Bittensor Explained: TAO, Subnets and a Governance Crisis

Bittensor spent the first half of 2026 proving two things about itself at once: that a blockchain network can generate real, measurable revenue from software people actually pay to use, and that its governance can look shakier than almost any other project of its size. TAO, the network’s native token, trades around $195 as of mid-July 2026, according to CoinGecko, well off its March 2024 all-time high of $757.60, yet the network underneath that price chart looks almost nothing like it did two years ago.

What started in 2021 as a single blockchain rewarding machine learning models has grown into a sprawling economy of well over a hundred specialized subnets, a market-priced token for each one, and a growing list of AI applications generating real usage revenue rather than pure speculation. It has also become a case study in how hard genuine decentralization is to deliver once real money and strong-willed founders are involved. Bittensor is also increasingly discussed alongside the wider push toward autonomous AI agents transacting on-chain, since several of its subnets exist specifically to serve that kind of workload. This is a comprehensive look at how Bittensor actually works, what changed in 2026, and where the risks sit.

What Is Bittensor?

Bittensor is an open, blockchain-based network that pays contributors in TAO for producing useful machine learning work, rather than for mining blocks or securing a ledger in the traditional sense. The idea traces back to 2019, when Jacob Steeves, known across the community by his handle Const, and Ala Shaabana began building what they described as a peer-to-peer market for machine intelligence. Steeves had worked as a machine learning engineer, including a stint at Google, before committing to the project full time.

Mainnet launched in January 2021 under the Opentensor Foundation, the nonprofit that has stewarded protocol development ever since. Bittensor’s underlying chain, known as Subtensor, has been rebuilt more than once since then to fix performance and consensus issues; the current version, called Finney, has been live since March 2023. Steeves recently stepped back from his formal leadership role at the Opentensor Foundation, though he has said he intends to keep writing chain code and building subnets, a move framed internally as part of the network’s broader push toward decentralization.

Rather than one blockchain doing one job, Bittensor is structured as a marketplace of markets. Anyone can launch a subnet dedicated to a specific AI task, from language model inference to protein folding to financial forecasting, and compete for a share of TAO’s daily token issuance based on how useful that subnet’s output turns out to be. That structure, and the fact that subnets are individually investable through their own tokens, is what separates Bittensor from a simple compute rental marketplace.

The pitch underneath all of this is that centralized AI labs concentrate an enormous amount of capital, data and decision-making power in a handful of companies, and Bittensor is meant to be a permissionless counterweight: nobody needs anyone’s permission to launch a subnet, run a miner, or become a validator, and no single company can unilaterally shut the network down the way it could pull a plug on its own servers. Whether the resulting subnets actually produce AI systems that compete with what OpenAI, Google or Anthropic ship is a separate question from whether the incentive structure itself works, and the two get conflated constantly in how the project is marketed.

Inside the Machine: Subnets, Miners and Validators

Every Bittensor subnet has the same basic cast of participants. Miners run the actual computation, whether that means serving a language model, scraping a data set, or generating trading signals, and submit their output on-chain. Validators query those miners, judge the quality of what comes back, and assign scores. The Opentensor Foundation and independent subnet teams write the specific scoring logic, called an incentive mechanism, that defines what good work looks like for that particular subnet. The chain’s code, including the dTAO and subnet-registration logic described below, is maintained publicly on GitHub, where anyone can review or propose changes to how the system works.

Subnets vary enormously in what they actually reward. Some focus purely on inference, answering user queries as fast and cheaply as possible; others focus on training, contributing compute toward a shared model; still others reward data collection, storage, or increasingly, orchestrating other AI agents. The common thread is that a validator, not a customer, decides whether the work was good enough to pay for, which pushes the quality-control problem onto the protocol itself rather than leaving it to whoever happens to be buying.

