Genesis Grid: the stack neocloud operators license

The economics

Serve both markets from one cluster

You sign long-term contracts because they make a cluster financeable and the operation predictable. You sell the hours those contracts leave open into the short-term on-demand market, where published rate cards put the premium at roughly a fifth to a half per hour — though in an oversupplied quarter it narrows, and on a clearing market it can invert. You serve both from the same cluster under one control plane, and that is how idle time gets pushed towards a minimum. Basis: published research, not our own operating records. Execution-idle is measured at 19.7% of in-execution time on a large academic AI cluster (arXiv:2604.04745, April 2026), and the compute share comes from a production deployment of more than 20,000 GPUs, where SM activity rose from 16% to 33% (MuxFlow, arXiv:2303.13803). The revenue figure is a model calculation from those measurements and published rental prices — the assumptions are on the monetization page. If your cluster is already shared and contracted end to end, this number is not yours and we will say so on the first call.

20%

of the hours you already sell, the accelerator sits idle

real compute share per accelerator once one cluster serves many tenants

30%

additional revenue per installed accelerator, no new accelerators or fabric

What changes for the operator

What changes the day you switch it on

Your hardware becomes a product: You put racked accelerators in front of buyers as a self-service cloud under your own brand, on demand or on contract, instead of negotiating every deal by hand.

One cluster serves many tenants: You sell the same cluster to many buyers at once instead of locking it to a single contract. InfiniBand partition keys on the fabric, per-tenant network segmentation, and namespaces with per-tenant credentials and quotas in Ceph and VAST enforce the boundary.

Facility class follows tenant class: Software resilience changes how much redundancy the building has to carry for a given class of tenant. It does not replace it. A deployment is one cluster in one building: lose the power to that building and the cluster goes with it. If you sell availability that depends on maintenance while the site runs, you need a facility that delivers it.

You keep the meter, the SLA and the customer: You price and bill the way the tenant wants to buy — per node, per hour, per token, on demand or on contract — and every unit of consumption is measured. The facility, the hardware, the power contract and the pricing stay yours.

You need fewer specialists, not none: You still staff an operations team on a rota: Linux, network and storage, node health and vendor RMAs. What you do not staff is a platform engineering organisation that builds the cloud itself.

Who you are buying from: An independent software vendor. No chip maker owns Genesis or its roadmap, so your platform and procurement choices stay yours.