At the beginning of January, AWS quietly raised the prices of EC2 Capacity Blocks for machine learning. This represents an increase of approximately 15 percent for GPU-based instances used for heavy ML training.
It is noteworthy that AWS implemented the change over the weekend without a separate announcement to customers.
Specifically, the p5e.48xlarge and p5en.48xlarge instances have become more expensive. According to The Register, these instances, equipped with multiple NVIDIA H200 GPUs, are used for large-scale machine learning and AI workloads. In most regions, the hourly price of the p5e.48xlarge rose from $34.61 to $39.80. For the p5en.48xlarge, the rate went up from $36.18 to $41.61 per hour. In the US West Northern California region, prices are even higher, with an increase to nearly $50 per hour for the p5e variant.
Capacity Blocks are intended for customers who need guaranteed GPU capacity at a predetermined time. Unlike spot instances, which can suddenly disappear, organizations use them to reserve capacity for a fixed time window, ranging from one day to several weeks. This service is mainly used by companies that perform business-critical ML training, and for which interruptions can have major financial consequences. The target group, therefore, consists mainly of large organizations with substantial cloud budgets.
According to Amazon, the price adjustment is due to changes in supply and demand ratios. The rates for Capacity Blocks would be adjusted quarterly based on expected market conditions. AWS emphasizes that this dynamic has long been part of the pricing model for this specific service.
No clear pricing policy
The increase comes a few months after AWS announced significant price reductions for GPU instances, particularly for On-Demand and Savings Plans. Capacity Blocks were not included in those reductions at the time. This underscores that different purchasing models within AWS are priced and adjusted independently of each other.
The change may affect organizations with Enterprise Discount Programs. These contracts typically offer discounts on public prices, but when the base price increases, the absolute amount that customers pay also increases. This may prompt renewed discussions between large customers and AWS about pricing agreements.
The price increase is occurring amid a global shortage of advanced GPUs. Demand for computing power for AI and machine learning continues to grow, while supply is limited. Competitors such as Microsoft Azure and Google Cloud are actively positioning themselves in this market, but are also facing limited availability.