This is a meaningfully different design from a rental marketplace like a GPU cloud, where a customer pays a fixed price for raw compute and the platform does not grade the output. Bittensor instead tries to price the quality of what compute produces, which is also why it tends to get compared to other AI-focused crypto networks rather than to storage or bandwidth infrastructure projects.

Subnets register on the root chain with a unique identifier and compete against each other for emissions, the daily allotment of new TAO the protocol mints and distributes. That competition is refereed by two linked mechanisms: Yuma Consensus, which turns validator scores into actual token payouts inside a subnet, and, since early 2025, a staking market called dynamic TAO that lets anyone, not just validators, influence which subnets get funded relative to the others.

Yuma Consensus: How Bittensor Scores Machine Intelligence

Yuma Consensus is the algorithm that decides who gets paid inside a subnet once validators have submitted their scores. Each validator ranks the miners it evaluates and submits those rankings as a weight vector. Yuma Consensus then combines every validator’s weights into a stake-weighted median, discards outlier scores that look like an attempt to game the system, and pays miners in proportion to the resulting consensus score.

Validators are paid too, but only if their scoring lines up with the rest of the validator set. The mechanism uses exponentially smoothed bonds that reward validators for consistently agreeing with consensus over time and penalize those who deviate, which is meant to discourage a validator from grading its own miners generously or colluding with a small group to redirect emissions. The design is intended to hold up as long as no single actor controls more than half of a subnet’s validator stake, the same collusion threshold that shows up across most stake-weighted consensus systems.

In practice, a subnet’s quality bar is only as good as its validator set. A subnet with few validators, or validators that are economically tied to the miners they are scoring, can drift toward rewarding mediocre or even fabricated work, which is one reason serious builders watch validator concentration on a subnet almost as closely as they watch its token price.

Yuma Consensus is also why Bittensor markets itself as resistant to a single validator or miner quietly taking over a subnet. Because rewards depend on agreement across the validator set rather than one grader’s opinion, an attacker would need to control a majority of stake within that specific subnet, not just run a lot of miners, to consistently bend the results in their favor. That does not make manipulation impossible, particularly on small subnets with only a handful of validators, but it raises the cost of trying well above what a single dishonest participant could pull off alone.

Dynamic TAO: Markets Decide Which Subnets Win

For its first four years, Bittensor allocated emissions between subnets through a vote: a small set of root validators decided how much of the daily TAO issuance each subnet deserved. That changed in February 2025 with the launch of dynamic TAO, or dTAO, which replaced the vote with a market, as CoinGecko’s explainer on the system lays out.

Under dTAO, every subnet has its own token, generally called an Alpha token, paired against TAO in an automated market maker that works the same way a Uniswap pool does, using a constant product formula where the size of one reserve multiplied by the other stays fixed. Staking TAO into a subnet is technically a swap: TAO goes into that subnet’s liquidity pool and the staker receives Alpha tokens in return. The more TAO flows into a subnet’s pool relative to its rivals, the larger a share of new emissions that subnet earns, and the more valuable its Alpha token becomes. Pull TAO back out, and both the emission share and the Alpha price fall together.

The system has been tuned twice since launch. Starting in November 2025, the protocol shifted toward a flow-based model the community calls Taoflow, weighting emissions by net staking inflows rather than a simple price ratio. Then on May 13, 2026, an update called the Emissions Refactor concentrated rewards among roughly the 30 most productive subnets and tied future issuance more tightly to prior emissions and buyback activity, a deliberate move away from spreading rewards evenly across every registered subnet regardless of output.

By March 2026, the combined market capitalization of every subnet’s Alpha token had reached roughly $1.12 billion, equal to about a quarter of TAO’s own market cap at the time, a figure that continued climbing past $1.5 billion within the following few months as more subnets launched tokens and attracted staking capital. It is worth contrasting this economic approach to quality control with how verifiability gets handled elsewhere in decentralized AI; projects working on zkML try to mathematically prove a model ran correctly using cryptography, while Bittensor instead relies on staked capital and validator reputation to police quality after the fact, an approach that is cheaper to run but leans more heavily on economic incentives holding up under pressure.

