Ingest Pricing Now Meters Compute, Not Bytes: How to Forecast and Bill It
Ingest Pricing Now Meters Compute, Not Bytes: How to Forecast and Bill It
Ingest Pricing Now Meters Compute, Not Bytes: How to Forecast and Bill It
Ingest Pricing Now Meters Compute, Not Bytes: How to Forecast and Bill It
Ingest Pricing Now Meters Compute, Not Bytes: How to Forecast and Bill It
Team Flexprice
Editorial
Your bill stops tracking volume and starts tracking work. When bulk ingest pricing changes from a fixed cost per GiB to a usage-based cost based on compute consumed during ingest, the same 10 GiB load costs different amounts depending on schema width, ingest mode, and transformations.
Key Takeaways
Compute-based ingest pricing charges for the CPU time spent parsing, transforming, and indexing your data, so two same-size loads bill differently.
Microsoft's Fabric Eventhouse analysis puts cost per GB ingested near 0.57 sample cost units on one cluster and near 6 on another.
Snowflake runs both models: Snowpipe bills a fixed credit per GB, Snowpipe Streaming Classic bills per-second compute plus client time.
Per-GB pricing hasn't gone anywhere. Datadog lists log ingest from $0.10 per GB, Elastic Serverless from $0.07.
Flexprice meters ingest at up to 1 million events per second under 60ms P99, with an event debugger showing every event.
What is compute-based ingest pricing?
Compute-based ingest pricing charges you for the compute your vendor burns loading your data, not the bytes you sent. The unit becomes CPU-seconds, capacity units, or credits.
One Fabric Eventhouse cluster ingested 1 TB in two hours, used 23 capacity unit hours, and landed near 0.57 SCUs per GB. Another ingested 2.5 TB a day with 10,500 queries, used 817 CU hours, and landed near 6. Microsoft names the drivers:
Schema design, since wide tables need more compute
Transformations: update policies, materialized views, partitioning
Streaming versus queued ingestion
Ingestion volume, since cost spreads as clusters grow
How do you meter compute consumed during data ingest?
Emit one usage event per ingest job with the compute measurement attached. Don't back into compute from byte counts afterward, because that relationship isn't stable.
Emit at job completion, keyed by an idempotency key so retries don't double-bill.
Carry the customer ID, the compute measure in CPU-seconds, and bytes as a second property.
Tag pipeline, region, and ingest mode, so you can price streaming apart from batch.
Sum the compute measure, and keep the byte total for the invoice.
Snowflake shows why step 2 matters: Snowpipe Streaming Classic bills per-second serverless compute and per-second client time, so one job yields two measures. Flexprice's Usage Metering takes both on one event, and our usage metering guide goes deeper.
Your bill stops tracking volume and starts tracking work. When bulk ingest pricing changes from a fixed cost per GiB to a usage-based cost based on compute consumed during ingest, the same 10 GiB load costs different amounts depending on schema width, ingest mode, and transformations.
Key Takeaways
Compute-based ingest pricing charges for the CPU time spent parsing, transforming, and indexing your data, so two same-size loads bill differently.
Microsoft's Fabric Eventhouse analysis puts cost per GB ingested near 0.57 sample cost units on one cluster and near 6 on another.
Snowflake runs both models: Snowpipe bills a fixed credit per GB, Snowpipe Streaming Classic bills per-second compute plus client time.
Per-GB pricing hasn't gone anywhere. Datadog lists log ingest from $0.10 per GB, Elastic Serverless from $0.07.
Flexprice meters ingest at up to 1 million events per second under 60ms P99, with an event debugger showing every event.
What is compute-based ingest pricing?
Compute-based ingest pricing charges you for the compute your vendor burns loading your data, not the bytes you sent. The unit becomes CPU-seconds, capacity units, or credits.
One Fabric Eventhouse cluster ingested 1 TB in two hours, used 23 capacity unit hours, and landed near 0.57 SCUs per GB. Another ingested 2.5 TB a day with 10,500 queries, used 817 CU hours, and landed near 6. Microsoft names the drivers:
Schema design, since wide tables need more compute
Transformations: update policies, materialized views, partitioning
Streaming versus queued ingestion
Ingestion volume, since cost spreads as clusters grow
How do you meter compute consumed during data ingest?
Emit one usage event per ingest job with the compute measurement attached. Don't back into compute from byte counts afterward, because that relationship isn't stable.
