November 20, 2026

Cloud Pricing Trends, October–November 2026

Summary: Cloud World Model's October and November 2026 pricing snapshots found no provider list-price changes among 33 tracked rates on AWS, GCP, Azure, OCI and DigitalOcean. Four records changed because we corrected them: two Azure SQL rates in October and two GCP Compute Engine rates in November. The GCP e2-medium record fell 39%, but that's a fix to our own arithmetic, not a GCP price cut. We modeled the same load balancer, two app servers and MySQL database as in September at 100 and 300 requests per second. Modeled monthly cost ranged from about $199 to about $738, depending on provider mapping and traffic.

We check cloud list prices every month across AWS, GCP, Azure, OCI and DigitalOcean. This write-up covers two snapshots, October 1 and November 1, 2026. We also reran September's architecture through the simulator on three providers, so the snapshots show up as modeled cost per hour, not just a list of rates.

Why we track pricing history

A simulation is only as good as its inputs. We check rates against official provider pricing sources and save each month as a dated snapshot. That way the history is a record, not something we overwrite. When we find a mistake in an earlier record, we log the fix as a correction instead of rewriting the past. It covers five providers, four categories (compute, database, storage, network) and the same 33 resources every month, so months can be compared directly. The full series is on Pricing History and in the public GET /api/pricing-history endpoint.

What changed in October and November 2026

We found no provider list-price changes among the 33 tracked rates in either month. Four recorded values changed because we corrected them. The November rates were rechecked against provider pricing sources on November 18, 2026.

ResourceProviderPrevious record ($/hr)Corrected record ($/hr)ChangeSnapshotType
SQL Database Basic (5 DTUs), East USAzure$0.0068$0.00670833 ($0.161/day ÷ 24)−1.35%October 2026Record correction
SQL Database Standard S2 (50 DTUs), East USAzure$0.1008$0.10083333 ($2.42/day ÷ 24)+0.03%October 2026Record correction
e2-medium (2 vCPU, 4 GB), us-central1GCP$0.0553$0.03350571−39.41%November 2026Record correction
n2-standard-4 (4 vCPU, 16 GB), us-central1GCP$0.194$0.194236+0.12%November 2026Record correction

What each correction is:

  • GCP e2-medium is a record correction, not a GCP price cut. Our earlier record priced e2-medium with dedicated-core arithmetic. e2-medium is a shared-core machine type, and Google publishes a bundled hourly price for it: $0.03350571 in us-central1. The November record now uses that published price. Google didn't lower the price. Our old record overstated it by about $15.91 a month per instance. If you budgeted e2-medium from our earlier history, use the corrected figure for all months.
  • GCP n2-standard-4 was stored rounded to $0.194. The record now uses the full published rate, $0.194236.
  • Azure SQL Database Basic and Standard S2. Azure publishes these as daily prices ($0.161 a day and $2.42 a day). We previously stored approximate hourly conversions. The October records now divide the published daily price by 24.

Corrections don't count as price changes. They're labeled "Record correction" on Pricing History and marked changeKind: "correction" in the API, and the per-provider change counts leave them out.

Two more things to know about these snapshots:

  • Eight rows are marked unverified. AWS CloudFront, Azure Blob Storage, and OCI Container Instances, Autonomous Database, Object Storage, Block Volume, File Storage and Load Balancer. Each one is a blended or usage-based estimate. The provider prices it by quantity (stored GB, transfer, bandwidth or configuration), so there's no single official hourly price to check it against. We keep the rows so the series stays comparable, but treat them as rough estimates.
  • The October snapshot carries 20 rates forward. For 20 AWS, GCP, OCI and DigitalOcean rates, we couldn't get official evidence of the rate in effect on or before October 1, so the October record keeps the September value. We rechecked all 20 on November 18, 2026. Eighteen match September. The other two are the GCP corrections above.

Earlier changes still apply. GCP Cloud SQL db-n1-standard-2 has stayed at $0.1351/hr since its June 2026 drop (from $0.1917/hr in May). The DigitalOcean Managed Database corrections from September stand.

