GCAIIAI infrastructure, priced as one system
One subscription covers the hardware, platform, updates and the on-call team. No capital outlay to start, nothing proprietary to unpick if you leave.
How the pricing works
Two things recur. Everything else is either inside them, or happens once before them. There is no third invoice line.
Flat, any node count
Scales with your footprint
One figure, nothing to reconcile
The pilot fee credits back in full — a prepayment, not another line.
Why the subscription exists
The control plane and 24×7 rota cost the same at three nodes or forty — that cost doesn’t follow your hardware, so it isn’t priced against it. Fold it into the nodes and small customers look extortionate while large ones get the platform free.
Why the nodes are separate
Compute is the part that grows with you. Priced per node, so you forecast it by counting boxes and growth stays a line item, not a surprise. Bundle it into one figure and every expansion becomes a renegotiation.
Why anything comes first
Neither line can be sized honestly on day one. The way in finds which constraint actually binds — latency, residency, bandwidth or connectivity — before you commit. Skip it and you’re quoted against a guess.
1 · The subscription
How much of the platform we run for you — one per organisation, whatever your node count. Every level is the same control plane; what changes is how much of the layer above it we operate. The hardware and platform are ours; your code, models and data are yours.
- Nodes and clusters operated for you — bare metal or virtualised
- Scheduling and queues, so allocation is fair and jobs finish (Slurm, Kubernetes, OpenShift)
- Environments ready to use — CUDA, PyTorch, TensorFlow, RAPIDS
- Access control — your SSO for people, roles, MFA
- Health and utilisation in one console, business-hours support
- Everything in Core, plus the 24×7 rota
- Workspaces people actually open — notebooks, VS Code, experiment tracking
- Pipelines run for you, data preparation through deployment
- Cost- and load-aware scheduling, long jobs checkpointed automatically
- Domain packs: geospatial and climate, life science, robotics
- Curated datasets with APIs, reproducible by default
- Everything in Accelerate
- Guaranteed capacity reserved for your team
- Custom workflows built and operated (Airflow, Spark, Dask, Nextflow)
- Data engineering, versioning and lineage
- Compliance accelerators: HIPAA, FERPA, GDPR, export control
- Domain specialists and a named advisor
2 · The nodes
The footprint at each site — pick a size, then count them. Same platform, console and SLA at every size; size is a footprint decision, not a different product. Hardware is in the fee and stays ours to maintain, which is how we put an SLA on gear in your lab, not just the room around it. Prices are 36-month; 12-month terms are 25% higher.
Fanless and low power. Sensors, kiosks, classroom inference, field cases with cellular failover.
Mid-range. Vision and speech at the site, full-rate capture, local model serving.
GPU-accelerated. Local fine-tuning, multi-camera perception, retrieval over past runs.
Training pod, dataset lake, registries, identity and policy. The estate’s centre of gravity.
A worked example
A robotics department — one shared GPU node, four instrumented labs, six bench and field kits. Eleven nodes, one estate, one number.
Two inputs, both yours to change: which level, and how many of each size.
What it replaces — the comparison that matters
One senior platform engineer costs $180,000–260,000 a year fully loaded — and cannot cover a 24×7 rota alone, cannot be two people during annual leave, and does not arrive with hardware.
- Eleven nodes of hardware, refreshed across the term
- The platform, the pipelines and every update
- A named engineer, and the rota behind them
- Statements a grant administrator will accept
- A written exit clause you can test once a year
How you start
This happens once, before the other two exist — nobody signs a subscription first. Design and Pilot run in either order: Design if you need a case for a budget holder, Pilot if you’d rather see it run. The pilot fee credits back in full — a prepayment, not an extra cost.
Two conversations about the workload and the constraint that actually binds. You keep the findings either way, and there is no obligation attached to them.
A reference architecture and a business case you can put in front of a sponsor or a board. Starts with a working session with the people who have the sharpest constraint.
One or two nodes at one site, run as a managed service, no capital outlay. The fee credits in full against your one if you continue.
What stays yours
Priced as one system, but never locked one.
Your code, models and data
Yours at every tier, exported on request in documented open formats. We never hold them hostage and never make them a dependency.
Open standards throughout
Nothing proprietary to attest to. Workloads move in or out without a rewrite, onto your own hardware or a cloud if that is where you end up.
A written exit
Every agreement states what is handed over, in what format and within how many days. If it ends, you keep a working system.
Burst and overflow GPU capacity is available on a consumption basis, for teams whose demand genuinely spikes. It isn’t how we price the platform — metering hours would charge you for the one thing we want you to do more of.
We size the plan to the workload, not the other way around.
Tell us what you are running, where it has to run, and why it cannot all live in the cloud. We will tell you what it costs — and whether we are the right people for it.
Book the assessment