TAO Tokenomics: A 21 Million Supply Cap and Its First Halving

TAO follows a supply schedule modeled directly on Bitcoin’s, an unusual choice for a network built around AI rather than payments. Maximum supply is fixed at 21 million TAO, issuance halves on a roughly four-year cycle, and the halving triggers automatically once half of the remaining supply has been mined rather than through a governance vote. The first halving activated on December 12, 2025, cutting daily issuance from about 7,200 TAO, or 1 TAO per block, to about 3,600 TAO, or 0.5 TAO per block. The mechanism is not purely symbolic: because miners and validators are paid in newly issued TAO, halving the daily issuance directly halves the raw token flow available to reward network participants, forcing subnets to compete harder for a shrinking pool unless offsetting demand for TAO shows up elsewhere, whether from stakers, from subnet buybacks, or eventually from ETF flows.

Grayscale Research argued ahead of the halving that the reduced issuance schedule would matter for TAO for the same reason it has mattered for Bitcoin. Will Ogden Moore, an analyst at Grayscale Research, wrote that “Bitcoin’s history shows that reduced supply can enhance network value despite smaller rewards, as its network security and market value have strengthened through four successive halvings,” adding that “Bittensor’s first halving marks a key milestone in the network’s maturation as it progresses toward its 21 million token supply cap,” according to a note covered by The Block.

Whether halving cycles are actually predictive of price, rather than a narrative that becomes self-fulfilling only when enough traders believe it, is a live debate that goes well beyond Bitcoin at this point. Readers who want the wider context should see HOGE Wire’s own breakdown of halving cycle math tested across multiple coins, since TAO now gives that pattern a real test case running on an entirely different kind of network.

The Subnet Leaderboard: Chutes, Targon, Templar and the Real Revenue

Bittensor’s subnet count roughly doubled in 2026. An upgrade the community calls Robin Tau expanded the network’s subnet capacity from 128 to 256 slots in May 2026, opening room for new teams to register once the original 128 slots had filled up. Live subnet data, including which subnets are earning enough emissions to avoid deregistration, is tracked publicly by taostats, Bittensor’s most widely used block explorer. Launching a subnet is not free, either: teams register through a recycling auction priced in TAO, and the cost rises and falls with demand for the limited slots, which is part of why the extra capacity mattered so much: it lowered the barrier for new teams to get a slot rather than having to outbid an existing, possibly underperforming, subnet for one.

Not every subnet matters equally, and the gap in usage between the top handful and the long tail is large. Subnets that cannot attract enough staked TAO to stay competitive on taostats’ own rankings risk deregistration, effectively getting pruned to make room for a new entrant, which keeps the total subnet count from simply growing forever regardless of quality.

Chutes, subnet 64, is an AI inference marketplace that has consistently ranked as the network’s largest single source of external revenue, serving inference requests at prices its operators claim undercut large cloud providers by a wide margin. Targon, subnet 4, run by Manifold Labs, focuses on confidential compute for enterprise workloads using hardware-based security. Apex, subnet 1, is one of the network’s oldest subnets and focuses on improving language model inference through agent-style workflows and fine-tuning data generation.

Templar, subnet 3, briefly became the most-watched project in the entire ecosystem in March 2026, after its operator, Covenant AI, announced it had trained a 72-billion-parameter model called Covenant-72B across more than 70 independently owned nodes using a gradient-compression technique called SparseLoCo, which the team said cut bandwidth needs by roughly 97% without a meaningful loss in accuracy. That technical achievement is also what makes what happened to Covenant AI one month later so notable, and is covered in detail below.

Across the network, subnets that sell an actual product rather than simply existing to farm emissions generated an estimated $43 million in AI service revenue in the first quarter of 2026 alone, led by Chutes and Targon. That is a small number next to a single large centralized cloud provider’s quarterly revenue, but it is real, on-chain, verifiable usage rather than token incentives paid out to insiders, which is the metric Bittensor’s supporters point to when arguing the network has moved past being a pure speculation vehicle.

