Pre-commit GPU, vCPU, and storage capacity for materially lower unit cost — without leaving your jurisdiction. The price curve of a 3-year reserve, the residency posture of a sovereign cloud, none of the multi-year-lock-in baggage.
The longer you commit, the deeper the discount on the underlying compute. Numbers below are normalised against a $1,000/month on-demand baseline so the per-row deltas read at a glance — contract pricing is sized to your actual workload and jurisdiction.
Indicative anchors · final pricing confirmed at contract · regional adjustments apply
Committed capacity is a pricing model, not a separate product — every workload still runs on the same single control plane that addresses bare metal, VMs, GPU pools, and storage in your jurisdiction.
Pick the floor of capacity you'll consume — by SKU, by region — and the term that fits your budgeting cycle. Discount scales with horizon: deeper for 3Y, lighter for 1Y. Mix horizons across SKUs.
Your committed floor is reserved on Navon hardware in your jurisdiction — never overcommitted, never reassigned to a noisy neighbour. Billed at the discounted tier from day one. No upfront payment.
Anything above the committed floor pays at the on-demand rate. No penalty for being under the floor on a given day — the commit is monthly aggregate, not minute-by-minute. Convert up at any time.
For the long tail of unpredictable consumption, on-demand is the right answer. For the predictable floor underneath it — the training run that you know runs every week, the production inference fleet, the working-set storage — committing earns its keep four ways.
The depth of the discount tracks the term. 3-year horizons unlock the floor of the unit-cost curve — closer to colocation economics than to public cloud — without you having to operate the rack.
Committed capacity isn't a quota on a shared pool — it's hardware reserved against your tenant in advance. When the region tightens, your training jobs still run because your slot was already cut.
Committed customers move to the higher SLA tier for availability and incident response, with named account engineering — the engineer who knows your workload picks up the phone, not a queue.
The committed portion lands as a fixed monthly invoice — finance can model the year ahead without forecasting raw consumption. Burst is metered separately so the surprise stays bounded.
The thing that makes hyperscaler reservations painful is that they're rigid — wrong horizon, wrong region, stuck with it. Navon's commit terms give you three escape valves so the floor tracks the actual business.
Move from a 1Y commit to a 3Y at any point. Spend already locked at the lighter discount carries forward against the deeper one — no reset, no penalty.
Scheduled maintenance and announced regional events don't draw down your committed floor — you don't pay for capacity you couldn't have used.
Sovereign customers with multiple legal entities under one parent — ministries, banking groups, holding companies — can share the committed floor across tenants under the same agreement.
Direct answers, no hedging. If something material to your decision isn't here, the capacity desk will answer it on the first call.
Committed capacity is a monthly floor — measured in GPU-hours, vCPU-hours, or storage TB per SKU per region — that you agree to consume over a 1-year or 3-year term. It's billed at the discounted tier from day one. Anything above the floor on a given month bills at on-demand.
The floor isn't a quota on a shared pool: hardware is reserved against your tenant in advance, in the facility you nominated. You could draw down 100% of it, every minute of every day, and it would still be there.
The commit is a monthly aggregate, not minute-by-minute. You're billed for the full floor whether you consume it or not — that's the trade you took for the discount.
Two things blunt the edge: planned-window pauses (announced maintenance and regional events don't burn the commit), and the flexibility to convert the unused floor into a different SKU under the same agreement once per quarter.
Yes — that's the design. Most customers commit the predictable floor (production inference fleet, recurring training jobs, working-set storage) and let the variable layer above it bill on-demand. The metering knows which is which; you don't manage two pools.
The commercial threshold is configured by region and SKU — generally a meaningful floor that warrants the operational overhead on both sides. The capacity desk will quote on the first call; there's no published gate.
For sovereign/government deployments under master-services agreements, the floor scales with the agreement, not with the SKU.
Commit is per region. Capacity is reserved in a specific Navon facility in a specific jurisdiction, because that's how the sovereignty guarantee works — it's not legally fungible across borders.
Multi-region customers run a portfolio of regional commits under one master agreement. Pricing is consistent across regions; absolute numbers reflect regional power, fibre, and tax structures.
By physics first, contract second. The hardware your floor runs on sits in a Navon-operated facility inside your jurisdiction. Encryption keys are held in-country. Operational logs land in-region. No cross-border replication unless you ask for it in writing.
The commercial commitment doesn't change the residency posture — committed capacity is the same sovereign cloud, just at the discounted unit price.
Send us your forecast — even rough — and the capacity desk will come back with an indicative floor, an indicative discount, and the deployment window for your region.