You have the hardware, the power contract and the customers in the pipeline. Genesis Grid supplies what sits between them — portal, tenancy, quotas, metering, identity — under your own brand, so your engineers stop building a cloud and start selling one. You still run the platform, with a smaller and less specialised team than a stack of your own would need.
Nobody loses this race to another vendor. They lose it to the control plane, the tenancy model, the metering pipeline and the portal their own team has to build first, while the accelerators depreciate in the rack. Genesis Grid is eight years and two complete rewrites of that work. What the capacity is then worth still depends on demand you can reach: the stack makes capacity sellable, not sold.
from signed contract to first invoice, pilot and cluster onboarding included
to stand the stack up on a further cluster
Because the build is not the portal, it is everything under it: scheduling, isolation, quotas, metering, invoicing and upgrades. That is twenty to thirty engineers permanently, against four to six to run Genesis Grid under a licence. It is also work with few finished answers in the open: we stripped Kubernetes back until it orchestrates virtual machines instead of containers, made InfiniBand multi-tenant over partition keys in a way that surprised NVIDIA's own engineers, and run VPP and SPDK in user space to hold line rate from 200 to 800G.
.svg){kind=small}
Yes, and the mechanism is specific rather than a policy — partition keys on the fabric, segmentation on the network, namespaces in storage, all listed above. Each tenant runs against the service level you sold it rather than a best-effort share of whatever is free. What a tenant environment is available for rests on your building and your hardware, so that number is yours to set. What Genesis commits to is the control plane, the portal and the meter at 99.9 %.
Mixed vendors and mixed generations are workable, and so are facilities below the top topology class — with one honest limit. Uptime Institute's Tier I to IV describe redundancy and what can be maintained while the site keeps running, not availability percentages. Software resilience shifts how much redundancy the building must carry for a given class of tenant; it does not replace it. A deployment is one cluster in one building: if that building loses power, the cluster goes down. We say so before you put a number in a contract.
You keep the facility, the hardware, the commercial terms and the tenant, and you run the platform yourself in both contract forms. Genesis backs your team with escalation support: 30 minutes on severity one, 24/7, with escalation to the engineers who build the stack. The split is written down line by line before anything is signed.
Volta licensed Genesis Grid and is building its platform on it. On 5 August 2026 Volta raised USD 300 million at a USD 2.4 billion valuation, co-led by Andreessen Horowitz and Altimeter with NVIDIA and Michael Dell participating, alongside USD 5 billion of customer financing and a six-year, USD 10 billion cloud contract delivered with Bitdeer from a 133 MW data centre in Norway. Founders Ricard Boada and Sofia Gumuzio came from Brookfield. Volta is the licensee we are allowed to name.
No. Genesis Cloud is a software company today: we license the stack and sell no capacity, so there is no rate card of ours to bid against yours. We were the operator before that — on the market since 2018, more than 20,000 users, and demanding buyers chose the cluster we ran. Photoroom took 256 H100s, and its co-founder Eliot Andres wrote up the decision publicly on 8 July 2024, describing Genesis Cloud as "a German provider that offers an Infiniband cluster that runs on CO2-free hydroelectricity". He states plainly that it is not an endorsement, which is what makes it worth reading.
