The demand for artificial intelligence computing power remains robust, as evidenced by Amazon.com (AMZN)'s cloud computing division, AWS, announcing another price increase for its machine-learning-focused EC2 Capacity Blocks reservation service, effective July 1st. This adjustment affects several cloud computing instances powered by NVIDIA (NVDA) GPUs and marks the second such price revision since January, underscoring the sustained strength of the AI compute market.
AWS's Latest Price Adjustments
The price changes apply to multiple instance types, including P6-B300, P6-B200, P5, P5e, P5en, and P4de. The P6-B300 and P6-B200 instances are equipped with NVIDIA's Blackwell B300 and B200 GPUs, respectively. The P5 series utilizes H100 and H200 GPUs, while the P4 series is powered by A100 GPUs.
Under the new pricing structure, the hourly rate per GPU for the P6-B300 instance rises to $14.04, and for the P6-B200 instance to $12.355. The P5 instance price increases to $5.191 per GPU hour in US regions and $4.72 in non-US regions. Prices for P5e and P5en instances have also been raised, with the P4de instance now priced at $2.214 per GPU per hour. AWS stated that prices for all other services remain unchanged and noted that EC2 Capacity Blocks pricing will be periodically adjusted based on market supply and demand dynamics.
Broader Market Context for AI Compute
Analysts view this pricing action as another indicator of an ongoing supply-demand imbalance for AI computing resources. The rapid development of generative AI, large language models, and AI agents has driven a sustained surge in corporate demand for GPU processing power. NVIDIA's H100, H200, and the latest Blackwell series GPUs have been in a state of prolonged scarcity, prompting cloud service providers to continually adjust their compute leasing rates.
Detailed Instance Pricing Examples
AWS's published pricing also reveals that the P6e UltraServer, configured with 72 Blackwell GB200 GPUs, carries a hefty hourly rental fee of $761.9 in the US Dallas local zone, equating to approximately $10.58 per GPU per hour. Furthermore, a P5 instance with 8 H100 GPUs costs about $34.61 per hour in key US regions, while a P5en instance with 8 H200 GPUs is priced around $45.77 per hour. Instances in the P4 series, equipped with 8 A100 GPUs, range from $11.8 to $14.77 per hour.
Pricing for AWS's Proprietary AI Chips
In addition to NVIDIA GPU-based instances, AWS disclosed pricing for its custom AI chip instances. The Trn1 instance, featuring its Trainium chip, is priced at about $0.596 per accelerator per hour, while the newer Trn2 instance carries a corresponding price of approximately $2.235.
Implications for the Cloud Computing Sector
Market observers believe that AWS's consecutive price hikes for AI compute products highlight both the continued robust demand for AI infrastructure and the ongoing benefits cloud providers are reaping from tight GPU supply. As major technology firms like Microsoft (MSFT), Meta Platforms (META), Amazon, and Alphabet (GOOGL) continue to expand their investments in AI infrastructure, the market anticipates that prices for high-performance GPU cloud services will remain elevated for the foreseeable future. The AI compute leasing business is poised to remain a significant growth driver for cloud computing providers.
Comments