AI computing power demand continues to be hot! Amazon.com, Inc. (AMZN.US) Cloud computing business AWS once again raised GPU capacity reservation service prices.

Amazon.com, Inc. (AMZN.US) announced that its cloud computing business AWS will increase the prices of machine learning-specific EC2 Capacity Blocks reservation services starting from July 1, involving multiple cloud computing instances with NVIDIA Corporation (NVDA.US) GPUs. This is the second price adjustment for related products by AWS since January of this year, reflecting the persistent strong demand for AI computing power. AWS stated that the price adjustment mainly involves several instance types including P6-B300, P6-B200, P5, P5e, P5en, and P4de. The P6-B300 and P6-B200 instances are equipped with NVIDIA Corporation Blackwell B300 and B200 GPUs respectively, the P5 series uses H100 and H200 GPUs, and the P4 series is equipped with A100 GPUs. According to the latest prices, the hourly price per GPU for P6-B300 instances has been increased to $14.04, and for P6-B200 to $12.355; P5 instances have had their GPU hourly prices raised to $5.191 in the U.S. region, and $4.72 in non-U.S. regions; P5e and P5en instance prices have also been increased, and P4de instance prices have been raised to $2.214 per GPU per hour. AWS stated that apart from the above products, the prices of other services will remain unchanged. The company pointed out that the prices for EC2 Capacity Blocks machine learning reservation services will be adjusted periodically based on market supply and demand. Analysts believe that this price adjustment once again reflects the fact that AI computing resources are still in short supply. In recent years, with the rapid development of generative AI, large language models, and AI agents, the demand for GPU computing resources from enterprises continues to rise. NVIDIA Corporation's H100, H200, and the latest Blackwell series GPUs have been in short supply, which has also led cloud service providers to continuously adjust their computing leasing prices. The prices announced by AWS also show that the P6e UltraServer, which consists of 72 Blackwell GB200 GPUs, has an hourly rental cost of up to $761.9 in the Dallas local region in the U.S., equivalent to approximately $10.58 per GPU per hour. In addition, the P5 instance with 8 H100 GPUs has an hourly rental cost of approximately $34.61 in the main U.S. region; the P5en instance with 8 H200 GPUs has an hourly rental cost of approximately $45.77; and the P4 series instance with 8 A100 GPUs has an hourly price ranging from $11.8 to $14.77. In addition to NVIDIA Corporation GPUs, AWS also announced the prices of its self-developed AI chip Trainium series instances. The Trn1 instance with Trainium chips has an accelerator price of approximately $0.596 per hour, while the new generation Trn2 instance has a corresponding price of approximately $2.235. Market observers believe that AWS's consecutive price hikes for AI computing products reflect both the strong demand for AI infrastructure and the continued benefits that global cloud computing companies are receiving from the tight supply of GPU resources. As Microsoft Corporation (MSFT.US), Meta (META.US), Amazon.com, Inc., and Alphabet Inc. Class C (GOOGL.US) and other tech giants continue to expand their investments in AI infrastructure, the market expects high-performance GPU cloud service prices to remain high in the foreseeable future, with AI computing leasing services expected to continue to be an important growth driver for cloud computing companies.