Do you also feel that spot instance prices have been surging recently? If so, you’re not alone. We have conducted an in-depth analysis of spot instance pricing over the past 12 months, pulling data from every public cloud region for AWS, GCP, and Azure. Our findings provide a fascinating look into the current dynamics of spot instance pricing.
We based our research on records from all public cloud regions for AWS, GCP, and Azure. CAST AI keeps a constant inventory of all spot instance prices in all cloud regions using the cloud provider API for retrieving inventory pricing. We focused on the changes over the past year, and the differences we noticed are staggering.
Let’s review the findings and identify what is really happening to spot instance pricing.
A comparative analysis of AWS, Azure, and GCP
When looking across the three main clouds, we find that on aggregate, worldwide, the following is true:
Azure: 108.01%
AWS: 21.09%
GCP: -25.95%
Azure spot VM prices have led the way with a 108% increase from 2022 to 2023.
With a rise of 21%, AWS comes second. GCP bucked the trend with a nearly 26% decrease in spot VM prices.
Let’s dive deeper into each cloud to check where these hikes come from and consider ways to mitigate their impact on your spend.
Spot instance pricing hikes in Azure
Delving deeper into Azure results, let’s see where these price increases come from on a regional basis.
We broke down the top 10 regions that are experiencing the most significant spot instance pricing increases:
germanywestcentral: 150.01%
australiaeast: 131.83%
eastus2: 131.14%
southeastasia: 130.36%
westeurope: 130.13%
southcentralus: 128.92%
westus2: 128.70%
northcentralus: 128.22%
uksouth: 127.58%
northeurope: 127.14%
Many of these are smaller regions. The spot instance pricing hikes may result from cloud users escaping large regions, hoping to find lower rates in smaller regions. This influx may create more demand and drag prices up.
The regions with the largest percentage increases may still be a good bet. However, the one region that stands out is germanywestcentral – the third most expensive region in Azure globally for spot instances.
The instance types that drive these changes the most are:
Standard_Dpds: 1011.15%
Standard_Dplds: 1009.28%
Standard_Dpls: 1005.91%
Standard_Dps: 961.96%
Standard_Eps: 950.78%
Standard_Epds: 857.69%
Standard_M-s: 542.48%
Standard_NVas: 443.86%
Standard_NV: 413.03%
Standard_NVs: 340.29%
If you were running one of these instance types without changing your strategy over the past year, you likely saw significant increases in your spend. These changes were due just to spot prices drifting over time.
Let’s consider another interesting example by looking at the D-family, which showed massive differences in spot instance pricing increases:
Standard_Dads_v5: 31.36%
Standard_Das_v5: 32.94%
Standard_Das_v4: 53.35%
Standard_Da_v4: 64.07%
Standard_Ds_v5: 73.21%
Standard_Dd_v5: 75.54%
Standard_Dds_v5: 77.11%
Standard_Dds_v4: 78.61%
Standard_Ds_v4: 80.93%
Standard_Ds_v3: 86.25%
Standard_Dd_v4: 91.47%
Standard_Dps_v5: 961.96%
Standard_Dpls_v5: 1005.91%
Standard_Dplds_v5: 1009.28%
Standard_Dpds_v5: 1011.15%
The variation between the lowest and highest price hike was 980%!
And while sometimes you may need specific features, even the difference between very similar Das_v5 and Das_v4 was 20%.
They are the same style of machine, only a different generation. If you configured your node pool for Das_v4 last year, you’re likely paying 20% more than you need today.
By allowing more flexibility in your spot instance selection, you could swap high-priced instances for much better options. We’ll tell you how to pull it off in a moment, but first, let’s consider the other two cloud providers.
Spot instance pricing changes in AWS
In AWS, the ten regions that have recorded the highest spot instance price increases include:
af-south-1: 24.98%
ap-southeast-4: 24.81%
ap-northeast-3: 23.19%
us-gov-west-1: 22.74%
eu-south-1: 22.60%
ap-east-1: 22.46%
eu-north-1: 22.39%
ap-northeast-2: 22.26%
ca-central-1: 22.03%
me-south-1: 21.80%
There’s also an interesting pattern when it comes to the instance families affected by the most significant price changes in AWS.