The Covenant AI Fallout: Bittensor’s 2026 Governance Crisis

On April 9, 2026, Covenant AI founder Sam Dare announced his team was leaving Bittensor entirely, and did so in the most public and damaging way possible, in a dispute later covered in detail by The Block. “The entire premise of Bittensor, the promise that drew builders, miners, validators, and investors into this ecosystem, is that no single entity controls it,” Dare wrote. “That promise is a lie.”

Dare’s specific complaint centered on what he called Bittensor’s triumvirate, a three-person multisignature group that controls protocol upgrades and had been presented to the community as distributed governance. He argued that Jacob Steeves effectively controlled all three signers and used that control to punish Covenant AI during their dispute, alleging Steeves suspended emissions to Covenant’s subnets, stripped the team’s moderation rights over its own community channels, deprecated its infrastructure without process, and sold TAO in a way timed to apply financial pressure. “When a single actor can suspend a subnet’s emissions, override an owner’s authority over their own community spaces, publicly deprecate projects without process, and use token sales as a coercive mechanism to compel compliance, that is not decentralization,” Dare said, summing the arrangement up as decentralization theatre.

Steeves pushed back directly on the specifics rather than the broader charge, in comments captured by CryptoTimes. “I do not have the ability to suspend emissions,” he said. “What I did do is sell some of my alpha holdings on his three subnets, because they were not running, and were on near 100% burn code.” On the moderation dispute, he said simply, “I did not remove his moderator role.”

The market did not wait for a resolution. TAO fell roughly 15% in about two hours after Dare’s announcement, sliding from around $338 to a low near $285, and Covenant AI reportedly sold approximately $10 million worth of TAO on its way out. Whatever the truth of any individual allegation, the episode exposed a real structural fact about Bittensor: a network built to celebrate decentralization still routes its most consequential decisions through a small group of people, and its most technically credible subnet team walked away publicly calling that arrangement fraudulent.

Opentensor’s Response: The Decentralization Roadmap

Bittensor’s leadership did not ignore the fallout. In the weeks after the Covenant AI dispute, Steeves proposed a staking lock-up mechanism intended to tie subnet ownership to long-term commitment rather than short-term token sales, a direct response to the accusation that he had used sell pressure as leverage against a dissenting team.

The bigger structural answer came on June 22, 2026, when Opentensor published an 18-month decentralization roadmap targeting full decentralization of the network’s economic and governance layers by December 2027. The plan calls for three main changes:

• Reintroducing competition among validators to reduce stake concentration within individual subnets
• Opening subnet liquidity pools to two-way investment so capital can move more freely in both directions
• Building a belief mechanism that gives Alpha token holders direct voting power over subnet decisions rather than leaving control with a small operator group

Steeves also stepped back from his own formal role, stepping down as chief executive of the Opentensor Foundation while remaining active writing protocol code and building subnets, a move the project has framed as evidence that decentralization is proceeding rather than stalling. Whether that framing survives contact with the next governance dispute is, at this point, an open question rather than a settled one.

Security History: The 2024 Supply Chain Hack

Bittensor’s most serious security incident to date did not target the blockchain itself. On July 2, 2024, an attacker compromised a version of Bittensor’s Python package distributed through the PyPI package repository, the standard way developers install the software needed to interact with the network, as CoinDesk reported at the time. Users who downloaded the tampered package between May 22 and May 29, 2024 unknowingly handed their wallet credentials to the attacker, who waited weeks before draining funds starting at roughly 7:06 p.m. UTC on July 2.

The theft totaled around $8 million, including one wallet drained of 32,000 TAO, according to a technical breakdown published by security firm Halborn. On-chain investigator ZachXBT was among the first to flag the unusual outflows publicly. TAO’s price fell about 15% on the news. The Opentensor Foundation’s response was to put the chain into what it called safe mode, halting all on-chain transactions for roughly a day while it investigated, removed the malicious package from PyPI, and audited its other published code for similar tampering. Co-founder Ala Shaabana confirmed the halt at the time.

