AWS raises AI cloud compute prices again as memory shortage deepens
Analysts say impact could extend well beyond individual customers
Amazon has increased the price of renting some of its most sought-after AI cloud computing services.
Amazon Web Services (AWS), said customers using its EC2 Capacity Blocks for Machine Learning service will face hourly price increases of around 20% from July.
The service enables businesses to reserve access to high-performance graphics processing units (GPUs) in advance, an option widely used by companies training or fine-tuning large AI models.
The latest rise comes just six months after AWS increased prices for the same service by about 15%, meaning the cost of securing reserved AI computing capacity has climbed significantly since the start of the year.
"Amazon EC2 Capacity Blocks for ML reservation prices are updated periodically based on supply and demand," AWS said in a statement announcing the changes.
The company introduced Capacity Blocks in 2023 after demand for AI chips surged, making GPUs increasingly difficult to obtain through traditional on-demand cloud pricing. The reservation model was designed to guarantee customers access to scarce computing resources during the AI boom.
The latest increase applies only to reserved Capacity Blocks rather than AWS's broader portfolio of AI computing services. The company said pricing for other purchasing options would remain unchanged, while its Trainium AI chips are not affected by the latest adjustments.
Although the change targets a relatively small segment of AWS's cloud offerings, analysts say its impact could extend well beyond individual customers.
As the world's largest cloud infrastructure provider, AWS underpins thousands of software platforms and AI services used by businesses worldwide. Higher infrastructure costs could eventually feed through to AI applications, enterprise software and cloud-based services.
Hardware constraints
The rapid expansion of AI sector is increasingly running up against physical hardware constraints, rather than a lack of software capability. At the centre of the bottleneck is high-bandwidth memory (HBM), a specialised type of memory packaged alongside advanced AI processors.
Limited manufacturing capacity for HBM has restricted the number of GPUs that chipmakers can produce, slowing the pace at which cloud providers can expand their AI datacentres.
Peter Berezin, chief economist at BCA Research, said the shortage gives major cloud providers greater leverage over pricing.
Berezin added that while memory shortages have increased infrastructure costs, they have also kept demand for AI computing above available supply, allowing hyperscale cloud providers such as AWS, Microsoft, Google and Oracle to pass higher costs on to customers.
Price hikes across tech
The pressure is being felt across the wider technology industry.
Apple last week raised prices for several products, citing soaring memory component costs, while Microsoft has also increased Xbox console prices.
In a blog post, Microsoft said memory costs for consoles had risen by more than two-and-a-half times and were expected to climb further through 2027 despite efforts to absorb higher expenses.
"The race to build out AI data centres is resulting in a swift and significant increase in demand that chip makers are rushing to meet," said Danni Hewson, head of financial analysis at AJ Bell.
However, the shortages have proved highly profitable for memory manufacturers. Investors have driven companies such as Micron and SK Hynix to record valuations amid expectations that demand for high-bandwidth memory will continue to outstrip supply for several years.