Calculator
Committed use discount calculator
Estimate what Google Cloud committed use discounts are worth for your machine family and region, at real, current rates. Then see why that number is wrong the moment your demand is a book of contracts rather than one flat line. For the wider picture, see how CUD optimization works.
Rates
n2 · us-central1 · per vCPU / month
List price, 2026-07-30 · sustained-use applied
If you commit to this steady demand
$9,692saved per year
on a 36-month commitment — $1,038 a month, $37,383 in total, which you cannot cancel, resell or exchange
Covering 100 vCPU of n2 in us-central1 with the cheaper term, versus on-demand for the same usage.
That assumes all 100 vCPU runs every month for three years. If any of it sits on contracts that end — or hasn't started yet — this number is wrong in both directions. See what that costs ↓
Monthly cost at this demand
100 vCPU · n2 · us-central1Solvicus solves this across every resource you run — driven from your AI assistant, with a dashboard to review and edit the plan. Free to start, and it never touches your cloud console.Create a free account →
That assumed none of it ever ends.
Some of it does, on dates you already know. Real demand is a book of contracts. Each has a termination date, a renewal you expect or don’t, and signed work not yet started. Review your bill every month and you will still commit three years of capacity to a customer who leaves in month seven. An average of what you have used cannot show you a contract that is about to end.
That asymmetry is the whole thing. Demand arriving late costs you a month of on-demand, and you recover. Demand leaving early costs you a commitment you cannot cancel, resell or exchange, for up to three years.
Here is what that costs, on three books. Nothing to fill in.
High-churn agency
5 rows · running today 840 vCPU · peaks at 840 in month 1 · 317 by month 42.
Show the 4 contracts and 1 pipeline deal
Every figure is editable — change a renewal rate or move a termination date and the plan re-solves.
Priced with n2 · us-central1 · vCPU — change the family or region at the top of the page.
Which of these looks like your book?
Large contracts terminating with moderate renewal. Reviewing the bill each month, you buy at the top and hold. A trailing average cannot show you a termination that has not happened yet, and a commitment cannot be cancelled once it has. Expected demand runs 840 vCPU today, peaks at 840 in month 1, and settles at 317 by month 42. Reviewed monthly from billing alone, it commits to capacity the book says goes away.
Where the two plans differ
n2 · us-central1 · vCPU · 42 monthsCoverage falls away at the right-hand edge of both plans. Neither is giving up: a fresh 1- or 3-year commitment bought to cover the handful of months still inside the window would bill long past it, and cost more than paying on-demand for those months. Buying stops before the window does — for the same reason it would in your own last quarter before a re-plan.
What it costs, and how it gets there
840 vCPU · 36-month from m2. What good billing-based practice buys — and it saves most of what there is to save.
408 vCPU · 36-month from m1, 250 vCPU · 12-month from m1.
What this plan bets, and what the bet costs
Renewals and pipeline are expectations. Commitments bill whether you use them or not. If nothing renews and nothing closes, this plan leaves 4,873 unit-months of committed capacity unused, and you pay at least $27,000 more than you would have paid knowing that in advance. At least, because the plan it is measured against is this calculator’s approximation, not a true optimum. That is the bet. It is why the renewal percentages are worth arguing over before you sign.
Google list price as of 2026-07-30 — $23.08 on-demand, $14.54 1-year, $10.38 3-year per vCPU per month at 730 h/mo. Verify current pricing on Google Cloud before you commit. The figures above compare against the sustained-use rate, not list: Google applies that discount to n2 automatically once a resource runs most of the month, and these figures assume yours runs all of it.
Data: Google Cloud Billing Catalog API. Solvicus is not affiliated with or endorsed by Google LLC.
Scoring runs the same arithmetic kernel as the product. The commitment mix is this calculator’s own approximation, costing at least 1.4% more than the best plan we can find. The comparison is a buyer reviewing their bill monthly, given everything except the contract book.
How these numbers are calculated — the demand model, the kernel, and what the comparison assumes
Contracts contribute their full amount to their termination date, then an expected renewal tail: the same contract length again, weighted by the renewal percentage, compounding at each roll. Pipeline deals are weighted by likelihood throughout. Plan scoring — monthly committed and on-demand cost, coverage, and totals — runs the identical arithmetic kernel that powers Solvicus’s own product, verified against the same conformance test vectors used internally. It runs nominally, without present-value discounting, and without rounding on-demand up to whole units, so figures here can differ slightly from the same book run inside the product. Discounting was tested rather than assumed away: at a 12% annual rate the ranking is unchanged and the lead book’s edge moves 16.8% to 14.8%. Plan generation — the suggested commitment mix — is this calculator’s own heuristic approximation, not the product’s LP/MIP solver. On these three books it costs at least 1.4% more than the best plan we can find, on the worst of them. That is a floor, not a ceiling: the comparison is against an independently-optimised plan, which itself only bounds the true optimum from above. The error is one-signed, since the heuristic never beats the optimum, so every saving shown here is understated rather than overstated.
The comparison plan is modelled as someone reviewing their bill monthly and sizing to the trailing 60-day average, on a one-month reporting delay, buying the whole shortfall and never selling. It is given clean data, optimal terms, and the same planning horizon as the book plan. The only thing withheld from it is knowledge of what is already scheduled. That is deliberate: it isolates the value of the forward information rather than the value of paying attention. Note the asymmetry it leaves in place. The heuristic above applies to the book plan but not to this arm, which chooses terms exactly, so the gap shown is understated on that count too. Both plans are evaluated over the same fixed 42-month window, the span these books were authored over. It is not an input, because the answer is not stable in it. Past month 42 every contract here has ended and only compounding renewal expectations remain, and the comparison arm’s term choices cross the 12/36-month lattice at horizons that have nothing to do with the businesses being modelled.
Commitments are costed for their full term even where that runs past the window, because you owe the money either way. Where that happens the ledger names the amount, for both plans. Buying late means buying past the window, and on a growing book that is most of the difference between them.
Your book isn’t one of these three
And it doesn’t belong on a marketing page. Contract values, termination dates and renewal expectations stay yours. Bring them to Solvicus instead. It solves the cost-optimal mix of 1- and 3-year commitments across every resource you run, net of what you already own, and lets you edit the plan yourself. It recommends and evaluates. It never buys anything or touches your cloud console.
What you bring is what you just edited: amounts, dates, renewal odds. You describe it to your assistant rather than typing it into a schema. You split demand by machine family and region yourself, since a resource-based commitment in one cannot cover another. They go in together and come back as one plan across the whole estate.
And if your book turns out to be the stable one, where the plan barely moves, the number is not the only thing that changes. A commitment you cannot cancel is a decision someone has to defend. The plan records which demand each commitment covers, so the rationale travels with the number. Same purchase, stronger basis.