Portfolio 01 · cloud bill translation

Multi-cloud AI compute normalizer

Keep one workload definition, enter measured runtime for each candidate, and translate native provider meters into a comparable per-run and monthly bill.

Decision this answers: Which measured candidate has the lowest useful cost after native rounding, startup waste, interruption loss, storage, and transfer?

Scenario

Decision summary

Best cost per run

$113.53exact-derived

Native bill plus fixed storage and transfer.

[1]

Best monthly cost

$2724.74exact-derived

Per-run bill multiplied by workload cadence.

[1]

Useful accelerator-hour

$5.91exact-derived

Cost divided by measured useful accelerator-hours.

[1]

Startup and replay waste

0.15 hmodel estimate

Expected non-useful billed hours per run.

[3]

Direct-labeled result profile

Best cost per run Best monthly cost Useful accelerator-hour Startup and replay waste

The tables below are the accessible source of truth; bar lengths never carry identity alone.

Provider-native candidate matrix

Provider-native candidate matrix
Candidate	Native SKU	Region	Native unit	Runtime	Bill	Normalized	Price kind	Evidence
Google Cloud	measured-gcp-accelerator	us-central1	accelerator-hour	2.40 h measured	$113.53/run · $2724.74/month	$5.91/useful accelerator-hour	negotiated	negotiated[1]
Amazon Web Services	measured-aws-accelerator	us-east-1	instance-second	2.30 h measured	$121.14/run · $2907.36/month	$6.58/useful accelerator-hour	negotiated	negotiated[3]
Databricks	measured-databricks-classic	us-east-1	DBU plus VM-hour	2.60 h measured	$128.88/run · $3093.12/month	$6.20/useful accelerator-hour	negotiated	negotiated[4]
Microsoft Azure	measured-azure-accelerator	eastus	VM-hour	2.70 h measured	$129.06/run · $3097.54/month	$5.98/useful accelerator-hour	negotiated	negotiated[2]

Official surface versus this site

Provider pricing catalogs

Use the official surface to confirm native prices, counting rules, limits, and product eligibility.

This offline workbench

Provider catalogs expose native meters. This page keeps candidate-specific measured runtime and cross-provider operational waste beside those meters; it does not claim hardware equivalence.

Sources and methodology

Every result-affecting reference is visible here without JavaScript and is retained in the JSON export.

[1] Google Cloud Billing Catalog API

Authority: Google Cloud
Native identifier: services.skus.list
Unit: native SKU pricing expression
Evidence: negotiated
As of: 2026-08-01
Retrieved: 2026-08-01
Freshness: reference
Normalization: SKU identity and native units are preserved; the displayed amount is an explicit editable override.

Open the authoritative source

[2] Azure Retail Prices API

Authority: Microsoft
Native identifier: Retail Prices API
Unit: retail meter
Evidence: negotiated
As of: 2026-08-01
Retrieved: 2026-08-01
Freshness: reference
Normalization: ARM SKU, region, unit, and purchase term remain provider-native; the scenario rate is editable.

Open the authoritative source

[3] AWS Price List GetProducts API

Authority: Amazon Web Services
Native identifier: Pricing.GetProducts
Unit: price dimension
Evidence: negotiated
As of: 2026-08-01
Retrieved: 2026-08-01
Freshness: reference
Normalization: Product attributes and billing dimensions are retained; interruption loss is a separate scenario estimate.

Open the authoritative source

[4] Databricks pricing system table

Authority: Databricks
Native identifier: system.billing.list_prices
Unit: DBU
Evidence: negotiated
As of: 2026-08-01
Retrieved: 2026-08-01
Freshness: reference
Normalization: The historical effective interval is retained and infrastructure is not double counted.

Open the authoritative source

Limits and non-claims

Rates in the opening scenario are explicit editable overrides, not claims of equivalent public hardware.
Availability, quota, benchmark performance, and negotiated discounts are outside this static calculator.
A lower bill is not a performance or quality recommendation.