Secure · Sovereign-ready · Low carbon
AI workloads executed on distributed UK infrastructure powered by renewable and low-carbon generation — with energy provenance measured at the level of the individual job, not bought as a certificate after the fact.
Most providers sell you an hour of a graphics card and leave the rest to you. We think that is the wrong unit. An hour is only meaningful if you already know exactly what silicon is behind it, and it tells you nothing about what you actually got.
So Beyond Grid prices the work: documents processed, tokens produced, jobs completed. The unit stays honest as hardware changes, it is the thing your finance team can compare against what you pay today, and it is what our ledger already measures for every job we run.
Give us a deadline instead of a schedule. We run your work when clean power is available and pause when it isn't — chasing surplus generation across the estate.
The core product. Ordinary turnaround on UK nodes, with the same per-job energy record and Carbon Ledger entry as everything else we run.
Commit to a quantity of work over a delivery window. We plan capacity around it, and you get a better rate and a guaranteed place in the queue.
Dedicated nodes, no shared workloads, customer-held keys, and placement rules the scheduler enforces — approved hardware, approved sites, carbon ceilings.
We are not publishing rates yet, and we would rather say so than post a number we cannot stand behind. Our first production node is being commissioned now, and until it has run real workloads we do not have an honest figure for what an hour of it is worth.
What we will do instead: take your actual workload, benchmark it, and quote against what you pay today. That is a more useful conversation than a rate card, and it is the only one where both sides know what is being compared.
Buying a renewable certificate a year in arrears tells you almost nothing about the electricity that actually turned your prompt into an answer. We do it the other way round: meter the machine, attribute the energy to the job that caused it, and record where, when and on what supply it ran.
Every workload produces a ledger entry. It is not marketing — it is the row our own scheduler acted on, kept so it can be audited afterwards.
We say measured, not certified. A company marking its own homework is not an assurance regime, and we are not going to pretend otherwise.
The methodology — metering, energy attribution, carbon accounting, ledger issuance and audit trail — is designed to be handed to an external assessor. Bringing one in is a stated commitment, not something already done.
The figures above illustrate the format of a ledger entry. They are not a customer's results.
A conventional cloud asks one question: where is there a spare machine? Fabric asks two — where is there a spare machine, and where is there power worth using right now. That second question is the entire business.
Nodes pull work outbound over HTTPS, so they sit behind a farm's broadband with no VPN, no static address and no inbound firewall rule. Every job is scored against every online node on cost, carbon, spare capacity and surplus generation — and the reason it went where it went is written down and kept.
Cost, carbon intensity, queue depth and surplus generation, weighted by how urgent the work is. Interactive work buys speed. Batch work buys clean power.
Work with a deadline and no urgency is held rather than burned on grid import. In our last recorded run that behaviour kept every completed job off the grid entirely.
The scheduler's own decision, the meter reading, the site and the timestamps become the ledger entry. Nothing is reconstructed later.
Losing a node loses no work. A machine that goes quiet is presumed lost, its power figures are zeroed rather than left stale, and whatever it was holding goes back on the queue for somewhere else to pick up.
Britain is full of generation that cannot get to market. Anaerobic digesters, solar arrays and hydro schemes with output the grid will not take, or will only take at a price that makes the export barely worth metering. Connection queues are measured in years.
A kilowatt-hour exported earns pennies. The same kilowatt-hour turned into computation is worth many times that. So we put the load next to the generation, behind the meter, and let Fabric decide what work is worth running on it.
That is why the energy story is not a marketing layer bolted onto a compute business. It is the reason the compute can exist at all.
Fabric only works if all three sides show up. Whichever one you are, the conversation starts the same way.
You have an AI workload — document processing, extraction, embeddings, batch inference, evaluation, fine-tuning — and you care where it runs and on what power. Bring us the workload you are paying for today and we will benchmark against it.
Reserve capacity →You run a digester, an array, a hydro scheme or a site with generation the grid will not take. We put compute behind your meter and pay for electricity that currently earns you an export rate you would rather not discuss.
What we look for →You have compute sitting idle, or you want to host a node. Fabric handles enrolment, scheduling, metering and billing; you supply the machine and the roof over it.
Enrol a node →Beyond Grid Fabric — the control plane, the scheduler, the job ledger, the failover and the Carbon Ledger — is built and tested. Sentinel, our site intelligence platform, is live and is how we choose where nodes go.
Our first production node is being commissioned now. Until it has run real customer workloads we will not quote a rate, and we will not describe anything as certified until an external assessor has been through the methodology.
If that candour is useful to you, we are probably the right people to talk to early. If you need capacity at scale this quarter, we are not — and we would rather say so now.
We partner with specialists in hardware, operations, compute platforms and data strategy — building a scalable ecosystem for distributed AI infrastructure.
If you have a workload, send us a description of it — volume, model, turnaround, and what you pay for it now. We will come back with a benchmark rather than a brochure.
