Stop paying servers for work your customers’ devices can do.
Hadron runs each job in your app at the cheapest place that can do it, starting on your customer’s own device.You add one call. Hadron tries nine places, cheapest first: four on the device, then the servers you register, out to hyperscale GPUs last.
This is a real job, running on your device right now.This is a real job, running in this tab right now.
Your device
Costs you leastServers you control
The cloud
Costs you mostWhat just happened
One small job runs on this device. The result appears here.
What a whole company could saveWhole products, job by job
See every companyThree kinds of software company, and the server work their users’ own devices could do instead.Each company’s jobs: how many can run on users’ devices and how many stay on servers. Device jobs run and are checked in its demo; server jobs are never run from this site.
Team workspace
A team workspace app with about 100 million usersTeam workspace, about 100 million users (reported, rounded down)
Saves up to £23.3 million a year
Units a year x list price x device share
At public list prices, with 9 in 10 of its users’ devices doing the work.At public list prices: about 1 million people paying for its AI features, device share 0.9.
8 jobs can run on users’ own devices. 6 stay on servers.8 device jobs; 6 server jobs, never run here.
Video meetings
A video meeting platform with billions of meeting hours a yearVideo meetings, billions of meeting hours a year
Saves up to £344 million a year
Units a year x list price x device share
At public list prices, with 9 in 10 of the devices in each call doing the work.At public list prices: 10 million people with AI notes on, device share 0.9.
5 jobs can run on participants’ own devices. 8 stay on servers.5 device jobs; 8 server jobs, never run here.
Supplier compliance
A supplier compliance platform reading 1 million pages a monthSupplier compliance, 1,000,000 pages a month (an example volume)
Saves up to £432,000 a year
Units a year x list price x device share
At public list prices, with its suppliers’ own computers doing the work. The volume is an example.At public list prices: forms extraction, reading and sorting, device share 0.9; the volume is an example.
5 jobs can run on suppliers’ own devices. 7 stay on servers.5 device jobs; 7 server jobs, never run here.
Public list prices for each kind of cloud job, checked October 2026, and published usage figures for apps of this size, rounded down. Work that has to stay on servers is left out. A company that runs this work on its own servers pays less than list price.Each figure: units a year x public list price x device share, at 0.75491 pounds to the dollar, rounded down to 3 figures. Prices checked 15 October 2026. Server jobs are left out; a company that runs this work on its own servers pays less than list price.
See every companyCheapest place first
Every job starts on the device in your customer’s hands. Work done there is work you don’t pay a server for, and the rest runs on the servers you choose.You write one call. Hadron goes through the nine places, cheapest first, and every place gives back the same kind of answer.
- Where the job ran
- Ready if needed
- Down or missing, so the line passes through
Every server on this line is one you set up. Hadron sends work only to the places you choose.Work goes only to servers you register. With none registered, as on this site, every job stays on the device.
Costs you least
-
On your customer’s device
The phone, tablet or laptop they already have. Nothing to pay for, nothing to upload, no waiting on a network.The browser tab itself. No network round trip and no server bill.
Graphics chipWebGPU
Built for heavy number work, such as pictures and long lists.
Graphics chip, in batchesWebGPU, in passes
The same chip again, for very large work.
Background processorsWeb Workers
Spare power beside the page, which stays smooth.
Main processorMain thread
The part that runs the page itself. Every device has one.
-
Servers you control
Nearby server
Edge endpointA small server you run close to your customers, so the trip is short.Edge functions you run, close to your users.
Your own servers
On-prem endpointMachines you already run, for work you’d rather keep in your own building.Hardware you host yourself, registered as an endpoint like any other.
-
The cloud
Cheapest cloud capacity
Spare or lower-cost capacitySpare or lower-cost cloud capacity, so the same work costs less.Spare capacity or a lower-cost region, so the same work costs less.
Standard cloud servers
Cloud endpointYour usual cloud, so nothing you run today has to move.Your regional services, registered once.
Hyperscale cloud GPUs
GPU cluster endpointThe biggest machines, last in line, so you pay for them least.The biggest machines, last in line.
Costs you most
If a place is down
The job moves on to the next place and still finishes. Your app keeps working.
What changes for your product
Smaller cloud billsFewer server calls
Work that runs on your customers’ devices is work you don’t pay a server to do, and the rest goes to the cheapest place that can take it.Work the device can do never reaches your backend, and the rest goes to the cheapest place that can take it.
Faster appsNo round trip
No trip to a server and back, so results appear while people are still typing.Nothing waits on the network for work the device can do itself.
Files stay with your customersData stays in the page
Photos, statements and logs are handled where they already are. Nothing to upload, nothing to store.Inputs are processed in the browser. Nothing leaves unless you send it to an endpoint you set up.
Keeps working when a server doesn’t
If one place is down, the job moves to the next and still finishes.
Try it on this device
See all demosEach demo does real work here, on the device you’re using, then shows you the result.
The cheapest computer is the one your customers already have.Add one call. Keep your servers for the jobs that need them.
Have a job that’s slow or costly to run? Bring it to us and we’ll show you where it could run instead.Want to try Hadron on one of your own workloads? We’ll pair with your team on it.