The lesson from the incident had little to do with Bittensor’s consensus mechanism and everything to do with software supply chains and the credential-theft techniques attached to the tampered package, the same broad category of attack behind a large share of the phishing-driven theft crypto users deal with every year. HOGE Wire’s look at how modern wallet-drainer campaigns actually operate covers that category from the victim’s side, even though this specific breach moved through a poisoned software package rather than a malicious signature request.

Who’s Backing TAO: Venture Capital and the Nvidia Question

Bittensor never ran a conventional token sale, but it has attracted an unusually concentrated set of large holders for a network its size. Fortune reported that Polychain Capital, the venture firm founded by former Coinbase employee Olaf Carlson-Wee, backed the project early and has been said to hold TAO exposure worth around $200 million as of the first quarter of 2026. Digital Currency Group, Barry Silbert’s crypto conglomerate, has been reported as the single largest holder of TAO among institutional investors, with roughly 500,000 tokens, worth an estimated $175 million at the time it was reported. The investment fund dao5 has also been reported to hold a position worth around $50 million.

Part of the appeal for early backers is structural rather than purely speculative: staking TAO into a subnet under dTAO produces an ongoing yield in Alpha tokens on top of any price appreciation, which gives large holders a reason to stay locked in rather than trade in and out the way they might with a typical layer-1 token. That dynamic helps explain why funds that got in early have generally preferred to keep building positions rather than exit into strength.

The more striking claim circulating in crypto trade press in early 2026 was that Nvidia had deployed approximately $420 million into TAO during the first quarter, with roughly 77% of that position locked into staking. That figure spread widely across crypto news outlets starting in April 2026, but it rests on on-chain analysis and secondary reporting rather than a confirmed disclosure from Nvidia itself, so it is worth treating as a widely repeated claim rather than an audited fact. Whatever the precise numbers, the pattern is consistent: sophisticated, well-capitalized investors have been willing to take large, multi-year TAO positions well before the token had any path to a regulated wrapper that ordinary investors could buy through a brokerage account. That gap is what the next section is about.

The ETF Race: TAO Meets Wall Street

Grayscale had already been offering accredited investors exposure to TAO through a private placement trust. On December 30, 2025, it filed a Form S-1 with the SEC seeking to convert that trust into a spot exchange-traded fund that would list on NYSE Arca under the ticker GTAO, with an amended filing following on April 2, 2026. Bitwise filed its own application around the same period for a TAO-focused product structured as a hybrid, holding roughly 60% of assets directly in TAO and the remainder in a TAO-linked exchange-traded product, part of a larger batch of spot crypto ETF applications the firm submitted covering eleven different tokens, as widely reported when the filings landed.

Neither filing guarantees approval. The SEC’s review clock for this kind of registration statement typically runs in stages over several months, and crypto trade press has pointed to around August 2026 as a plausible window for a decision, though the agency has repeatedly used its full statutory review period on other crypto ETF applications rather than moving early. Readers who want the fuller regulatory backdrop, including how the SEC’s posture toward crypto has continued shifting through 2026, can see HOGE Wire’s tracker of US crypto enforcement and rulemaking for the wider picture beyond this one filing.

An approved TAO ETF would matter less for the amount of money it brings in on day one than for what it signals: that a token built entirely around funding decentralized machine learning subnets has become mainstream enough for a regulated fund issuer to bet on. It would also hand TAO something Bitcoin and Ethereum ETFs already have and that most other AI-crypto tokens still lack, a clean, custody-free way for a retirement account to get exposure.

