Cloud customers are beginning to feel the 2026 memory crunch through higher infrastructure pricing, with AWS EC2 Capacity Blocks for machine learning reported to have risen in January and again for July and OVHcloud forecasting broader increases. The development matters because companies that rent compute still pay for scarce DRAM and GPU-linked memory, even when the charge does not appear as a separate invoice line.

Public reporting has identified specific AWS price moves, not a blanket cloud-wide increase. ITPro reported in January that the p5e.48xlarge EC2 Capacity Block rate rose from $34.61 to $39.80 an hour in most regions, a move it described as roughly 15%. Business Insider later reported that AWS announced another roughly 20% increase for the same purchasing model starting in July 2026.

AWS has framed Capacity Blocks pricing as variable and tied to supply and demand, and Business Insider reported that the change applies to one purchasing option for reserving GPU capacity in advance. That distinction matters: there is no confirmed universal AWS, Azure or Google Cloud price rise in the reporting reviewed for this article, but the Capacity Blocks changes give cloud buyers a visible example of how scarce AI infrastructure can get more expensive.

OVHcloud has been more explicit about broader pressure. TechRadar reported that Octave Klaba of OVHcloud forecast 5% to 10% cloud price increases between April and September 2026 as RAM and NVMe disk prices rise, with server costs expected to climb 15% to 25%. The Thorsten Meyer AI report supplied for this article uses that chain to argue that a severe DRAM shock can reach users as a smaller, less obvious cloud-bill increase.

Cloud Buyers Lose Cost Cover

The practical effect is that renting infrastructure does not remove memory exposure. Cloud providers buy servers from the same OEM supply chain as enterprises, so higher server DRAM, SSD and accelerator costs can be absorbed for a time, repriced later or spread across instance families, regions and managed services.

That matters most for AI training, analytics, Redis and in-memory databases and memory-optimized instances such as AWS r-series, Azure E-series and Google Cloud highmem shapes. These workloads can carry a larger share of memory cost, while steady high-utilization workloads may force finance and engineering teams to compare reserved cloud capacity, committed-use discounts, colocation and owned servers more closely.

AI Demand Tightens Memory Supply

The price pressure traces back to DRAM and high-bandwidth memory demand from AI data-center buildouts. The Thorsten Meyer AI source material says Samsung, SK Hynix and Micron raised server DRAM prices by around 60% to 70% compared with late 2025; separate market reporting from Tom’s Hardware described server DRAM contract increases of up to 50% in late 2025.

The cost then moves through the hardware stack. Memory is a large part of server cost, so a jump in DRAM can become a 15% to 25% server-cost increase, according to OVHcloud’s forecast cited by TechRadar, before cloud providers decide how much to pass on. Analysts cited by Barron’s said meaningful new memory capacity is not expected until 2027, leaving supply tight through 2026.

“reservation prices are updated periodically based on supply and demand”

— AWS, in an announcement cited by Business Insider

Provider Pass-Through Remains Uneven

Several points are still unresolved. Microsoft and Google have not announced matching broad increases in the material reviewed, and AWS has tied the confirmed changes to EC2 Capacity Blocks for ML, a specific product with variable pricing. It is also unclear how much of any future cloud-bill rise will come from DRAM rather than GPUs, power, data-center capacity, networking equipment or regional demand.

The Thorsten Meyer AI report estimates that a 60% to 200% DRAM shock can surface as a 5% to 10% invoice increase after being diluted through server and cloud-provider costs. That is a model, not a disclosed provider formula. Actual impact will depend on contract terms, workload mix and region.

July Rates Test Buyer Budgets

July AWS Capacity Blocks pricing will be the next visible marker for buyers running large AI workloads, while OVHcloud’s April-to-September forecast keeps attention on mid-2026 cloud adjustments. Enterprises are likely to watch whether Azure, Google Cloud, Oracle and smaller providers make similar moves or hide changes inside discounts, regions and instance availability.

For customers, the near-term task is to map memory-heavy workloads to their cheapest reliable venue, reduce idle RAM, review committed spend before renewal and price out owned or hybrid capacity for steady high-utilization systems. The next hard signal will be provider pricing pages, renewal quotes and earnings commentary through the third quarter of 2026.

Key Questions

Is this a confirmed cloud-wide price increase?

No. The confirmed public reporting covers AWS EC2 Capacity Blocks for ML and OVHcloud’s forecast, not a universal rate rise across all providers. The broader issue is the risk of pass-through from memory and server costs.

Why can a RAM shortage affect cloud invoices?

Cloud providers buy physical servers that use DRAM, SSDs and accelerators. When those input costs rise, providers can absorb them, renegotiate purchasing or adjust instance prices, managed service fees, regional rates or discounts.

Which cloud workloads are most exposed?

GPU reservations, AI training and in-memory databases have the most direct exposure because they rely on scarce memory or accelerator capacity. Memory-optimized instances are also exposed, though the impact depends on contract terms and region.

Should companies move workloads out of the cloud?

Not automatically. Cloud still fits spiky or uncertain demand, and on-prem servers also face higher component costs. The stronger move is workload sorting: use cloud where elasticity pays, and compare owned or hybrid capacity for steady, high-utilization workloads.

Source: Thorsten Meyer AI