What this means for budgets

October and November introduced no provider list-price changes to carry into your budget. If you budget from our history, three corrections are worth a look:

  • GCP e2-medium: use $0.03350571/hr, about $24.46 a month per instance, instead of the old $0.0553/hr record (about $40.37).
  • GCP n2-standard-4: about $0.17 a month more per instance than the rounded record.
  • Azure SQL Basic and Standard S2: a few cents a month either way.

What stable rates look like on a real architecture

We reran September's three-tier architecture: a load balancer in front of two app servers and one MySQL database. On AWS it's the same setup we use on our accuracy benchmark, in us-east-2. On GCP and DigitalOcean we used the same closest classes as in September:

RoleAWS (benchmark)GCPDigitalOcean
Regionus-east-2us-central1nyc3
Load balancerApplication Load BalancerCloud Load Balancing (1 forwarding rule)Load Balancer
App servers (×2)m5.large (2 vCPU, 8 GiB)n2-standard-2 (2 vCPU, 8 GB)g-2vcpu-8gb General Purpose (2 vCPU, 8 GiB)
MySQLRDS db.r5.large (2 vCPU, 16 GiB)Cloud SQL db-n1-highmem-2 (2 vCPU, 13 GB), Enterprise edition, zonalManaged MySQL db-s-2vcpu-4gb (2 vCPU, 4 GB)

The DigitalOcean database is still a smaller class than the other two, because it's the largest DigitalOcean MySQL size the simulator currently models.

We ran steady traffic at 100 and 300 requests per second (RPS), with no injected failures:

ProviderTrafficModeled cost/hrModeled cost/month (730 h)Fixed infrastructure/hrLoad balancer usage/hrModeled egress/hrEgress share of total
AWS100 RPSabout $0.59about $432about $0.45about $0.016about $0.12about 21%
AWS300 RPSabout $0.89about $651about $0.45about $0.047about $0.39about 44%
GCP100 RPSabout $0.61about $445about $0.40about $0.016about $0.20about 33%
GCP300 RPSabout $1.01about $738about $0.40about $0.048about $0.57about 56%
DigitalOcean100 RPSabout $0.27about $199about $0.27—$0 (within included transfer)0%
DigitalOcean300 RPSabout $0.27about $199about $0.27—$0 (within included transfer)0%

Fixed infrastructure is the app servers, the database and the load balancer's hourly fee. The GCP database line is the Cloud SQL instance price only; the simulator doesn't yet model Cloud SQL storage, so the GCP total leaves storage out.

Why these numbers differ from September

None of the differences come from provider list prices. They come from a newer simulator engine (1.2.80, up from 1.2.15), which models transfer and load balancers in more detail:

  • Egress is now based on bytes. The simulator assumes a 5,000-byte response (plus a 1,024-byte request) per request and applies each provider's tiered internet transfer rates and included monthly allowance. In September it applied a flat cost per request and deducted no allowances.
  • Load-balancer usage is now billed. AWS ALB capacity units and GCP load-balancer data processing are now separate charges on top of the hourly fee. In September only the hourly fee was modeled.
  • DigitalOcean's included Droplet transfer is now credited. Each g-2vcpu-8gb Droplet includes 4,000 GiB of outbound transfer a month, pooled to 8,000 GiB for two. At 300 RPS the model sends about 3,671 GiB a month, so there's no overage. In September this allowance wasn't credited.
  • Catalog precision. n2-standard-2 now uses Google's published us-central1 rate, $0.097118/hr, instead of a $0.10 estimate. The Cloud SQL db-n1-highmem-2 instance now uses a verified Cloud Billing price, $0.1761/hr, instead of a $0.16928 estimate. The g-2vcpu-8gb Droplet is now $0.0863/hr, from $0.0865. AWS m5.large and db.r5.large are unchanged at $0.096 and $0.24, and now come from exact us-east-2 price list entries.

Net effect versus September: AWS is about $3 a month lower at 100 RPS and about $8 higher at 300 RPS. GCP is about $52 higher at 100 RPS and about $135 higher at 300 RPS, mostly from egress. DigitalOcean is about $13 lower at 100 RPS and about $40 lower at 300 RPS, because egress is now covered by included transfer.

How egress is modeled

  • What the simulator reports: one hourly "egress and data transfer" cost for the client response path, plus any cross-zone transfer. Cross-zone transfer was $0 in all three setups.
  • What it charges for: modeled served requests. That's 100 and 300 RPS on GCP and DigitalOcean. On AWS, egress is charged on the requests the calibrated model serves, which is slightly below offered traffic.
  • Assumed sizes: 5,000 bytes per successful response and 1,024 bytes per request. Both are assumptions, not measurements.
  • Modeled monthly transfer: about 1,224 GiB at 100 RPS and about 3,671 GiB at 300 RPS on GCP and DigitalOcean. On AWS it's about 1,088 GiB and about 3,261 GiB, because fewer requests are served.
  • Rates applied:
    • AWS: $0.09/GiB to internet after a 100 GiB monthly allowance (us-east-1 reference tariff).
    • GCP: Premium Tier us-central1 to North America, after the first 1 GiB a month, $0.12/GiB up to 1 TiB and $0.11/GiB from 1 TiB to 10 TiB.
    • DigitalOcean: $0.01/GiB only beyond the pooled Droplet allowance.
  • Effective egress cost per million served requests: about $0.39 at 100 RPS and $0.41 at 300 RPS on AWS, about $0.55 and $0.53 on GCP, and $0 on DigitalOcean. We divided hourly modeled egress cost by hourly modeled served requests, then multiplied by one million. The figures shift with traffic because allowances and tiers apply per month.
  • Allowances assume nothing else uses them. The model assumes no other workload in the account has used the included allowance. If yours is already used up, your egress will be higher.
  • Your egress will differ. Real egress depends on your response sizes and where your clients are. Treat this line as a placeholder to replace with your own numbers.

What we'd take from it

These points apply to this architecture at these traffic levels:

  • Traffic, not list price, drives the change. Tripling traffic from 100 to 300 RPS raises the modeled cost by about 51% on AWS (about $219 a month) and 66% on GCP (about $293 a month). About 90% of each increase is modeled egress, and the rest is load-balancer usage. On DigitalOcean the modeled cost doesn't move, because the transfer fits inside the included allowance.
  • This period's corrections don't touch this setup. e2-medium, n2-standard-4 and the two Azure SQL rates aren't part of this architecture, and the AWS rates it uses didn't change.
  • Modeling assumptions move the total more than prices did. Between September and now, no list price in this setup changed, yet the modeled GCP cost at 300 RPS rose by about $135 a month because egress is now modeled by bytes and tiers. Check the assumed response size against your own traffic before deploying or changing infrastructure.
  • The database class is a smaller lever here. Swapping the GCP database for the db-n1-standard-2 we track ($0.1351/hr), with less memory (7.5 GB vs 13 GB), brings the estimate to about $0.57/hr at 100 RPS and about $0.97/hr at 300 RPS. These two figures come from snapshot rates, not a separate simulation.

This isn't a ranking. The setups differ in database memory, egress allowances and evidence quality, so the gap between providers comes mostly from the mapping, not from which cloud is better.

