Multi-Cloud Cost Optimization: Beyond AWS

Running more than one cloud spreads your risk and gives you room to negotiate at renewal. It also splits your cost data across three billing models that do not agree on what a unit of compute is, how a discount gets applied, or when a month ends. The optimization techniques transfer between providers. The tooling does not.

Where multi-cloud cost work gets hard

Patterns that carry across providers

1. Standardize tagging across clouds

One schema, applied everywhere. Azure calls them tags, GCP calls them labels and enforces lowercase keys, so choose names that survive the strictest of the three.

{
  "CostCenter": "CC-1001",           // Same across all clouds
  "Project": "CustomerPortal",      // Consistent naming
  "Environment": "production",      // Standardized values
  "Team": "WebDev",                // Same team structure
  "CloudProvider": "aws|azure|gcp", // Track provider
  "Region": "us-east-1",           // Normalized regions
  "Application": "customer-portal"  // App identifier
}

2. Aggregate cost data in one place

Whether you buy a platform or write the collector yourself, the target is one table with a provider column. The shape of the query is much the same in all three SDKs:

import boto3
from azure.mgmt.consumption import ConsumptionManagementClient
from google.cloud import billing

def get_multi_cloud_costs(date_range):
    costs = {}

    # AWS Costs
    aws_ce = boto3.client('ce')
    aws_response = aws_ce.get_cost_and_usage(
        TimePeriod=date_range,
        Granularity='MONTHLY',
        Metrics=['UnblendedCost']
    )
    costs['aws'] = extract_aws_costs(aws_response)

    # Azure Costs
    azure_client = ConsumptionManagementClient(credential, subscription_id)
    azure_costs = azure_client.usage_details.list(
        filter=f"properties/usageStart ge '{date_range['Start']}'"
    )
    costs['azure'] = extract_azure_costs(azure_costs)

    # GCP Costs
    gcp_client = billing.CloudBillingClient()
    gcp_costs = gcp_client.list_billing_accounts()
    costs['gcp'] = extract_gcp_costs(gcp_costs)

    return costs

3. Buy committed capacity on each platform

Every provider sells the same trade: a term commitment in exchange for a discount. The published ceilings differ, and so does how much flexibility you keep.

Those ceilings are vendor headline numbers, quoted for the longest term on the least flexible option. Treat them as an upper bound, not a forecast.

4. Watch data transfer between providers

Egress is where multi-cloud architectures quietly get expensive. Anything chatty that spans two providers deserves a second look:

Workload placement

Each provider has areas where its pricing model is genuinely better. Worth knowing when you decide where a new workload lands:

Tooling

Platforms worth evaluating

Build or buy

A sensible order of work

Most teams have one dominant provider and a smaller footprint elsewhere. Fix the dominant one first: that is where the money is, and the habits transfer. On the AWS side, our AWS cost audit surfaces the quick wins first, at a fixed price.