Emit at job completion, keyed by an idempotency key so retries don't double-bill.
Carry the customer ID, the compute measure in CPU-seconds, and bytes as a second property.
Tag pipeline, region, and ingest mode, so you can price streaming apart from batch.
Sum the compute measure, and keep the byte total for the invoice.
Snowflake shows why step 2 matters: Snowpipe Streaming Classic bills per-second serverless compute and per-second client time, so one job yields two measures. Flexprice's Usage Metering takes both on one event, and our usage metering guide goes deeper.
Get your entire pricing and billing structured in days
Get your entire pricing and billing structured in days
How do you move from per-GiB pricing to consumption-based pricing?
Run both models in parallel before you retire either one. Splunk lists Workload Pricing and Ingest Pricing side by side: one meters workload types, the other data volume.
Shadow-bill a cycle: charge per GiB, calculate the compute number, store both. Pricing Models runs two rates at once.
Publish the ratio per customer.
Cap cycle one at the old bill.
Cut over by cohort, grandfathering annual contracts to renewal.
Snowflake moved the other way. Its Snowpipe docs say ingestion now bills at a fixed credit amount per GB, replacing per-second per-core compute plus a per-1,000-files fee. Compute pricing is more accurate, and accuracy costs predictability, the trade in every migration to usage-based pricing.
How do you tell customers their pricing model is changing?
Send them the new number before you send the new invoice.
Last cycle's spend recalculated under both models, per customer
The drivers they control: schema width, file size, transformation, ingest mode
A named cap or credit for cycle one, with the amount written down
A forecast endpoint, because a PDF can't answer "what if we double"
Skip that email and your support queue detonates on invoice day. Test the pricing on a cohort.
How do you forecast a bill under compute-based ingest pricing?
Forecast from a measured cost-per-GB ratio for your own workload, not the vendor's rate card.
Measure credits or CU-seconds per GB over a representative week, per pipeline.
Multiply by planned volume, pipeline by pipeline.
Add a variance band. Microsoft's two clusters differ by roughly 10x.
Re-measure after any schema change, since it reprices every future GiB.
Keep a per-GB anchor: Datadog lists log ingest from $0.10 per ingested or scanned GB per month, Elastic Serverless from $0.07 on Logs Essentials.
How do you show ingest cost breakdowns on customer invoices?
Put the compute measure, the bytes, and the applied rate on one line item. A charge the customer can't reconstruct becomes a support ticket.
Compute consumed, bytes ingested, and the rate shown together
A breakdown per pipeline, not one blended line
The underlying events reachable from the invoice
Credits, usage, and subscription fees on one document
How does Flexprice meter and invoice compute-based ingest?
Flexprice captures compute and bytes on one event, then puts both on the invoice line. Flexprice is enterprise-grade, open source usage based billing infrastructure for AI and SaaS companies. It can be deployed in your own VPC, on-prem, or on Flexprice's managed cloud. Against the four requirements above:
Billing and Invoicing puts usage, credits, and subscription fees on one invoice, and Usage Metering keeps the event trail behind every line.
Parent-child accounts, contract versioning, ramped commits with mid-cycle overages, RBAC, and multi-currency across Stripe, Razorpay, Moyasar, and Nomod.
SOC 2 Type 2 and on-prem or air-gapped deployment on the Mission Critical plan.
AGPL-3.0 with 4,300+ GitHub stars, and 20B+ events a month for 100+ customers.
CASParser cut a 15 to 30 minute metering lag to near real-time, live in two developer days.
One honest limit: if you bill flat per GiB and never plan to change it, a metering engine is overhead.
Frequently asked questions
What does compute-based ingest pricing mean on a bill?
Compute-based ingest pricing bills for the compute consumed loading your data rather than the volume you sent, measured in CPU-seconds, capacity units, or credits. Microsoft's Fabric Eventhouse figures put cost per GB between 0.57 and 6 sample cost units.
How do I forecast a bill under compute-based ingest pricing?
Measure your own credits or capacity-unit-seconds per GB over a representative week, per pipeline, then multiply by planned volume and add a variance band. Your schema and ingest mode decide how many units a GiB costs.
Why would a vendor move from per-GiB ingest pricing to compute-based pricing?
