RAM prices have multiplied: what it means for S/4HANA landscapes

What is happening to RAM prices

Every quarter, the market research company TrendForce publishes how contract prices are developing – the prices that large buyers negotiate with the memory manufacturers. Since the fourth quarter of 2025, these figures have been pointing steeply upwards:

Compounded, this means: based on these figures, the contract price for conventional DRAM in mid-2026 is three to four times the level of autumn 2025 (our own calculation). Since then, it has been rising more slowly, but it is still rising.

Users are already feeling the effect. The hosting provider Hetzner, for example, raised its prices on 1 April 2026, by up to 32% for cloud servers, citing sharply increased purchase prices for memory and NVMe SSDs.

Why AI is driving prices

Memory manufacturers are shifting their production to products for AI data centres: high bandwidth memory (HBM) for AI accelerators and memory for servers. According to TrendForce, the large US cloud providers have been securing volumes through early orders and long-term supply contracts since the end of 2025. That leaves less for everyone else, at higher prices.

There is no quick end in sight. TrendForce expects server DRAM to remain tight in the fourth quarter of 2026 as well, and sees the manufacturers continuing to prioritise their most advanced production for server products. For enterprise SSDs, TrendForce does not expect significantly more capacity until the end of 2027 or 2028. As long as the build-out of AI data centres continues, you should not plan for a return to the old prices.

Why S/4HANA is particularly affected

SAP HANA is an in-memory database: the data resides in main memory. How much RAM a system needs therefore depends directly on how much data it contains. In the past, the history was stored on comparatively inexpensive hard disks. Today, it occupies exactly the component whose price is currently multiplying.

Then there is the landscape. Alongside production, there is usually a QA system, a test or development system and a sandbox system, often more for projects or training. If they are filled from production by client or system copy, each one gets the full history and needs almost as much main memory as production.

Simplified example. Production holds 4 TB of data in main memory. QA, test and sandbox are full copies. Together, that is around 16 TB – three quarters of it in systems in which nobody does productive work.

Whether the systems run in your own data centre, with a hosting provider or in the cloud: at the latest with the next hardware renewal, memory expansion or contract renewal, the new market prices will feed through, either directly via the hardware or via the providers’ prices. Costs that hardly anyone had on their radar until now become an issue for budgeting and procurement.

Approach 1: shrink production

The obvious lever is production itself. Less data there means less main memory there – and in every full copy.

• SAP archiving: Completed business transactions are moved from the database to archive files and remain readable when needed.
• Data aging and Native Storage Extension: Rarely used data moves from main memory to cheaper disk storage while remaining in the database.
• Cleaning up technical tables: Logs, IDocs, workflow and application logs often grow unnoticed and can be cleaned up regularly.

This has a lasting effect, but it has its limits. Archiving is a project in its own right, with coordination with the business departments, retention periods and questions of access. And whatever is needed for ongoing operations, reporting or legal obligations stays in production.

Approach 2: size non-production systems by purpose

Where production cannot shrink any further, the bigger lever lies with the other systems. A QA or test system is often cheaper per gigabyte than production, for example because it manages without high availability. In total, however, it still costs a lot of money: in the example above, three quarters of the total main memory sits in these systems.

For tests, releases and support, the last few months are usually what matters. Yet years or even decades are copied, because the standard client copy has no concept of a time frame.

Instead, the Data Refresh Operator copies a time slice, for example the last 90 days, and automatically adds all dependent objects from earlier years until it is verifiable that nothing is missing. The target systems are fully functional from a business perspective and only as large as necessary. Production remains unchanged.

–75%

HANA RAM in non-production

–60%

Storage & backup

Guideline values from reference scenarios. The actual effects depend on system size, history and refresh cycles.

In the example above, that would mean: instead of 12 TB for QA, test and sandbox, around 3 TB would be enough. The landscape would then need 7 TB instead of 16 TB of main memory. And because smaller copies run faster, the systems can be refreshed more often. More on HANA memory in non-production systems

What you can do now

• Record the main memory of all systems: production and every copy of it
• Check the dates: when are hardware renewals, memory expansions or contract renewals due?
• Assess the potential for archiving and clean-up in production
• Define for each non-production system which time frame is really needed for testing
• Switch refreshes to time slices before the next expansion is due

We are happy to work out with you what this means for your landscape: Book a demo.

Sources

• TrendForce, 26 February 2026: Price Rally Drives 4Q25 DRAM Revenue Up 29.4%
• TrendForce, 5 January 2026: Memory Makers Prioritize Server Applications, Driving Across-the-Board Price Increases in 1Q26
• TrendForce, 31 March 2026: AI Server Demand to Drive Memory Contract Price Increases in 2Q26
• TrendForce, 9 July 2026: Server DRAM Contract Prices Expected to Rise 13–18% QoQ in 3Q26
• TrendForce, 30 September 2026: AI Server Demand Sustains Memory Contract Price Increases in 4Q26
• WinFuture, 23 February 2026: Up to 32 per cent more: Hetzner Online to raise hosting prices shortly (in German)