Arrochar Frontiers/Project Hydrabyte

Can a market take data centres off drinking water?A private market for recycled water, built so that the plant gets financed, every trade stays inside the plant's output, and a person decides every order.

Data centres need cooling water, and where they cluster the pipes are already full at peak. A recycled-water plant would solve it, but nobody finances a plant without contracted demand, and no market exists to make that demand bankable. Project Hydrabyte asks whether a licensed precinct whose members trade their recycled-water rights on a private exchange, with a trading agent preparing each member's orders and a person deciding them, can finance the plant and move cooling off the drinking-water network. A working prototype of the exchange runs on a fictional precinct.

Field: Urban water markets
Status: Prototype built, market economics untested
Started: October 2026
Current model: Exchange prototype and workflow guide version 0.3, 9 October 2026

The context

Where the problem stands

A single data centre can ask a utility for 5 to 40 megalitres a day. Where the supply zone is already full, there is no connection to give, however much the operator is prepared to pay. In Sydney the zone serving the Macquarie Park cluster is at capacity and the council has warned that clustering is straining the infrastructure homes depend on. Data centres are under one per cent of Greater Sydney's demand today, about 3.5 billion litres a year, and could add 20 to 25 per cent to total demand by 2035 if proposed sites cool with water. Most of that water is drinking water, and drinking water is the wrong thing to cool with.

The field's answer so far is to meter more and to price the peak. Policy now points to a smart meter at every data centre site and better demand data at the approval stage, and the state's pricing regulator is examining whether charges should follow peak demand rather than average use, so that the few large users who drive augmentation pay for it. Neither creates supply. A recycled-water plant would, and licensed private recycled-water utilities already serve precincts in Sydney under the state's water industry competition regime. But a plant is financed against contracted, creditworthy demand, and without a market in which that demand can be recorded, traded and metered, the plant that would solve the problem is never built. Australia has run tradeable water entitlements in its rural basins for decades. No city runs one for recycled water.

The project reads the problem through three bodies of work. Public reporting on data centre water demand and supply-zone capacity supplies the scale of the problem and the policy response. The private-utility licensing regime, and the regulator's register of who already holds a licence, supplies the legal frame a precinct would run inside. Arrochar Labs' own operating-model work, which compared a government-led market with a private precinct-led one and chose to start private and design for public, together with its evidence chain and its agent-under-mandate engineering, supplies the design position. Where these disagree, the model records the disagreement as an experiment.

The one rule that frames the project

The reserve comes first, and the operator never trades. Before any right is issued, a share of the plant's output is held back for amenity and turndown, so drinking water and housing capacity are never in the market at all. And the party that runs the register and the order book never holds a position in it: a market operator that also traded would have a conflict, so the software is built so that it cannot. Arrochar Labs provides the exchange and its agents. It never trades, never holds water rights and never builds plants. It is the same shape as Project SpaceChain's rule for evidence. There, the record must be sealed against the party that holds it. Here, the market must be closed to the party that runs it.

The hypothesis

W0, the master hypothesis

A licensed recycled-water precinct whose members hold long-term offtake, interruptible tiers and verified offsets as rights on one register, and trade them on a private exchange with a reserve set before any right is issued, can produce contracted demand bankable enough to finance the plant and move data-centre cooling off the drinking-water network, with a trading agent preparing each member's orders under a mandate and a named person deciding them, provided three conditions hold: a tariff exists at which the plant's financiers and the data centres both sign, below the cost of a potable connection and the augmentation it would force; a market of a few members clears often enough for leases and swaps to be worth having, so that the exchange can be paid for by subscription rather than by trading volume; and the whole arrangement can be licensed under existing private-utility law, with no new legislation.

It can be falsified

W0 fails if the gap between what a plant costs per megalitre and what a data centre will pay cannot be closed at any reasonable scale; or if a market of a few members on a simulated year produces so few trades that the register is only a contract ledger; or if a lawyer and the regulator conclude that a tradeable offtake right is a financial product, or that the precinct cannot be licensed as designed. Any of these closes W0 as stated. The register, the metering and the settlement would still be useful to a precinct operator, and the page would say so, but the claim that a market finances the plant would be withdrawn.

The evidence is lopsided, and the shape matters

The evidence for W0 is permissive: every element exists separately. Rural water entitlements trade in Australia at scale, private recycled-water utilities are licensed and operating in Sydney precincts, take-or-pay offtake finances plants in other utilities, and an agent that prepares an order for a person to decide is what Arrochar Labs builds. The evidence against is obstructive: nobody has combined them, a precinct market is thin by nature, the plant is expensive before the first member signs, and the legal status of the traded right is untested. Exactly one objection could kill W0 outright: that the plant is uneconomic against potable water at any tariff a data centre would accept, so that no market, however well run, has anything to finance. It is unproven, unrefuted, and the first thing the programme attacks, because it can be answered with public tariffs and published plant costs.

What the hypothesis commits us to

W0 says the market is unbuilt, not undiscovered, which puts it with Project SpaceGuard and Project SpaceChain rather than Project Chrononaut: the work is economics, law and measurement. It also commits the programme to saying plainly what the prototype proves. The exchange runs: register, order book, metering, curtailment, settlement, reporting and trading agents, on one hash-chained evidence trail. That proves the software can be built. It proves nothing about whether a market will form, and the page will not pretend otherwise.

What we research next

The experiments that can start now

The register is ordered so that the cheapest decisive experiment comes first. The first runs on public tariffs and published costs, the second on the prototype that already exists, the third on paper with adversaries. Only the fourth touches a real precinct, and it does so in shadow, with no money changing hands.

