TL;DR
AWS is raising EC2 Capacity Blocks for ML prices by roughly 20% in July 2026, after a roughly 15% increase in January. The change is limited to one purchasing option, but it shows how higher memory and AI hardware costs can reach cloud customers through instance pricing rather than a clear memory surcharge.
Amazon Web Services is raising prices for EC2 Capacity Blocks for ML by roughly 20% starting in July 2026, after an earlier increase of about 15% in January, according to Business Insider and ITPro. The move matters because it shows how the AI-driven memory squeeze is reaching cloud customers through GPU reservation prices, even when invoices do not label the increase as a memory cost.
The confirmed AWS change applies to Capacity Blocks for machine learning, a reservation model used by organizations that need booked GPU capacity for training or tuning AI models. AWS told Business Insider that these reservation prices are updated based on supply and demand, and said the change applies to one purchasing option rather than all AWS compute pricing.
The pressure is not limited to AWS. OVHcloud chief executive Octave Klaba forecast cloud price rises of 5% to 10% between April and September 2026, citing higher RAM and NVMe costs, according to TechRadar. Hetzner has also raised some cloud and server prices, with Tom’s Hardware reporting increases of up to 37% from April 1, 2026.
The hidden part of the bill is the pass-through path. Memory makers raise DRAM prices; server vendors absorb higher component costs; cloud providers buy that equipment; customers then see increases in instance families, regions, reservation products, storage tiers or managed services. The exact memory share is rarely separated on invoices, which makes the cost harder for customers to audit.
Cloud’s hidden memory bill
Thought the cloud lets you dodge the squeeze — you rent the RAM, you don’t buy it? You’re still paying for every gigabyte. You’ve just stopped being able to see the bill.
No escape from the shortage anywhere — on-prem servers also cost +15–25%. But providers hedge scarce hardware better than you can, and you can’t buy half a cluster for two weeks.
8×H200 ≈ $15–20/hr owned (3-yr amortized) vs $39.80 rented — roughly half. 83% of CIOs plan to repatriate some workloads. Hybrid is the new default.
The cloud doesn’t make the memory tax disappear — it launders it, turning a violent fab shortage into a few innocuous percentage points scattered across a bill you can’t easily audit. “I’m in the cloud, I’m safe” is the most expensive misconception in this series. Refuse to pay for idle RAM, sort each workload to its cheapest venue, and lock pricing before the Q2–Q3 adjustment. The escape hatch was never cloud-vs-on-prem — it’s discipline-vs-drift. Next: the local-inference rig.
Cloud Budgets Feel Hardware Pressure
For cloud customers, the AWS move weakens the assumption that renting capacity shields them from hardware price shocks. Companies still pay for RAM, storage and accelerators, but those costs arrive through bundled cloud pricing rather than a line item marked memory.
The impact is likely to be uneven. AI training, memory-heavy analytics, Redis-style caching, in-memory databases and large managed data services are more exposed than light compute workloads. Customers with steady, high-use workloads may compare cloud pricing against owned hardware, while spiky or short-term demand can still favor cloud capacity because providers can pool scarce equipment across many users.
AWS EC2 GPU reservation
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
AI Demand Tightens DRAM Supply
The pricing pressure follows a broader memory supply crunch tied to AI data centers. Samsung, SK hynix and Micron dominate much of the global DRAM market, while high-bandwidth memory used with AI accelerators has absorbed more factory output and capital.
Samsung and SK hynix have warned that memory shortages could last into 2027, according to Tom’s Hardware. Barron’s also reported KeyBanc analyst John Vinh’s view that added industry capacity is not expected to arrive in a meaningful way until 2027, leaving data center demand to keep pressure on prices through 2026.
“Amazon EC2 Capacity Blocks for ML reservation prices are updated periodically based on supply and demand.”
— Amazon Web Services, quoted by Business Insider
machine learning cloud instance
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Provider Pricing Remains Patchy
It is not yet clear whether Microsoft Azure, Google Cloud or other large providers will announce comparable changes across similar AI or memory-heavy services. AWS has confirmed a change to EC2 Capacity Blocks for ML, but a broad industry-wide price schedule has not been established.
It also remains unclear how much of each increase comes from DRAM versus GPUs, storage, energy, networking, financing or regional capacity constraints. Customers may see a higher cloud bill without a clean way to isolate the memory component.
high memory cloud server
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
July Bills Set First Test
The next test comes when July 2026 charges begin showing up for customers using AWS Capacity Blocks. Finance and infrastructure teams will be watching renewals, reservation terms, region pricing and managed-service rates to see whether the increases stay contained or spread.
Cloud buyers are also likely to review whether each workload belongs on reserved cloud capacity, on-demand instances, private infrastructure or a hybrid setup. The main near-term milestone is whether more providers publish Q3 2026 price changes tied to memory and AI hardware costs.
GPU cloud computing reservation
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Key Questions
Did AWS raise all cloud prices?
No. The confirmed change concerns EC2 Capacity Blocks for ML, a GPU reservation product. AWS says other AI workload options remain available.
Why are memory costs showing up in cloud bills?
Cloud providers buy servers that contain DRAM, storage and GPUs. When component costs rise, providers can pass those costs through instance prices, reservation rates or managed-service fees.
Which cloud workloads are most exposed?
The most exposed workloads are likely AI training, high-memory databases, large caches, analytics jobs and GPU-heavy reservations. Light compute workloads may see less direct pressure.
Does moving on-premise avoid the memory squeeze?
Not fully. Owning hardware can help for steady high-use workloads, but buyers still face higher server memory and storage costs. Cloud can still make sense for short, uneven or uncertain demand.
What should customers watch next?
Customers should watch July AWS invoices, contract renewals, cloud provider notices and Q3 pricing updates. The open question is whether this remains a narrow AI reservation issue or spreads further.
Source: Thorsten Meyer AI