Because a GiB of narrow batched data and a GiB of wide streamed data cost very different amounts to load. Not everyone moves that way: Snowflake's Snowpipe docs say ingestion now bills a fixed credit per GB.
Measure your own cost-per-GB ratio for a week before you announce the change. If you're building the metering side, the API is at docs.flexprice.io.
P.S. We re-checked every vendor figure on 2026-08-14 against the vendor's own docs.
How do you move from per-GiB pricing to consumption-based pricing?
Run both models in parallel before you retire either one. Splunk lists Workload Pricing and Ingest Pricing side by side: one meters workload types, the other data volume.
Shadow-bill a cycle: charge per GiB, calculate the compute number, store both. Pricing Models runs two rates at once.
Publish the ratio per customer.
Cap cycle one at the old bill.
Cut over by cohort, grandfathering annual contracts to renewal.
Snowflake moved the other way. Its Snowpipe docs say ingestion now bills at a fixed credit amount per GB, replacing per-second per-core compute plus a per-1,000-files fee. Compute pricing is more accurate, and accuracy costs predictability, the trade in every migration to usage-based pricing.
How do you tell customers their pricing model is changing?
Send them the new number before you send the new invoice.
Last cycle's spend recalculated under both models, per customer
The drivers they control: schema width, file size, transformation, ingest mode
A named cap or credit for cycle one, with the amount written down
A forecast endpoint, because a PDF can't answer "what if we double"
Skip that email and your support queue detonates on invoice day. Test the pricing on a cohort.
How do you forecast a bill under compute-based ingest pricing?
Forecast from a measured cost-per-GB ratio for your own workload, not the vendor's rate card.
Measure credits or CU-seconds per GB over a representative week, per pipeline.
Multiply by planned volume, pipeline by pipeline.
Add a variance band. Microsoft's two clusters differ by roughly 10x.
Re-measure after any schema change, since it reprices every future GiB.
Keep a per-GB anchor: Datadog lists log ingest from $0.10 per ingested or scanned GB per month, Elastic Serverless from $0.07 on Logs Essentials.
How do you show ingest cost breakdowns on customer invoices?
Put the compute measure, the bytes, and the applied rate on one line item. A charge the customer can't reconstruct becomes a support ticket.
Compute consumed, bytes ingested, and the rate shown together
A breakdown per pipeline, not one blended line
The underlying events reachable from the invoice
Credits, usage, and subscription fees on one document
How does Flexprice meter and invoice compute-based ingest?
Flexprice captures compute and bytes on one event, then puts both on the invoice line. Flexprice is enterprise-grade, open source usage based billing infrastructure for AI and SaaS companies. It can be deployed in your own VPC, on-prem, or on Flexprice's managed cloud. Against the four requirements above:
Billing and Invoicing puts usage, credits, and subscription fees on one invoice, and Usage Metering keeps the event trail behind every line.
Parent-child accounts, contract versioning, ramped commits with mid-cycle overages, RBAC, and multi-currency across Stripe, Razorpay, Moyasar, and Nomod.
SOC 2 Type 2 and on-prem or air-gapped deployment on the Mission Critical plan.
AGPL-3.0 with 4,300+ GitHub stars, and 20B+ events a month for 100+ customers.
CASParser cut a 15 to 30 minute metering lag to near real-time, live in two developer days.
One honest limit: if you bill flat per GiB and never plan to change it, a metering engine is overhead.
Frequently asked questions
What does compute-based ingest pricing mean on a bill?
Compute-based ingest pricing bills for the compute consumed loading your data rather than the volume you sent, measured in CPU-seconds, capacity units, or credits. Microsoft's Fabric Eventhouse figures put cost per GB between 0.57 and 6 sample cost units.
How do I forecast a bill under compute-based ingest pricing?
Measure your own credits or capacity-unit-seconds per GB over a representative week, per pipeline, then multiply by planned volume and add a variance band. Your schema and ingest mode decide how many units a GiB costs.
Why would a vendor move from per-GiB ingest pricing to compute-based pricing?
Because a GiB of narrow batched data and a GiB of wide streamed data cost very different amounts to load. Not everyone moves that way: Snowflake's Snowpipe docs say ingestion now bills a fixed credit per GB.
Measure your own cost-per-GB ratio for a week before you announce the change. If you're building the metering side, the API is at docs.flexprice.io.
P.S. We re-checked every vendor figure on 2026-08-14 against the vendor's own docs.
Share it on:
