Direct answer: at Azure Container Instances public list prices on 10 November 2026 (eastus, USD, excluding tax and discounts), Linux standard container groups bill per second: vCPU $0.0405/hour and memory $0.00445/GB-hour; Spot $0.01215 / $0.001335 (about 70% off); Confidential $0.04455 / $0.004895; Windows software adds $0.000012/vCPU-second ($0.0432/vCPU-hour); GPUs bill K80 $0.01/100 seconds, P100 $0.0244/100 seconds, V100 $0.0793/100 seconds. Billing is at the container group: vCPU requests round up to a whole number, memory to 0.1 GB, with a 1 vCPU/1 GB minimum, so a 1/1 group running 730 hours costs $32.81. Examples: 1/1 for 5 minutes/day (2.5h/month) $0.11; 10 vCPU/20 GB for 1 hour/day (30h/month) $14.82; the same Spot $4.45; 4/8 Windows 30h $11.11; 4/16 always-on 730h $170.24, Confidential $187.26. Against Container Apps Consumption: the same 10/20 30h is $27.00 after the free grant; 1/1 always-on and continuously active is $65.56; but a 2.5h tiny job is $0 on CCA vs $0.11 on ACI—short batches favor ACI, spiky HTTP favors CCA.
This guide is for architects, platform/SRE, finance and procurement choosing among ACI, Container Apps and AKS. It separates official price facts, editorial calculations and exclusions, covering per-second billing, container groups and rounding, minimum bills, Spot/Windows/Confidential/GPU, a same-workload comparison with Container Apps, hidden costs and a 90-day checklist. Prices come from the ACI pricing page, Azure Retail Prices API and Microsoft Learn (10 November 2026). This is not procurement advice.
(1) Billing is per container group, not per container: all containers share the group's requested vCPU/memory, and sidecars count. (2) The minimum is 1 vCPU/1 GB: a 0.5/0.5 request bills as 1/1, $32.81/month if always on; vCPU rounds up to a whole number, memory to 0.1 GB. (3) The clock starts at image pull: large images and restarts bill too; restart policy Always keeps restarting and billing, so batch jobs should use Never/OnFailure and stop the group.
Meters: per-second vCPU, memory and premiums
ACI bills the container group's vCPU and memory requests per second; Linux standard, Spot, Confidential, Windows and GPU have separate prices. Spot is about 30% of standard (vCPU $0.01215, memory $0.001335) and suits retryable/interruptible work; Confidential is about 10% above standard; Windows adds $0.0432/vCPU-hour; GPUs bill vGPU per 100 seconds plus vCPU and memory. ACI is also eligible for Azure savings plans (1/3-year), with the actual discount depending on commitment and region.
| Meter | Price | Equivalent/note |
|---|---|---|
| Linux standard vCPU | $0.0405/hour | Per second, rounded up to whole vCPU |
| Linux standard memory | $0.00445/GB-hour | Per second, rounded up to 0.1 GB |
| Spot vCPU | $0.01215/hour | About 30% of standard |
| Spot memory | $0.001335/GB-hour | About 30% of standard |
| Confidential vCPU | $0.04455/hour | About 10% above standard |
| Confidential memory | $0.004895/GB-hour | About 10% above standard |
| Windows software | $0.000012/vCPU-second | $0.0432/vCPU-hour |
| GPU K80 | $0.01/100 seconds | $0.36/hour plus vCPU/memory |
| GPU P100 | $0.0244/100 seconds | $0.8784/hour plus vCPU/memory |
| GPU V100 | $0.0793/100 seconds | $2.8548/hour plus vCPU/memory |
Sources: ACI pricing page and Retail API (eastus, 10 November 2026). GPUs are available only in select regions; Spot can be interrupted and its SLA follows the official terms; Windows applies to Windows container groups; excludes storage, logs, egress and tax.
Rounding and minimum bill: 0.5 vCPU bills as 1
ACI rounds vCPU requests up to a whole number and memory up to 0.1 GB, with a 1 vCPU/1 GB minimum per group. A 0.5 vCPU/0.5 GB group therefore bills as 1/1 at $0.04495/hour, or $32.81 for 730 hours; 1.2 vCPU/2.05 GB bills as 2 vCPU/2.1 GB at $0.090345/hour, or $65.95 for 730 hours. Packing small containers into one group shares resources, but the group's total request is still the billing basis—don't set fractional requests expecting a smaller bill.
| Request | Billed | Hourly | 730 hours |
|---|---|---|---|
| 0.5 vCPU / 0.5 GB | 1 / 1 GB | $0.04495 | $32.81 |
| 1.2 vCPU / 2.05 GB | 2 / 2.1 GB | $0.090345 | $65.95 |
| 4 vCPU / 8 GB | 4 / 8 GB | $0.1976 | $144.25 |
| 10 vCPU / 20 GB | 10 / 20 GB | $0.494 | $360.62 |
Editorial calculation: vCPU rounded up to a whole number, memory to 0.1 GB; Linux standard PAYG, 730 hours; excludes Windows/GPU/Spot, tax and discounts. The 1 vCPU/1 GB minimum comes from the pricing page.
Scenario TCO: short batch, long batch, always-on
Per-second billing favors short jobs: 1 vCPU/1 GB for 5 minutes/day (2.5h/month) is $0.11; 10 vCPU/20 GB for 1 hour/day (30h/month) is $14.82; the same Spot is $4.45; 4/8 Windows 30h is $11.11 (compute $5.93 + Windows software $5.18). Always-on 1/1 for 730h is $32.81, 4/16 $170.24, 10/20 $360.62. The point of ACI for batch is "run and stop, don't pay for idle"; leave a group running and per-second quietly becomes a monthly fixed fee.
| Scenario | Hours/month | Monthly | Note |
|---|---|---|---|
| 1/1, 5 minutes/day | 2.5h | $0.11 | 1/1 minimum |
| 10/20, 1 hour/day | 30h | $14.82 | Standard Linux |
| 10/20 Spot, 1 hour/day | 30h | $4.45 | Interruptible |
| 4/8 Windows, 1 hour/day | 30h | $11.11 | Includes Windows software |
| 1/1 always-on | 730h | $32.81 | Minimum always-on cost |
| 10/20 always-on | 730h | $360.62 | High-utilization service |
Editorial calculation: Linux standard PAYG; the Windows row includes $0.0432/vCPU-hour × 4 vCPU × 30h; Spot uses $0.01215/$0.001335; excludes storage, logs, egress, tax and discounts.
Comparison with Container Apps: free grant and scale-to-zero change the answer
Container Apps Consumption has higher active rates (vCPU $0.000024/second = $0.0864/hour, memory $0.000003/GiB-second = $0.0108/GiB-hour) but a monthly free grant of 180,000 vCPU-seconds, 360,000 GiB-seconds and 2 million requests per subscription, plus scale-to-zero. Same workloads: a 2.5h 1/1 job is ACI $0.11 vs CCA $0 (within the grant); 10/20 30h is ACI $14.82 vs CCA $27.00; 1/1 always-on and continuously active is ACI $32.81 vs CCA $65.56; 1/1 for 60h plus 1M requests is ACI $2.70 (start/stop) or $32.81 (always-on) vs CCA $1.26. Bottom line: short batch and sustained high utilization favor ACI; spiky HTTP/events favor CCA; if an always-on app is mostly idle, CCA's idle rate ($0.0108/vCPU-hour) can beat ACI.
| Same workload | ACI | CCA Consumption | Lower |
|---|---|---|---|
| 1/1 × 2.5h | $0.11 | $0 (free grant) | CCA |
| 10/20 × 30h | $14.82 | $27.00 | ACI |
| 1/1 × 730h continuously active | $32.81 | $65.56 | ACI |
| 1/1 × 60h + 1M requests | $2.70 (start/stop) | $1.26 | CCA |
Editorial calculation: CCA active $0.000024/vCPU-second, $0.000003/GiB-second, $0.40/million requests, less the 180,000 vCPU-second/360,000 GiB-second/2M request free grant; excludes CCA private-endpoint/environment fees ($0.10/hour). ACI excludes storage, logs, egress and tax.
Spot, Confidential and GPU
Spot is about 30% of standard: 10/20 for 30h drops from $14.82 to $4.45, suiting retryable batch, rendering and CI; interruptions need retry logic. Confidential 4/16 always-on is $187.26 vs standard $170.24, about 10% more, for hardware-level isolation in select regions/SKUs. GPU pricing for "1 GPU + 4 vCPU + 16 GB" for one hour: K80 $0.55, P100 $1.07, V100 $3.05; GPUs are available only in select regions, so check region and quota before deployment.
| Configuration | Duration | Monthly/hourly | Note |
|---|---|---|---|
| Spot 10/20 | 30h | $4.45 | 30% of $14.82 standard |
| Confidential 4/16 | 730h | $187.26 | Standard $170.24 |
| GPU K80 + 4/16 | 1h | $0.55 | Plus vCPU/memory |
| GPU P100 + 4/16 | 1h | $1.07 | Plus vCPU/memory |
| GPU V100 + 4/16 | 1h | $3.05 | Plus vCPU/memory |
Editorial calculation: Spot $0.01215/$0.001335; Confidential $0.04455/$0.004895; GPU 100-second prices converted to one hour plus 4 vCPU/16 GB; excludes storage, logs, egress, tax and discounts. GPU/Confidential availability follows the official docs.
Selection matrix and hidden costs
Match the workload: one-off/short batch/scheduled ETL uses ACI, Spot where retryable; spiky HTTP/events needing scale-to-zero use CCA; sustained high-utilization services can use ACI or AKS (scale/reservations); Windows containers use ACI; GPU/confidential workloads compare ACI and CCA serverless GPU by region availability. Seven hidden costs: image-pull time (billing starts at pull), the group's total request (sidecars and idle capacity count), the 1/1 minimum and rounding, restart=Always (continuous restart billing), Spot interruption retries, Windows software, and logs/ACR/Azure Files/egress. 90 days: weeks 1–2 inventory job duration, image size, group requests and restart policy; weeks 2–4 run a 1/1 or real config for 7 days; weeks 4–6 compare ACI/CCA including the free grant and idle; weeks 6–8 move retryable batches to Spot and shrink images; weeks 8–10 test cold start and interruption retries; weeks 10–12 reconcile seconds, rounding and premiums against the bill.
| Workload | First choice | Check first |
|---|---|---|
| One-off/short batch | ACI (Spot optional) | Image pull, group request |
| Scheduled ETL/CI | ACI | Restart policy, stop the group |
| Spiky HTTP/events | CCA Consumption | Free grant, scale-to-zero |
| Sustained high utilization | ACI or AKS | Unit price vs reservations/ops |
| Windows containers | ACI | Windows software premium |
| GPU/confidential | ACI GPU/Confidential or CCA GPU | Region, quota, SLA |
Editorial summary, not a vendor recommendation; availability, SLA, quotas and regions follow the official docs.
FAQ
How is Azure Container Instances billed?
By the container group's vCPU and memory requests, per second: Linux standard vCPU $0.0405/hour, memory $0.00445/GB-hour; Spot $0.01215/$0.001335; Confidential $0.04455/$0.004895; Windows software $0.0432/vCPU-hour; GPUs use 100-second K80/P100/V100 prices plus vCPU/memory.
What is the minimum bill?
Each group has a 1 vCPU/1 GB minimum, with vCPU rounded up to a whole number and memory to 0.1 GB. A 1/1 group costs $0.04495/hour, or $32.81 for 730 hours; a 0.5/0.5 request still bills as 1/1.
Am I charged after stopping the group?
A stopped group no longer bills compute; duration runs from the start of the first image pull (or restart) until the group stops. Logs, ACR, Azure Files and egress continue to bill through their services. Delete the group to avoid leftovers.
