Vision

Universal Compute Income

A possibility worth testing. Not a prediction, and not a promise.

1 · The idea

Hardware people already own, doing paid work while it waits

Many households own computing hardware that sits idle for long stretches of the day. If enough paid work could safely reach that hardware, its owners could be paid for the time it spends working.

Spread across enough homes, that could become a small, steady income for owning a computer: a Universal Compute Income.

Each part of that sentence is a condition, and most of them are not yet met.

2 · What would have to be true

Ten constraints, any one of which could sink it

These are the reasons the idea might not work. We list them first because they decide everything else.

Ten constraints on the idea, why each matters, and what would ease it
ConstraintWhy it mattersWhat would ease it
Electricity costThe owner pays for the power a job uses. At ordinary household prices, much of what a device could earn goes straight to the electricity bill.Using only hours a device is already on, efficient hardware, and work priced for what it really costs to run.
Hardware wear and noiseSustained load makes fans loud and parts warm. People will not tolerate a machine that is noticeably worse to live with.Owner-set temperature and power caps, quiet hours, and work that pauses the moment the owner needs the device.
AvailabilityHome devices come and go without warning. A job cannot depend on any one of them staying online.Small, restartable tasks, checkpoints, and scheduling that treats a disconnect as normal rather than as a failure.
BandwidthHome upload speeds are limited and uneven. Work that moves a lot of data per unit of compute does not fit.Choosing workloads with a high ratio of compute to data, and caching inputs that many tasks share.
Security and isolationBuyer code runs on hardware the buyer does not control, inside homes where the owner's files and network must stay out of reach.A hardware-virtualised sandbox per task, no access to the home network, and an independent security review before any third-party code runs.
Platform restrictionsPhones, consoles and cars are closed platforms. Their makers decide what can run in the background, not us.Only the platform holders can change this. Until they do, those devices stay out of scope.
DemandThere have to be enough buyers with interruptible, parallel work willing to use consumer hardware. Without them, nothing is earned.Pilots with real buyers to find out which workloads fit, and saying no to the ones that do not.
Competition from data centresCentralised data centres buy power and hardware at scale and run them efficiently. A scattered network starts at a cost disadvantage.Competing on work data centres are not optimised for, and on hardware that is already paid for and already switched on.
Verification overheadResults from devices nobody controls have to be checked. Every check costs compute, which comes out of the price or the contributor's share.Workloads with cheap checks, known-answer tasks, and reputation built from a device's history.
Payments and tax in many countriesPaying many small amounts to people in many countries brings fees, minimum payouts and tax reporting that can swallow small earnings.Starting in a small number of countries, a sensible minimum payout, and an established payout provider.

3 · Capacity

Theoretical capacity is not commercial capacity

It is easy to multiply the number of devices in the world by their speed and call the result a supercomputer. That number means very little. What matters is how much survives each of the steps below.

  1. Devices that exist

    Unknown: no reliable public data.

  2. Devices capable of useful paid work

    Unknown: no reliable public data.

  3. Switched on and idle at the same time

    Unknown: no reliable public data.

  4. Owners who choose to opt in

    Unknown: nobody has been asked yet.

  5. Hours with paid work to run

    Unknown: there are no buyers yet.

  6. Value left after electricity and fees

    Unknown: there is no market price yet.

Bar widths are schematic, not measurements. We have found no reliable public figure for any of these stages, so we show none.

Every stage removes most of what came before it. A device has to be capable, switched on, idle, owned by someone who wants to take part, matched with paid work, and still worth running after its electricity and fees are paid. Only the last stage is income.

4 · If it worked

What it would look like if it worked

One flagship consumer GPU, computed with our calculator at its default assumptions, and again with twice the paid hours at twice the rate.

Illustrative

Hardware profile: Gaming desktop, Flagship GPU, 24 GB+

Illustrative monthly figures for one flagship gaming GPU, at default assumptions and with paid hours and rate both doubled
FigureDefault assumptionsIf paid hours and rates both doubled
Hours with paid work, per month60120
Gross, per month$10.80$43.20
Extra electricity−$5.18−$10.37
Platform fee−$1.62−$6.48
Net, per month$4.00$26.35

The honest reading: at today's assumptions, the best consumer hardware nets about $4.00 a month. Doubling both the paid hours and the rate gets it to $26.35. Neither is an income. Most other household devices would net about nothing. Universal Compute Income is a long way off, and it may not be reachable at all.

Default assumptions
Gaming desktop (Flagship GPU, 24 GB+), available 8 hours a day, with paid work for 25% of that time at a hypothetical $0.18 an hour, electricity at $0.18 per kWh, counting only the extra electricity of a device that is already on, and a 15% platform fee.
Doubled scenario
Gaming desktop (Flagship GPU, 24 GB+), available 8 hours a day, with paid work for 50% of that time at a hypothetical $0.36 an hour, electricity at $0.18 per kWh, counting only the extra electricity of a device that is already on, and a 15% platform fee.
Where the rate comes from
Our planning assumption, inside the range of owner payouts we have seen posted for comparable consumer GPUs. We have not verified those figures. Not a SpareGrid price: there is no market yet.

5 · Falsifiable

How we would know we are wrong

Three signals. If any of them holds up in a pilot, we intend to say so on this page.

  1. 01

    Pilots cannot find workloads that fit

    If buyers with real work cannot cut it into small, interruptible, checkable tasks that tolerate consumer hardware, there is nothing to run. Then the idea fails at the start.

  2. 02

    Verified work does not cost less than cloud spot capacity

    Once checking, retries and fees are paid for, the cost per verified unit of work has to come in below what a buyer could get from interruptible cloud capacity. If it does not, buyers have no reason to come.

  3. 03

    Contributors lose money at typical electricity prices

    If a typical contributor's net is negative once their own electricity is counted, the network would be asking people to subsidise it. Then it should not exist in this form.

6 · Stages

Where this stands

No dates. Each stage happens only if the one before it works.

  1. Now

    Implemented

    This website, its calculator, and research into which workloads and devices could fit. The network itself does not exist.

  2. Next

    Planned

    A closed pilot with a handful of devices and one buyer, on non-sensitive work whose results can be checked.

  3. Later

    Planned

    A public beta for the devices the pilot shows are worth supporting, if the pilot works.

  4. Vision

    Conceptual

    Universal Compute Income: enough paid work reaching enough homes that owning a computer becomes a small, steady source of income. It may never happen.

7 · People

Who is behind this

Team and company details are not on this page yet. We intend to publish them before any beta, so that anyone deciding whether to trust SpareGrid with a device or a workload can see who they would be dealing with.

See the two halves it would take

Jetpack is the planned app for people with devices. SpareGrid Compute is the planned marketplace for businesses with work.