State of the art of Serverless - Q1 2026 - Cost estimation #3

Go back to list

To ground the cost comparison in a realistic use case, we modeled a video transcoding pipeline processing 2 million requests per month. Each transcoding job is assumed to run for 45 seconds, a representative duration for re-encoding a short-form video clip (typically 2 to 5 minutes) into a single target format and resolution. The workload is configured with 8 GB of memory, reflecting the substantial RAM footprint required to hold intermediate decoded frames in memory during the transcoding process.

Below, the baseline used for estimate costs for running such workloads

Provider
AWS
Azure
Google
Oracle
T Cloud
Scaleway
Service Lambda x86 Lambda arm Container Apps * Cloud Run Functions Functions FunctionGraph✦ Serverless✦
Region eu-central-1 West Europe europe-west3 FRA eu-de fr-par
Size 8GB 4CPU 8GB 2CPU 8GB 8GB 8GB 3CPU 8GB
Calculation
Requests $0.20 x 2 $0.56 x 2 $0.40 x 2 $0.2 x 2 $0.19 x 2 /
CPU / 0.000034 x 4 x 45 x 2000000 ✧ 0.0000336 x 2 x 45 x 2000000 / / 0.000011422 x 3 x 45 x 2000000
Memory 0.0000167 x 8 x 45 x 2000000 0.0000133 x 8 x 45 x 2000000 0.000004 x 8 x 45 x 2000000 0.0000035 x 8 x 45 x 2000000 0.00001417 x 8 x 45 x 2000000 0.000016667 x 8 x 45 x 2000000 0.0000011422 x 8 x 45 x 2000000
Estimation
Requests $0.40 $1.12 $0.80 $0.40 $0.38 /
CPU / $12,240 $6,048 / / $3,084
Memory $12,024 $9,576 $2,880 $2,520 $10,202 $12,000 $822
Total $12,024 $9,576 $15,121 $8,568 $10,202 $12,000 $3,906
  • *: Each container app manage 1 request maximum
  • ✧: Instances have a ratio of 1CPU:2GB, we selected 0.5CPU/1GB
  • ✦: From Euro during March 2026

Source