# TAO perpetual > The TAO perpetual on Hyperliquid: the token of Bittensor, a network that pays for machine-learning work in its own token, and what it is exposed to in the compute economy. Updated 2026-09-06. Canonical: https://gpuquant.com/docs/instruments/tao The token of Bittensor, a network that pays for machine-learning work in its own token. A view on decentralized AI compute, not a GPU rental price. ## What it is Bittensor is a network of subnets, each a market for one kind of machine-learning work. Miners produce the work, validators score it, and the network pays both in TAO, its own token, with a capped supply and halvings modelled on bitcoin’s. The value of the token is a claim on those emissions and on demand for what the subnets produce. This perpetual is a leveraged view on TAO, quoted in dollars on Hyperliquid, and not a claim on any subnet or on any compute. ## Its place in compute Bittensor pays for inference and model work in its own token rather than in dollars, so its economics run alongside the token price of inference the site publishes rather than the rental price of a GPU. Both the Silicon Data blended token index and Ornn’s per-lab indices say what a million tokens costs in dollars from the large labs, which is the market a subnet’s output competes in. > **Not a GPU price** > > This row is a claim on a token or a company, and a perpetual on it is a leveraged view on that claim. It is not a GPU rental price and it tracks none of the indices on Market Data. ## The contract A perpetual on Hyperliquid's main market, named `TAO`, quoted in US dollars and settled in USDC. It is a leveraged claim on the token's price and not on the token: it confers nothing the token does. A perpetual has no expiry. Instead, funding is paid every hour between longs and shorts, in whichever direction pulls the perpetual’s price toward the venue’s index of the underlying, and the panel shows the current rate and the countdown to the next payment. The venue's maximum leverage on it is **5x**, margin is cross by default, so the whole account backs the position, sizes are in TAO to 3 decimals, and the smallest order the venue accepts is $10 of notional. Prices are quoted to 2 decimals. Hyperliquid charges its own trading fees, 0.045% on a taker fill and 0.015% on a maker fill at its base tier, and the panel shows the rates that apply to your account where the venue reports them. GPUQuant adds 0.01% of the filled notional, charged by the venue on the fill and never on an order that does not fill. An account enabled from this site is referred by GPUQuant at the venue, which lowers the venue’s own fees; the terms page says by how much. ## Reading it beside the data - [Silicon Data token index](https://gpuquant.com/docs/market-data/silicon-data-token-index): What a million tokens of inference costs across the labs, weighted by spend. - [Ornn token index, OpenAI](https://gpuquant.com/docs/market-data/ornn-token-index-openai): The transacted price of a million tokens on one lab’s models, settled daily. ## Trading it here From the row on Tradable Compute, with an order signed in your own wallet and executed by the venue. The address you log in with is your account at the venue; your collateral stays there and never passes through GPUQuant. The first order signs two one-time approvals, a ceiling on GPUQuant's fee and an agent key held in your browser that signs orders and cannot withdraw; after that every order is signed locally, and what the venue reports as filled is what the page shows. Funding the account, the approvals, the order types and the fees are described once for every instrument on [Tradable Compute](https://gpuquant.com/docs/tradable-compute). ## Risks A perpetual is a leveraged contract. At the venue's maximum leverage a move of a fraction of the price against you liquidates the position, and a fast market can lose more than the margin posted. The liquidation price shown before you confirm is an estimate from the venue's own formula; the venue's engine decides. Prices on the page come from the venue's public feed and can lag or drop, and a market order fills at the book's prices up to the slippage limit shown, not at the price you saw. TAO moves with the crypto market and with the network’s own emission schedule and governance, none of which a compute price explains. Nothing on this site is investment, legal or tax advice. The full terms are on [Terms and risk](https://gpuquant.com/terms). ## Sources - [Bittensor](https://bittensor.com), the project or company behind the underlying. - [Hyperliquid](https://app.hyperliquid.xyz/trade/TAO), the venue's own view of the market. - [Open an account at Hyperliquid](https://app.hyperliquid.xyz/join/GPUQUANT), with GPUQuant's referral code. - [Tradable Compute](https://gpuquant.com/docs/tradable-compute), how every instrument is executed, funded and charged from this site.