Method and limits

  • Only AWS is based on our own measurements. The AWS setup matches our company-owned benchmark topology. At both 100 and 300 RPS, the simulator reported that the owned benchmark calibration applied. That calibration covers performance outputs such as app CPU and in-VPC load-balancer latency, not cost. For the GCP and DigitalOcean setups in this post, performance comes from documentation-based reference estimates. Our measured GCP results cover a different topology; see the accuracy post. Because this is a pricing post, we've left latency out. See the live accuracy page for methodology and scope.
  • Costs are modeled from public list prices and simulator catalog rates. They are not measured bills. They're on-demand with no discounts. Transfer is a reference estimate based on assumed request and response sizes. Cloud SQL storage and backups aren't included.
  • Not every rate is in our tracked 33. The n2-standard-2, db-n1-highmem-2 and g-2vcpu-8gb prices come from the simulator's own catalog.
  • Nothing was deployed. Simulations don't create resources in anyone's cloud account.
  • Snapshot rates: the October 2026 and November 2026 snapshots (dated October 1 and November 1, 2026; rates rechecked November 18, 2026). The GCP database swap is computed from snapshot rates, not simulated.
  • Run details: simulator engine 1.2.80, seed 20240601, run on November 20, 2026. Results from other engine versions may differ. The September figures came from engine 1.2.15.

Run the same comparison yourself

Agents can simulate this comparison through the Cloud World Model MCP server or the REST API. The AWS setup is unchanged from September. Here it is exactly as we created it with the MCP simulation.create tool:

{
  "name": "pricing-example-aws",
  "appWeight": "lean",
  "seed": 20240601,
  "traffic": 100,
  "minInstances": 2,
  "maxInstances": 2,
  "resources": [
    {"id": "alb", "type": "network", "name": "Application Load Balancer (internal)", "provider": "aws",
     "location": {"regionKey": "us-east-2"},
     "characteristics": {"serviceFamily": "alb", "loadBalancerScheme": "internal"}},
    {"id": "app-1", "type": "compute", "name": "app-1", "provider": "aws",
     "location": {"regionKey": "us-east-2"},
     "characteristics": {"size": "m5.large", "serviceFamily": "ec2", "autoscaling": false, "workload": "crud"}},
    {"id": "app-2", "type": "compute", "name": "app-2", "provider": "aws",
     "location": {"regionKey": "us-east-2"},
     "characteristics": {"size": "m5.large", "serviceFamily": "ec2", "autoscaling": false, "workload": "crud"}},
    {"id": "db", "type": "database", "name": "mysql-db", "provider": "aws",
     "location": {"regionKey": "us-east-2"},
     "characteristics": {"size": "db.r5.large", "serviceFamily": "rds", "workloadDatabaseEngine": "mysql"}}
  ],
  "connections": [
    {"sourceId": "alb", "targetId": "app-1"},
    {"sourceId": "alb", "targetId": "app-2"},
    {"sourceId": "app-1", "targetId": "db"},
    {"sourceId": "app-2", "targetId": "db"}
  ]
}

Then call:

  1. simulation.step
  2. simulation.cost_breakdown, which gives the 100 RPS cost
  3. simulation.inject_traffic with {"targetRps": 300}
  4. simulation.step
  5. simulation.cost_breakdown, which gives the 300 RPS cost
  6. simulation.delete when you're done

For GCP and DigitalOcean, use the same graph with these changes. Drop workload, workloadDatabaseEngine and loadBalancerScheme from both. Remove serviceFamily from the two app servers and keep "autoscaling": false.

  • GCP: set provider to gcp and regionKey to us-central1. Use load balancer serviceFamily cloud-load-balancing, apps n2-standard-2, and database db-n1-highmem-2 with serviceFamily cloud-sql. On the database, also add "cloudSql": {"engine": "mysql", "edition": "enterprise", "availability": "zonal", "diskType": "ssd", "allocatedGiB": 10}. Without it, the current engine can't price the Cloud SQL instance and leaves it out of the total.
  • DigitalOcean: set provider to digitalocean and regionKey to nyc3. Use load balancer serviceFamily load-balancer, apps g-2vcpu-8gb, and database db-s-2vcpu-4gb with serviceFamily managed-mysql.

Change the regions, instance classes or traffic to match your own setup before deploying or changing infrastructure. Start with the MCP quickstart. Price history is available at GET /api/pricing-history and GET /api/pricing-history/compare?from=2026-09-01&to=2026-11-01.

Previous: Cloud Pricing Trends, September 2026