ABOUT
Heavy AI work comes in bursts. Owning GPUs for the bursts means paying for them in between.
THE THESIS
Most teams need a lot of compute some of the time.
A retraining run, a backlog of footage, a sweep of ten thousand simulations: the work is real, it is heavy, and it is occasional. Hardware bought for the peak sits idle for the rest of the month.
E31 Network is being built so a team can rent exactly the compute a job needs, for exactly as long as it runs, and stop paying when it is done.
PRINCIPLES
Four commitments the design keeps coming back to.
The job is the unit
Everything on the network is scheduled, isolated, and billed per job. There are no instances to forget about.
Pay for what runs
The meter runs while compute is held and stops when the job finishes. Idle time is not the customer's problem.
Real-time is not ours
Control loops, safety systems, and millisecond inference stay on the customer's own systems. The network is not being designed to be in that path.
Say what is unfinished
Open problems are published alongside the design, and nothing here claims to have shipped.
ELEMENT 31
The parent line.
E31 Network is the on-demand compute line of Element 31. Element 31 also builds Czar, an appliance for teams that want AI compute in their own rack; the network is for the jobs that do not need to own it. This site sells neither.
THE GROUP
Built by the people who answer for it.
A small systems engineering group inside Element 31 that publishes design notes, names its open problems, and writes before it ships.
- ORGANIZATION
- E31 Network — a line of Element 31
- LOCATION
- San Francisco, California
- STATUS
- Technical preview · in active development
- FOCUS
- On-demand compute for batch AI workloads
TECHNICAL PREVIEW — NO JOBS CAN BE SUBMITTED YET; DESIGN-PARTNER INTAKE IS NOT YET OPEN