The discount is on the order form, the money is in the clauses
A Snowflake capacity contract looks like three decisions: how many credits, how many years, which cloud. What actually determines the cost over the term is whether the credit rate is locked, whether unused credits survive the year end, and whether the renewal increase is capped at signing.
Prepared by Redress Compliance · August 11, 2026 · Data cloud advisory. Based on 20 to 30 Snowflake negotiations, 2024 and 2025.
Executive summary
Credit burn ran 25 to 50 percent above plan in the first year of a new commit, while commitments were simultaneously oversized against optimistic adoption curves.
Those two findings look contradictory and they are not: consumption per workload exceeded the model, and the number of workloads that actually landed fell short of it. The result is an estate that overspends on what it runs and forfeits what it committed to and never used.
A three million dollar annual commit consumed at 2.4 million forfeits 600,000 dollars, and 20 percent under consumption compounds to 1.8 million across a three year term. The default capacity contract treats annual commits as use it or lose it, so the forfeiture is silent and automatic.
Rollover of 90 to 365 days is negotiable and some accounts have secured longer rolling windows, but it has to be asked for at signing.
The standard order form fixes the credit price for the initial annual commit only, and reserves the right to adjust list pricing during the term. A customer drawing down credits in year two or three can therefore pay a rate it never agreed to on a commitment it cannot reduce.
The fix is a fixed per credit rate clause referencing the signing date, specified by warehouse size and separately for the serverless features.
The renewal uplift is the cheapest clause to fix and the most commonly skipped. Binding the renewal to a capped increase at signing costs nothing at the moment of signature, because it concerns a number three years away that neither side is arguing about yet.
Left unbound, it becomes the vendor's opening position at exactly the point when the estate is least able to move.
The clauses that decide the economics
| Clause | The default position | The buyer side move |
|---|---|---|
| Credit price lock | Fixed for the initial year only | Fix for the full term at the signing rate |
| Annual rollover | Unused credits forfeited at year end | Negotiate 90 to 365 day carryforward |
| Renewal uplift | Unbound and set at renewal | Cap at signing, indexed |
| Storage pricing | Bundled into the headline discussion | Negotiate per terabyte separately from compute |
| Multi cloud freedom | Assumed but often unwritten | Confirm the cross cloud right on the order form |
| Data sharing | Ambiguous on cross account fees | Confirm sharing without additional fees |
| Cortex AI pricing | Priced into general credit consumption | Model separately, negotiate conversion rights |
The order in which these are fought matters, because they are not all equally contested.
The credit price lock and the rollover terms are the two that materially change the money, and they are also the two the vendor defends hardest, so they belong at the front of the negotiation while there is still commitment size to trade against.
The renewal uplift cap is the opposite case: it is cheap to secure at signing because it concerns a figure three years out that nobody is arguing about, and expensive to secure later because by then it is the live number.
Storage, data sharing, and multi cloud rights are clarifications rather than concessions, and asking for them costs nothing. Sequence accordingly.
Sizing the commit against two errors at once
- Model per workload consumption at a premium to plan, because credit burn ran 25 to 50 percent above plan in the first year and the gap is systematic rather than occasional.
- Discount the adoption curve, not just the rate, since capacity commitments were oversized against optimistic assumptions about how many workloads would actually land in year one.
- Price the forfeiture risk explicitly, as 20 percent under consumption on a three million dollar annual commit compounds to 1.8 million over three years under a use it or lose it default.
- Separate compute and storage in the model, because storage bills per terabyte per month on a different curve and disappears inside a single headline credit discussion.
- Model the AI workload on its own, since Cortex consumption follows a different pattern from analytical query load and negotiating credit conversion rights depends on having that number.
The Snowflake negotiation briefing
Clause by clause positions, commit sizing method, rollover language, and the buyer side moves that hold at signature.
Read the briefing →Two opposite errors, one contract
The most useful thing in the engagement data is a pair of findings that appear to contradict each other. Credit burn ran 25 to 50 percent above plan in the first year of a new commit, and at the same time capacity commitments were consistently oversized against optimistic adoption curves.
Both are true because they measure different things.
The consumption model underestimates how many credits a given workload burns once it is running in production against real data volumes and real concurrency, while the adoption model overestimates how many workloads will have migrated and stabilised by the end of year one.
So the estate simultaneously spends more than expected on the workloads that landed and fails to consume the commitment it sized for the workloads that did not. Under a use it or lose it default those two errors do not net off.
The overspend is billed and the shortfall is forfeited, which is why a three million dollar commit consumed at 2.4 million loses 600,000 in a single year and 1.8 million across a term at 20 percent under consumption.
