Warppool runs real work on idle GPUs in browsers whose owners opted in. Every root below is someone who said yes — the network grows one tab at a time, on hardware that already exists.
Researchers run free. Publishers earn per GPU-millisecond. No middleman takes 50%.
Nothing runs until you press this. Close the tab and it stops.
A weather lab runs 4,000-member regional precipitation ensembles overnight, distributed across volunteer browsers in the same time zone as the model domain.
See demos →A classroom batches 220k video frames through Apple's open FastVLM model for captioning research. Each task is one frame; tasks stream to whichever idle Warppool device picks them up first.
See demos →An ML team sweeps 20k learning-rate × batch-size combinations on a small classifier. Each task is one config × 5 seeds. Warppool returns the best 50.
See demos →A new data center answers demand with land, power and water. Warppool answers it with hardware that is already built, already plugged in, and already in someone’s hands.
| New data center | Warppool | |
|---|---|---|
| New construction | Land, a substation, a new building | None. The laptops already exist. |
| Cooling | Chillers, often evaporative towers drawing local water | Room air. The work spreads thin across thousands of rooms. |
| What runs on the machine | Anything the tenant installs | WebGPU compute shaders only, inside the browser’s sandbox. |
| Who said yes | A zoning board | The person at the keyboard, before a single kernel ran. |
Reports in which WebGPU returned a working adapter. That is the pool: GPUs already built, already powered on, already in someone’s hands.
Source: Web3D Survey415 TWh, set to more than double to around 945 TWh by 2030 — slightly more than Japan uses today.
Source: IEA, Energy and AIDriven mainly by building more data centres: the embodied carbon in steel and concrete, plus the chips, servers and racks inside them.
Source: Microsoft, 2024 Environmental Sustainability ReportEvery embarrassingly parallel job that runs on Warppool is compute that did not need a new building.
On a publisher’s site, a consent banner comes before the first kernel. On this page, the “Plant this tab” button is the yes.
Jobs are WebGPU compute shaders, run inside the browser’s own sandbox. No installs, no file access.
Nothing is installed and nothing keeps running in the background. The work lives exactly as long as the page does.
For content sites with engaged readers (avg. session >2 minutes), yes. A site with 100k monthly visitors and 50% opt-in rate earns roughly $0.04 per GPU-hour contributed. That works out to comparable or better revenue than programmatic display ads for most niche publishers — and you keep 70% of it instead of the 30–50% left after an ad network's cut. The math gets better as GPU utilization rates improve and compute demand grows.
No — and we built the network specifically to be the opposite. Cryptominers run inside ads, without consent, and benefit the site operator at the visitor's expense. Warppool requires explicit per-session opt-in, surfaces what's running, throttles to background-tab limits, and pays the credit to whoever the visitor authorized. The first thing on every roadmap is making fraud easy to spot.
Researchers with verifiable institutional affiliation (university, public lab, school) run free — no cap, just don't be silly with it. Everyone else is billed per millisecond of completed compute: you pay only for tasks that returned a verified result. Failed tasks, timeouts, canary tasks, and redundant re-runs cost nothing. Indicative rate is around $0.04 per GPU-hour. Embed the worker tag on your own site and you earn credits back at the same rate.
A mid-range integrated GPU (M-series, recent Intel) returns ~1.5 GFLOPS-equivalent on our Monte Carlo benchmark. A discrete RTX-class card returns 30–80×. Calibration runs on join, so the Warppool scheduler always knows roughly how big a task a given worker can swallow in ~2 seconds.