobody can tell you what the rules will be next Tuesday.” His partner Siam Kidd went further, calling dTAO “basically uninvestable” at that cadence of change. The second prong is subtler and, for the revenue thesis, more dangerous: a closed loop can manufacture the appearance of demand. If subnets pay each other with credits minted from emissions, on-chain activity rises without a single outside dollar entering the system. Distinguishing genuine external revenue from internal churn is already hard, as the verification gaps above show; Gamma could make it harder.

The decentralization clock still reads December 2027

Every forward-looking feature in Montreal, Gamma included, is shadowed by an unresolved governance question. Both founders stepped back from executive roles in 2026, Steeves from the Opentensor Foundation in February and co-founder Ala Shaabana the following month, though both stay on as contributors. The published roadmap targets a handover to a democratized governance model around December 2027, built on tougher validator competition, bidirectional liquidity pools, conviction-based voting and revised incentive-distribution logic. Steeves has acknowledged that the network is not yet decentralized in the way Bitcoin is, framing the centralized control as a temporary tool to keep pace with AI rather than a permanent arrangement.

Not everyone accepts that framing. The sharpest dissent came from Covenant AI, the team behind the Templar training subnet, which exited the ecosystem on 9 April 2026. Its founder, Sam Dare, did not mince words.

The entire premise of Bittensor, the promise that drew builders, miners, validators, and investors into this ecosystem, is that no single entity controls it. That promise is a lie. It is not. It is decentralization theatre.

Sam Dare, founder of Covenant AI, speaking to The Block

Dare alleged that Steeves had suspended emissions to Covenant’s subnets and depreciated its infrastructure unilaterally, the kind of discretionary control the Dec 2027 roadmap is meant to dissolve. TAO fell about 15% on the news, from $338 to a low near $285 within two hours before recovering. The relevance to Gamma is direct: a credit system that routes payments between subnets is only as trustworthy as whoever governs the credit, and today that is still a small group. Ask who can freeze a subnet’s Gamma balance and you are back to the same question Dare raised.

The ETF overhang and a frozen SEC

The other thing hanging over the token is the exchange-traded fund that has not arrived. Grayscale filed an S-1 on 30 December 2025 to convert its existing Bittensor Trust, which trades over the counter under the ticker GTAO, into a spot ETF listed on NYSE Arca. The trust charges a 2.5% expense ratio, holds a fixed amount of TAO per share, and creates and redeems in 10,000-share baskets through authorized participants using in-kind TAO or cash. It has continued to file routine updates with the regulator, the most recent an 8-K dated 23 September 2026, but there is still no approval and no listing.

Part of the delay is structural: the SEC is operating through a government shutdown that has effectively frozen new approvals, the environment we described in our report on the October freeze at a dark SEC. Part of it is specific to Bittensor. A US spot ETF almost certainly cannot stake the TAO it holds, which means an ETF holder would forgo the roughly 10% staking yield, the alpha-token emissions and any governance rights, while still eating the full dilution from the protocol’s issuance. That is the staking paradox: the regulated wrapper US investors can buy is the one that strips out the thing that makes holding TAO economically rational in the first place. European staked-TAO products that pass yield into their net asset value do not have that problem, which is an awkward look for the market everyone most wants to reach.

What to watch next

Strip away the conference staging and Bittensor’s situation in October 2026 is easy to state. The network has, for the first time, a credible dataset showing real customers paying real money, concentrated in compute and inference and led by Lium, Targon and Chutes. It also has a vision, the full-stack intelligence network, and a mechanism, Gamma, meant to turn that revenue into a self-sustaining internal economy. And it has a gap between the two that remains an order of magnitude wide, plus a governance question that will not settle until late 2027.

Four things will tell you whether the Montreal thesis is working. First, a Gamma specification with actual numbers: issuance, supply, redemption and, above all, who governs the credit. Second, the next SubConnect index, and whether external revenue climbs toward that $100 million projection while the subsidy ratio narrows instead of widening. Third, the ETF, which could move quickly once the SEC reopens and clears its backlog. Fourth, visible progress on the December 2027 handover, because every promise about collaboration and credits rests on the assumption that no single party can rewrite the rules mid-hand. At roughly $310, TAO is already pricing in a good deal of the vision. The job now is to ship the parts that are still slideware, and to do it without changing the table one more time.

Frequently Asked Questions

What is the Gamma token on Bittensor?

Gamma is a proposed on-chain credit, unveiled at the Exploit Summit in September 2026, that would let Bittensor subnets pay one another for services like compute, inference and storage. Subnets would convert part of their TAO emissions into Gamma credits and spend them inside the network. As of October 2026 it is only a concept, with no token issuance, supply cap, redemption mechanics or launch date disclosed.

How much revenue do Bittensor subnets actually make?

SubConnect’s first Bittensor Revenue Index, published in August 2026, estimates that 24 subnets generate $28 million to $35 million a year from paying customers, led by the compute subnets Lium, Targon and Chutes. That external revenue still covers only about 9% to 12% of the more than $300 million a year the protocol pays out in token emissions, and independent trackers measure some subnets well below the index estimates.

What was announced at the Bittensor Exploit Summit 2026?

The Exploit Summit, held in Montreal on 28 and 29 September 2026 and organized by the non-profit Bittensor Commons, centered on a full-stack intelligence network roadmap from co-founder Jacob Steeves, the Gamma credit proposal for inter-subnet payments, and a revenue dataset showing 24 subnets earning real money from customers. It was a builder conference of more than 100 teams, not a product launch.

Is there a Bittensor (TAO) ETF?

Not yet. Grayscale filed in December 2025 to convert its over-the-counter Bittensor Trust (ticker GTAO) into a spot ETF on NYSE Arca, and it continues to file updates with the SEC, but no US spot Bittensor ETF had been approved as of October 2026. A government shutdown that has frozen SEC approvals is part of the delay.

When will Bittensor be fully decentralized?

The Opentensor Foundation’s published roadmap targets a handover to a democratized governance model around December 2027, after both founders stepped back from executive roles in 2026. Critics, including the departed Covenant AI team, argue the network remains effectively controlled by a small group today, and Steeves has acknowledged it is not yet decentralized in the way Bitcoin is.

Marcus Okafor covers decentralized AI and market structure for the HOGE Wire markets desk.