This is the specific reason the rollover clause matters more than its prominence suggests. Rollover does not reduce the price of anything. It converts a timing error into a timing inconvenience, and timing error is the dominant failure mode in this contract.
A buyer who negotiates 90 to 365 days of carryforward has bought insurance against the exact mistake the data says they are most likely to make, and the ones who secured longer rolling windows removed it almost entirely.
The same logic explains the credit price lock: a commitment that cannot be reduced, drawn down at a rate the vendor may adjust mid term, exposes the buyer on both sides of the same transaction.
Fix the rate for the full term, specify it by warehouse size, and document it separately for the serverless features, because those bill on their own rates and are the easiest thing to leave undefined. The full pricing method sits in the Snowflake negotiation practice.
- Percentile standing for your exact deal size and industry, from real closed transactions
- Scenario simulation before the call: test alternative terms and see the financial impact of each
- A negotiation playbook, talking points, and a two page executive brief on day one
What we saw across Snowflake engagements, 2024 and 2025
Across roughly 20 to 30 Snowflake negotiations run between 2024 and 2025, credit consumption and commit sizing drove more value than the headline rate did, and the vendor's own cost guidance was rarely modelled before signing:
How far credit consumption exceeded the model on new commitments, driven by real data volumes and concurrency in production.
What buyers took against the publisher's first proposal where the clause set was negotiated rather than accepted on the standard template.
Three patterns recurred: credit burn running 25 to 50 percent above plan in the first year of a new commit, capacity commitments oversized against optimistic adoption curves, and rollover and expiry terms quietly stranding prepaid credits at year end.
The buyer side move is to fix the credit rate for the full term, negotiate carryforward, and cap the renewal uplift at signing. The wider library sits in the Snowflake practice.
Your first five moves
- Fix the per credit rate for the full term at the signing date rate, specified by warehouse size and separately for the serverless features, because the default fixes it for the initial year only.
- Negotiate 90 to 365 days of credit rollover, since the use it or lose it default turns a timing error into a permanent loss and 20 percent under consumption compounds to 1.8 million over three years.
- Cap the renewal uplift at signing, which costs nothing now because it concerns a number three years out and costs a great deal later when it becomes the live figure.
- Size the commit against both errors, modelling per workload burn at a premium to plan and the adoption curve at a discount, because the two do not cancel out under a forfeiture default.
- Negotiate storage and Cortex separately from compute, then confirm multi cloud and data sharing rights in writing on the order form. The Snowflake practice runs the full clause set.
Frequently asked questions
Which clause moves the most money?
The credit price lock and the annual rollover.
The standard order form fixes the credit price for the initial annual commit only and treats unused credits as forfeited at year end, so a buyer can pay an adjusted rate in year two on a commitment it cannot reduce, while losing the credits it did not use.
How much do forfeited credits actually cost?
A three million dollar annual commit consumed at 2.4 million forfeits 600,000 dollars. At 20 percent under consumption across a three year term that compounds to 1.8 million in stranded value. Rollover of 90 to 365 days is negotiable and some accounts have secured longer rolling windows.
Why did credit burn run above plan and the commit still go unused?
Because they measure different things. Consumption per workload exceeded the model once workloads hit real data volumes and concurrency, while fewer workloads than planned actually migrated in year one.
Under a forfeiture default the overspend is billed and the shortfall is lost, so the two errors do not net off.
What does the credit price lock need to specify?
The signing date rate, applied to the full term, excluding mid term list adjustments, specified by warehouse size since XS through 4XL each carry separate rates, and documented separately for the serverless features.
Snowpipe, Tasks, Materialized Views, and Search Optimization all bill on their own rates.
When should the renewal uplift be capped?
At signing. It is the cheapest clause to secure because it concerns a figure three years out that neither side is arguing about yet, and the most expensive to secure later, when it becomes the live number and the estate has the least ability to move away.
Is multi cloud flexibility automatic?
It is widely assumed and frequently unwritten. Confirm the cross cloud right explicitly on the order form rather than relying on the general description of the platform. The same applies to data sharing across accounts, where the position on additional fees should be stated rather than inferred.
How should AI workloads be handled?
Model Cortex consumption separately from analytical query load, because it follows a different pattern and disappears inside a single blended credit forecast. Negotiating credit conversion rights depends on having that number, and it is not recoverable once the commit has been sized without it.
Negotiating Snowflake When the Meter Is the Deal
Credit price lock, rollover, and the renewal uplift cap decide the cost of a capacity contract far more than the headline discount does.