01
Price the gap
Take published potable tariffs, connection and augmentation charges for a data centre cluster in Greater Sydney, and published capital and operating costs for recycled-water plants of the size a precinct would need. Compute the tariff per megalitre at which a plant is financeable on contracted take-or-pay demand, and compare it with what a data centre pays today once connection is refused or delayed.
What settles it. Whether a price exists that both the plant's financiers and the data centres would sign, and at what minimum contracted volume. No such price is the clean negative that closes the project early.
02
Run a year on the prototype
The exchange prototype already runs a fictional precinct with a plant, a reserve, four kinds of member, interruptible tiers, drought stages and a trading agent per member under a mandate. Run it through a simulated year of metered use and curtailment, with members' needs drawn from the demand profiles the first experiment produces, and record every proposal, every decision and every trade on the evidence chain.
What settles it. How often a market of a few members clears, whether leases and swaps carry value when the plant is tight, how curtailment lands tier by tier, and whether anything ever executes without a person except inside the value the person set in advance.
03
Put the model to the law
Write the precinct as a licence application and a set of market rules, and put it to two adversaries. A water lawyer argues whether the operator can be licensed as designed under the existing private-utility regime, and whether a tradeable offtake right is a financial product. A competition reviewer plays a precinct with three large buyers and tries to corner it.
What settles it. Whether the arrangement can run under existing law without new legislation, and whether the holding limits and the reserve are enough to keep a few large buyers from owning the market.
04
Shadow a real precinct
With one licensed precinct operator, run the exchange for a season on real metered data in shadow mode: members see the positions, the proposals their agents would have made and the trades that would have cleared, and nothing is binding and no money moves. The operator's regulator reports are produced from the shadow register alongside the real ones.
What settles it. Whether members would actually trade, whether the agents' proposals survive contact with people who know their own plants, and whether the regulator reports the register produces are ones a licensing regulator would accept.

The objective

The shadow precinct

The programme's near-term objective is one licensed precinct that has run the exchange in shadow for a full year on its own metered data, with the record published: how often the market cleared, what curtailment did by tier, how many of the agents' proposals the members accepted, rejected or changed, and whether a plant financier would have lent against the contracted demand on the register. Nothing on the register is offered as a market until that record exists. A live market with binding trades, and any wider market in which the public utility's own capacity is traded, remain beyond the horizon of the current roadmap, and the model does not pretend otherwise.

Four gates decide everything

1Does a tariff exist that finances the plant and that a data centre would pay? If not, there is nothing for a market to finance and the project closes.
2Does a market of a few members clear? If not, the exchange becomes a register and a settlement engine for a precinct, and the trading claim is withdrawn.
3Can it be licensed under existing law? If not, the project becomes a submission to the regulator and the legislature, and says so.
4Would members trade on a real precinct? If, in shadow, they would not, the model has described a market nobody wants, and the page will say so.

Where the project stands

A working prototype of the exchange exists: registry, continuous order book, daily metering, drought-stage curtailment, settlement, regulator reporting and one trading agent per member under a mandate, every state change on an append-only hash-chained evidence trail, with a kill switch and a human decision queue. It runs on a fictional precinct with invented organisations, prices and readings, and it is open to registered visitors. Nothing is deployed for a customer, no real water or money has passed through it, nothing has been licensed and no precinct operator has agreed to a shadow run. The economics of the market have not been tested against a real plant's costs, and the legal questions are open. From 10 to 27 October 2026 the prototype was published on this site as a product called Hydrabyte Exchange. It moved to Arrochar Frontiers because the open question is not whether the software can be built, which it has been, but whether the market can exist, and that is a research question with a falsifiable answer.

What we publish and what we keep

We publish the context, the master hypothesis, the next experiments and the prototype, to registered visitors with an organisation email. The mandate schema, the curtailment rules tables, the settlement engine, the operating-model comparison and the fee model stay internal. If W0 falls on the first or third experiment, this page will say so.

The prototype exchange

The second experiment runs on software that already exists. Registered visitors are signed in as the registrar of a fictional precinct, with the register, the order book, the agents' decision queue, settlement and a regulator report in front of them. It is a research prototype, not a product, and nothing in it is for sale.

Open to registered visitors with an organisation email. Each request is approved individually. It runs on fictional data.

Sources and data classification

The reading behind the project

Every claim on this page traces to one of the sources below. The published books and papers are public. The two Arrochar documents are internal and are not published.

1
Reporting on data centre water demand, supply-zone capacity and peak pricing in Greater Sydney
W.Media, April and September 2026
Public
2
Reporting on data centre connection requests, smart metering and demand data for planning
Smart Water Magazine, 2026
Public
3
Analysis of data centre water use and cooling demand (Pursuit)
University of Melbourne, July 2026
Public
4
Water Industry Competition Act 2006 (NSW) and the register of licensed private water utilities
NSW legislation; Independent Pricing and Regulatory Tribunal of New South Wales, accessed October 2026
Public
5
Murray-Darling Basin water markets inquiry, final report
Australian Competition and Consumer Commission, 2021
Public
6
Urban water trading operating models (Model A government-led, Model B private-led)
Arrochar Labs, October 2026. Internal. Not published.
Arrochar-owned
7
Hydrabyte Exchange workflow guide, version 0.3
Arrochar Labs, 9 October 2026. Internal. Not published.
Arrochar-owned

Working on any of this?

The next experiments need a water economist, a water lawyer who knows the private-utility regime, and one licensed precinct operator willing to run a season in shadow. If that is you, or you think the hypothesis is wrong and can show why, we want to hear it.

Talk to us All Frontiers projects