Bittensor vs. the Rest of Decentralized Compute

Bittensor gets lumped in with decentralized GPU marketplaces constantly, and the comparison is useful mainly because the differences are so specific. A network like Akash rents out CPU, GPU and storage capacity through a reverse-auction marketplace where providers bid down prices for standardized jobs. Render does much the same for GPU-heavy graphics rendering. io.net aggregates GPUs from multiple sources into large clusters for AI training and inference. All three primarily sell compute itself, priced like any other commodity.

Bittensor sells something closer to graded output. A subnet miner is not paid simply for having a GPU available; it is paid according to how validators score the actual result it produces, whether that is an inference response, a trained model checkpoint, or a curated data set. Gensyn, which reached mainnet in April 2026, sits somewhere in between: it focuses specifically on cryptographically verifying that distributed machine learning training was executed correctly, using a dispute-resolution system called Verde rather than Bittensor’s stake-weighted validator scoring.

For an investor, the practical distinction is where the yield comes from. Holding AKT, RENDER or IO is mostly a bet on rental demand for compute and on each token’s own supply mechanics. Holding TAO, or staking into a specific subnet’s Alpha token, is a bet on which categories of AI work end up being worth subsidizing through token emissions at all, a messier and more speculative judgment call, but one with a much higher ceiling if a subnet like Chutes keeps growing into a genuine alternative to renting inference from a large cloud provider.

None of these networks yet operate at a scale that seriously threatens centralized cloud AI infrastructure. The combined revenue of the entire decentralized GPU sector remains small next to a single large cloud provider’s quarterly numbers, a gap worth keeping in mind against Bittensor’s own $43 million quarterly figure. The pitch from all of them, Bittensor included, is not that they beat centralized providers on scale today, but that a permissionless, censorship-resistant alternative is worth building before it becomes urgently needed.

TAO Price Action: From $757 to a Halving-Era Reset

TAO’s price history is an extreme version of a pattern familiar across the AI-crypto sector: a fast run to an all-time high early in its life, a long drawdown, and repeated sharp rallies and drops tied to specific news rather than a smooth trend. The token hit its all-time high of $757.60 on March 7, 2024, during the broader crypto market’s first wave of AI-token enthusiasm, then spent most of 2024 and 2025 well below that level.

2026 has been unusually volatile even by TAO’s own standards. The token rallied around 27% in the first week of January 2026 and gained roughly 21.6% across the first quarter as a whole, helped along by the Nvidia and Polychain reports and the run-up to Grayscale’s ETF filing. The Covenant AI dispute erased a large chunk of those gains in a single afternoon in April. By late June 2026 the token was trading near $250, and by mid-July it had drifted down to around $195, broadly in line with a pullback across AI-linked tokens generally rather than a Bittensor-specific event.

TAO’s drawdown from its all-time high, still deep even after the 2026 rallies, is not unusual within the AI-token cohort; most tokens that led the 2024 AI narrative remain well below their own cycle highs even where the underlying products have kept shipping. That divergence between usage growth and price recovery is one of the more interesting open questions in the sector: either the market is underpricing real revenue growth across decentralized AI, or 2024’s AI-token prices were simply disconnected from any plausible near-term cash flow and are only now correcting toward something more grounded.

Public price forecasts for the rest of 2026 vary enormously depending on whether the analyst is modeling a successful ETF approval, a continuation of the halving-driven supply squeeze Grayscale Research has highlighted, or another governance shock along the lines of April’s. That spread, from targets not far above current levels on the cautious end to figures several multiples higher on the bullish end, says as much about how unsettled the AI-crypto sector’s valuation framework still is as it does about Bittensor specifically.

Risks and Open Questions

Bittensor’s growth in 2026 does not erase a set of risks specific to how the network is built. The clearest is governance: a project whose entire pitch rests on decentralization spent April 2026 in a very public argument about whether one person can unilaterally defund a team he disagrees with. The June 2026 roadmap is a real commitment on paper, but it targets full decentralization by the end of 2027, which leaves a year and a half during which the same concentration of power Covenant AI complained about stays in place.