How much does Spot save?
Spot vCPU is $0.01215 and memory $0.001335, about 30% of standard—roughly 70% off. 10/20 for 30h falls from $14.82 to $4.45, but Spot can be interrupted and suits retryable work; SLA follows the official terms.
ACI or Container Apps?
Short batch/one-off and sustained high utilization favor ACI; spiky HTTP/events that can scale to zero favor CCA. CCA includes 180,000 vCPU-seconds, 360,000 GiB-seconds and 2M requests free per month, so a 2.5h tiny job can be free; for 10/20 30h, ACI at $14.82 beats CCA at $27.00.
How are GPU containers billed?
GPU bills K80 $0.01, P100 $0.0244, V100 $0.0793 per 100 seconds for the vGPU, plus vCPU $0.000879/100 seconds and memory $0.000118/100 GB-seconds. One GPU + 4 vCPU/16 GB for one hour: K80 $0.55, P100 $1.07, V100 $3.05. Only select regions offer GPUs.
How much more do Windows containers cost?
On top of the Linux-equivalent vCPU/memory price, Windows adds $0.000012/vCPU-second ($0.0432/vCPU-hour). 4 vCPU/8 GB for 30h: compute $5.93 + Windows software $5.18 = $11.11; Windows applies to Windows container groups.
How do I avoid container-group billing traps?
Five things: (1) size the group request to actual need—sidecars count; (2) use Never/OnFailure for batch and stop/delete the group; (3) shrink images to cut pull billing; (4) don't use Always for one-off jobs; (5) start with a 1/1 group for 7 days, then reconcile seconds and rounding against the bill.
Research method and limitations
Official facts (pricing page, Retail API, Learn): Linux standard $0.0405/$0.00445, Spot $0.01215/$0.001335, Confidential $0.04455/$0.004895, Windows $0.000012/vCPU-second, GPU K80/P100/V100 100-second prices; container-group billing, vCPU rounding to a whole number, memory to 0.1 GB, 1/1 minimum; billing from image pull/restart to stop. Editorial calculations: 730 hours/month, 30 days, per-second conversion; CCA at $0.000024/vCPU-second, $0.000003/GiB-second and $0.40/million requests, less the free grant.
Seven limitations: (1) only eastus public list prices, excluding tax/CSP/enterprise discounts/actual savings-plan rates; (2) job duration, image size and request volume are editorial assumptions; (3) excludes ACR, Azure Files, logs, egress, private endpoints and tax; (4) no real cold-start, image-pull or Spot-interruption benchmark; (5) GPU/Confidential regional availability and quotas are not checked one by one; (6) the CCA comparison uses Consumption active rates only, without private-endpoint/environment fees or idle discounts; (7) prices and quotas change—recheck official pages and bills before purchase.
Sources and basis
- Microsoft Azure, Azure Container Instances Pricing: azure.microsoft.com/en-us/pricing/details/container-instances. Container-group/per-second billing, 1/1 minimum, Spot/Confidential/Windows/GPU, billing examples. Accessed 10 November 2026.
- Azure Retail Prices API (productName=Container Instances, eastus): prices.azure.com/api/retail/prices. $0.0405/$0.00445, Spot $0.01215/$0.001335, Confidential $0.04455/$0.004895, Windows $0.000012, GPU 100-second prices. Accessed 10 November 2026.
- Microsoft Learn, Container groups in Azure Container Instances: container-instances-container-groups. Shared lifecycle/resources/network, sidecars. Accessed 10 November 2026.
- Microsoft Learn, Resource availability & quota limits for ACI: container-instances-quotas. vCPU/memory limits, regional capacity. Accessed 10 November 2026.
- Microsoft Azure, Azure Container Apps Pricing: azure.microsoft.com/en-us/pricing/details/container-apps. Consumption active/idle, free grant, requests, scale-to-zero. Accessed 10 November 2026.
Data cut-off 10 November 2026; prices/availability follow Microsoft official pages and bills; this is not procurement, commercial or legal advice.