Cost optimization is a top priority for organizations in 2023, and cloud spend is one of the best opportunities for reducing costs. Vantage is a cloud cost management and optimization platform with a unique dataset of thousands of AWS accounts and how they are allocating resources. This Q4 2022 Cloud Cost Report follows up on our Q3 Report and uses anonymized, real-world data to quantify how cloud spending is shifting and changing. To discuss this report in more detail, join our growing Slack Community of over 1,000 FinOps and cloud professionals. This report analyzes interesting trends to provide engineering leaders, FinOps professionals, and CFOs with emerging and evolving patterns.
Organizations have been aggressive in purchasing commitments and it is paying off. On-demand spend as a percentage of all compute spend came in at just 31.36% to end Q4, down significantly from the 50.19% where it stood 6 months ago at the beginning of Q3. AWS themselves confirmed this trend in their Q3 earnings call:
"We did see a deceleration or a drop in op margin sequentially quarter over quarter. The broad disclaimer on AWS margins is that they will fluctuate over time as we balance investments versus renegotiating pricing with the long-term customer commitments."
- Brian Olsavsky, CFO at Amazon
How much lower can the share of on-demand spend go? We recommend teams aim for 95% coverage for their infrastructure so theoretically teams could save even more than they are now with committed use discounts.
TweetIn Q3 we reported that over 80% of S3 costs were coming from the Standard Storage tier. In Q4 we saw a dramatic shift to lower tiers with Standard Storage dropping to 64% of S3 costs as customers turned on S3 Intelligent Tiering and moved objects into the archival storage tiers. For this quarter we found 16% of S3 costs came from tiers which are managed by Intelligent Tiering. Glacier Flexible Retrieval also saw a spike in usage in our data from last quarter.
Overall, storage is the #3 cost category on AWS, behind compute and data transfer, and EBS and S3 rank #3 and #4 respectively on the Cloud Cost Leaderboard. Investing in cost optimization efforts in this area can have a major impact on overall AWS bills.
S3 Storage Class | Share of Costs | S3 Intelligent Tiering |
---|---|---|
Standard Storage | 64.07% | |
Standard Infrequent Access | 11.24% | |
Standard Frequent Access | 10.35% | Yes |
Glacier Flexible Retrieval | 6.28% | |
Archive Instant Access | 3.77% | Yes |
Infrequent Access | 1.89% | Yes |
Glacier Deep Archive | 0.95% | |
One Zone Infrequent Access | 0.91% | |
Archive Access | 0.02% | Yes |
Other | 0.52% |
AMD's c6a instances are gaining ground while Graviton instances continue their steady march. Note that the most expensive Graviton instance is only $2,354 per month while the priciest AMD instance reaches $14,394 per month. That means that this chart does not tell the full story of Graviton adoption (see Lambda below).
All this still leaves Intel with 75% of cloud compute spend to end the quarter. While we expect their share of costs to continue dropping, the new r6in instances with 3rd generation Xeon Scalable chips are sure to help them retain workloads.
TweetAmong the other services which offer RIs we see variations in what customers are willing to commit to. RDS averages around 65% on-demand while Redshift sees more than 50% of its costs covered by reservations.
ElastiCache and OpenSearch are sandwiched in the middle when it comes to buying committed use discounts.
TweetThis chart represents the daily percentage higher or lower than average for compute spend in Q4. Among Vantage customers, the single highest day for on-demand spend was the Thursday following the Black Friday/Cyber Monday weekend, reaching a peak of 23% higher daily costs than average. As the holidays kicked in, spend dropped dramatically, down nearly -30% by Christmas Day.
The 50% swing in spending between the highest and lowest days of December illustrates in sharp relief the advantages of the cloud for seasonal workloads.
TweetOur customers did better this quarter at upgrading to newer instances which offer superior price to performance ratios, with the c6 variety crossing a 10% share of costs. But the story remains the dominance of m5s and r5s for memory optimized workloads.
M5 and R5 instances, despite having later geneneration counterparts, still offer extremely good price-to-performance ratios which is why organizations continue to leverage them en masse.
TweetIn our Q3 report we examined how GP3 adoption was growing even as older GP2 EBS volumes still drove most costs. On RDS, GP3 SSDs are only driving 2% of storage costs while Provisioned IOPS (or IO1 SSDs) started the quarter at 50% of costs.
However, throughout Q4 we observed a marked shift as GP2 SSDs hit 60% of costs for RDS storage. AWS also launched support for GP3 Volumes on RDS in November 2022 and we began to see Vantage customers adopt those (although they cost the same per GiB as GP2 for RDS).
TweetThis quarter saw a 300% increase in usage of the newest g5 GPUs which contain NVIDIA's latest A10 cards. Older, p3 instances were the biggest loser from Q3 to now. p4 instances, which the Stable Diffusion generative model was trained on, did not see dramatically higher usage but they are only available in one expensive size so the greater optionality of the g5 series may be more appealing to customers.
TweetIn a blog post earlier this year we measured how Lambda adoption of Graviton was growing very rapidly. For this report we can report that Lambda costs were consistently above 50% on Graviton, a marked victory for customers looking for savings on Lambda.
Could Graviton adoption on Lambda grow even more? We think so. All programming languages in Lambda are now supported by Graviton. The last movers will be those with custom containers who may need to recompile or switch libraries to shift to Graviton.
Tweet