The dTAO market design cuts both ways, too. Letting capital vote with its feet on which subnets deserve emissions is more responsive than the old validator-vote system, but it also means subnets can lose funding quickly if staked TAO rotates elsewhere. Most subnets are not Chutes or Targon; they are small teams competing for a shrinking share of daily issuance against more than a hundred rivals, with real deregistration risk if their Alpha token price falls too far behind the pack. There is also a simpler concentration risk sitting underneath all of it: TAO’s own holder base. With a handful of venture funds and, if the reported figures hold up, a single chipmaker controlling a meaningful share of circulating supply between them, the float actually available to ordinary holders is thinner than the headline market cap suggests, which can make price moves sharper in both directions than a similarly sized but more widely distributed token would experience.

• Governance still runs through a small multisignature group despite the decentralization roadmap’s long timeline
• dTAO lets staked capital rotate out of a subnet quickly, leaving smaller teams exposed to falling emissions and eventual deregistration
• Centralized AI labs still command vastly more compute than the entire decentralized AI sector combined
• Neither the Grayscale nor the Bitwise ETF filing is guaranteed approval on the timeline the market appears to be pricing in

Competitively, Bittensor and every other decentralized AI network are still tiny next to the compute centralized labs can deploy, and nothing about subnet emissions changes that gap on its own. The bet embedded in TAO’s price is not that Bittensor out-computes a frontier AI lab, but that a permissionless alternative accumulates enough independent value, in inference pricing, in specialized models, in resistance to any single company or government controlling access, to matter regardless of scale. Grayscale’s and Bitwise’s filings are a meaningful vote of confidence in that thesis, but neither the SEC’s review timeline nor its eventual decision is guaranteed, and a rejection or extended delay would remove one of the more concrete catalysts investors have been pricing in through 2026.

Frequently Asked Questions

What is Bittensor (TAO)?

Bittensor is a decentralized network that pays contributors in its native token, TAO, for producing useful machine learning work across specialized subnets rather than for traditional block production. Launched in January 2021 by Jacob Steeves and Ala Shaabana under the Opentensor Foundation, it functions as a marketplace where miners compete to produce the best AI output and validators grade that output to determine rewards, all coordinated on its own blockchain, Subtensor.

How does dTAO (dynamic TAO) work?

Dynamic TAO, live since February 2025, gives every Bittensor subnet its own Alpha token traded against TAO through an automated market maker. Staking TAO into a subnet is effectively a swap into that subnet’s Alpha token, and the more TAO a subnet attracts, the larger the share of daily emissions it earns. This replaced an older system in which a small group of root validators voted directly on how emissions were split between subnets.

How many TAO tokens are there, and when is the next halving?

TAO has a fixed maximum supply of 21 million tokens, following the same model as Bitcoin. The first halving activated on December 12, 2025, cutting daily issuance from about 7,200 TAO to about 3,600 TAO. Because the halving triggers automatically once half of the remaining unissued supply has been mined, rather than on a fixed calendar date, the exact date of the second halving will depend on network activity between now and then.

Is there a Bittensor (TAO) ETF?

Not yet, but two are pending with the SEC. Grayscale filed a Form S-1 in December 2025 to convert its existing Bittensor Trust into a spot ETF trading on NYSE Arca under the ticker GTAO, and Bitwise has filed for a hybrid TAO product as part of a larger batch of crypto ETF applications. Neither had been approved as of mid-2026, and crypto trade press has pointed to around August 2026 as a possible decision window, though the SEC could take longer.

Has Bittensor been hacked?

Yes. In July 2024, an attacker compromised a Bittensor software package distributed through PyPI and used it to steal wallet credentials, draining roughly $8 million in TAO from affected users. The Opentensor Foundation halted the network in a safe mode for about a day, removed the compromised package, and audited its other code. The exploit targeted the software supply chain and individual user wallets rather than the Bittensor blockchain’s consensus mechanism itself.

By the HOGE Wire crypto desk.