Amazon has increased prices for several of its artificial intelligence cloud services, underscoring how persistent shortages of advanced memory chips are beginning to reshape the economics of AI infrastructure and raise costs across the technology industry.
The latest increase affects Amazon Web Services’ (AWS) EC2 Capacity Blocks for ML, a service that allows customers to reserve graphics processing unit (GPU) capacity in advance for AI training and inference workloads. Beginning in July, hourly prices for several server configurations will rise by approximately 20%, following an earlier **15% increase introduced in January.
The consecutive price hikes suggest that pressure on AI infrastructure costs is intensifying rather than easing, even as hyperscale cloud providers continue investing hundreds of billions of dollars in new data centers.
In announcing the changes, AWS said: “Amazon EC2 Capacity Blocks for ML reservation prices are updated periodically based on supply and demand.”
The move is enormous in weight because AWS is the world’s largest cloud computing provider, serving millions of developers, enterprises, startups, and government agencies. Many AI applications, enterprise software platforms, and consumer services depend on AWS infrastructure, meaning higher cloud costs could eventually be passed through to businesses and end users.
Unlike price increases for consumer electronics, such as smartphones or gaming consoles, higher cloud computing costs have the potential to ripple across the broader digital economy, increasing operating expenses for companies building AI products and potentially slowing adoption among smaller businesses with limited technology budgets.
While much attention has focused on breakthroughs in foundation models and software capabilities, the industry’s biggest bottlenecks are tied to physical infrastructure, particularly the supply of advanced semiconductors. At the center of those constraints is high-bandwidth memory (HBM), a specialized memory technology that sits alongside AI processors from companies such as Nvidia and AMD. HBM dramatically increases the speed at which GPUs can access data, making it essential for training and deploying today’s largest AI models.
However, production of HBM remains limited because manufacturing is highly complex and concentrated among a small number of suppliers, primarily Micron, SK Hynix, and Samsung Electronics. Demand from hyperscalers has far outpaced supply, driving up prices for AI servers and increasing the cost of expanding data center capacity.
Those supply constraints are now flowing directly into cloud pricing.
Industry-wide, similar trends are emerging. Apple recently increased prices on several products, citing sharply higher memory costs, while Microsoft raised Xbox prices. Elon Musk has also complained publicly about unprecedented increases in memory prices affecting AI infrastructure.
The implications extend well beyond individual hardware products. Cloud providers purchase thousands of AI servers equipped with advanced GPUs and HBM memory. As the cost of these systems rises, cloud operators face a choice between absorbing the higher expenses or passing them on to customers. Amazon’s latest pricing decision suggests that hyperscalers believe demand remains strong enough to support higher prices.
Peter Berezin, chief economist at BCA Research, argued that the industry’s biggest constraint is no longer software innovation but manufacturing capacity.
“As there is a limit to how much memory can be produced, then there is a limit to how many GPUs can be produced, which means that there’s a limit to how many data centers can be built,” Berezin wrote on X.
He added that the shortage gives large cloud providers unusual pricing power because customers have few alternatives when GPU capacity is scarce.
“While the memory shortage raises their costs, it also keeps the demand for compute above the available supply, which gives them greater pricing power over access to cloud computing,” Berezin said.
The pricing dynamics highlight how the AI boom is evolving. During the early phase of generative AI, competition centered on developing more capable foundation models. Increasingly, however, competitive advantage depends on securing access to scarce computing resources, advanced memory chips, and electricity needed to operate large-scale AI infrastructure.
Major cloud providers, including Amazon, Microsoft, Google, and Oracle, have collectively committed hundreds of billions of dollars to expanding AI data center capacity, yet supply continues to lag demand. That imbalance has enabled infrastructure providers to maintain premium pricing even as they incur higher capital expenditures.
The beneficiaries of this trend include memory manufacturers. Strong demand for HBM has propelled companies such as Micron and SK Hynix to record valuations, with investors betting that AI-driven demand will keep the memory market tight for years.
For enterprises building AI applications, however, the picture is more challenging. Higher cloud rental costs could increase the expense of training models, deploying AI agents, and running inference workloads, potentially slowing experimentation by startups while reinforcing the advantages enjoyed by larger technology companies with deeper financial resources.
Amazon’s latest increase, therefore, represents more than a routine pricing adjustment. Industry analysts see it as a sign of a growing reality in the AI economy: where the pace of innovation is becoming increasingly dependent on the availability of physical infrastructure, from advanced memory chips and GPUs to data centers and power, making compute capacity one of the industry’s most valuable and scarce resources.