The highest increases occurred for the instance types that are very old or have specific high-performance requirements, such as onboard NVMe like the I3 and I3en:
p4d: 180.23%
trn1: 143.51%
f1: 127.84%
t1: 100.91%
x2iezn: 100.20%
i3en: 73.85%
r3: 65.70%
i3: 65.54%
r4: 59.83%
trn1n: 57.95%
For comparison, let’s look at a commonly used instance type – the C family. These instances are largely interchangeable, except for the Graviton machines.
Here is how the price increases varied in just this one family, again showing the benefits of being flexible in machine types:
c6a: -11.82% (decrease)
c1: -6.71% (decrease)
c6g: -3.15% (decrease)
c7g: -2.07% (decrease)
cc2: 0.00% (no change)
c6gd: 0.02%
c5a: 3.24%
c6id: 5.37%
c3: 8.41%
c5ad: 10.91%
c5: 12.47%
c6gn: 15.95%
c6i: 17.41%
c5d: 17.87%
c4: 20.35%
c5n: 20.61%
c6in: 28.44%
c7gn: 49.58%
If you were locked in on using c5’s in 2022, this year you are paying on average 24% more than if you had dynamically moved to c6a’s as the spot prices drifted up.
In a moment, we’ll tell you how to make that happen.
GCP: the only cloud to see spot instance pricing drops
Google Cloud presents an interesting anomaly, as it didn’t have a single region that increased spot instance prices in aggregate. On the contrary, GCP has reduced spot instance pricing across all regions compared to a year ago.
As the only provider to show a net decrease in spot instance pricing, GCP is definitely an excellent choice for ephemeral workloads.
Spot instance pricing changes in GCP were indeed interesting, as the regions with the highest percentage increase in price were still… NEGATIVE!
Compared to 2022, GCP reduced spot instance pricing across all regions:
northamerica-northeast2: -20.24%
australia-southeast2: -20.34%
europe-west6: -23.73%
me-central1: -23.79%
asia-southeast2: -24.45%
southamerica-west1: -24.52%
europe-west1: -24.73%
us-south1: -24.77%
us-east4: -24.77%
asia-east1: -25.14%
But instead of getting lost in percentages, let’s consider one more interesting example.
Launched in March 2023, the C3 type is a younger sister of the C2 series.
Comparing their cost per compute core, you get the following:
- c2-standard: $.023/core
- c3-standard: $.016/core
If you set up your node pools to run on c2-standard spot instances in July 2022, you pay 30% more per CPU core for less performant machines.
When GCP launched the c3 instance types earlier this year, CAST AI automatically started provisioning these highly performant, low-cost spot instances. Users didn’t have to enroll for these new instance types, they became available to them by default.
One customer saved an additional 12.5% month-over-month from February to March purely due to CAST AI using the c3 instance type where applicable.
Automated provisioning is the answer
All the examples above prove that the landscape of spot instance pricing is ever-changing. Being aware of these dynamics can help you make more informed decisions about your cloud strategy.
By being flexible and using automatic provisioning tools, you can navigate these changes and save significantly on your cloud spend.
CAST AI’s feature called scheduled rebalancing watches for these price changes. It automatically swaps out your spot instances as they increase in price on the market for more competitive instances.
As a result, you benefit from lower prices, but also much better performance and match to your workload’s needs.
Final thoughts on spot instance pricing changes
Our analysis revealed some intriguing trends in the pricing strategies of major cloud providers. AWS showed a mix of regions and instance types with high price increases, while GCP demonstrated a trend towards price reductions in many areas.
It’s important to note that these findings are based on averages, and individual instance types within the groups may vary. Therefore, when deciding which cloud provider, region, or instance type to choose, it’s crucial to consider the specific requirements of your use case.
This is just a snapshot of the kind of insights from an in-depth analysis of cloud pricing data. With more detailed data, we could delve into more granular insights, such as the impact of reserved instances, spot instances, or variations in pricing across different operating systems.
In future analyses, we could also consider the performance characteristics of different instance types, as the cost-performance ratio is often more important than the raw cost.
Stay tuned for future updates as we explore the fascinating world of cloud pricing trends!
