The rise in memory prices can no longer be explained solely as another semiconductor shortage cycle. The expansion of artificial intelligence is changing where manufacturers place their production capacity and which products take priority. HBM, server memory, and enterprise storage are competing for factories that cannot ramp up production as quickly as demand grows, and that pressure ultimately affects RAM, SSDs, and, finally, the cost of deploying infrastructure.
The key points of the new server costs in 30 seconds
- The demand for AI is pushing manufacturing capacity toward HBM and higher-margin server products.
- TrendForce forecasts that conventional DRAM will increase another 13-18% quarterly in the third quarter of 2026.
- The situation might persist into 2027 because new factories will take time to deliver significant volume.
- Always buying the latest generation is no longer automatically the most cost-effective choice.
- Reducing, redesigning, and reusing hardware can become a strategy for infrastructure, not just savings.
This has particular implications for system administrators and infrastructure managers: for years, it was assumed that technological upgrades would allow for more memory and storage at lower costs.
That curve has stopped behaving as expected.
TrendForce believes that the supply of DRAM will remain tight through 2027. The analyst points directly to the capacity dedicated to High Bandwidth Memory (HBM), the growth of AI servers, and the increased amount of memory per server. It also estimates that the supply of RDIMM bits might grow only between 15% and 20% year-over-year, while demand from new server platforms continues to rise.
The industry is thus entering a paradoxical situation: never before has so much money been invested in computing capacity, yet some basic server components are becoming increasingly expensive.
AI doesn’t just require GPUs: it’s also absorbing memory
When discussing the cost of AI infrastructure, GPUs usually get most of the attention.
But a GPU doesn’t work alone.
Large accelerators need enormous amounts of high-bandwidth memory. The servers supporting them also include processors, conventional DRAM, NVMe storage, high-speed networking, and power and cooling systems.
The explosion of HBM has an especially important characteristic: producing it is more demanding in manufacturing capacity than manufacturing conventional DRAM. TrendForce notes that its production requires a significantly larger number of wafers, so increasing HBM capacity indirectly reduces available capacity for other segments.
Manufacturers also have economic incentives to prioritize higher-value products.
The result isn’t just that “AI buys all the RAM.” The industrial reality is more complex: manufacturers are reorganizing processes, investments, and capacity around server products, HBM, and high-performance applications.
And other markets are experiencing secondary effects.
| Component | What’s happening | Impact on infrastructure |
|---|---|---|
| HBM | Strong growth driven by AI accelerators | Absorbing more DRAM capacity |
| Server DRAM | High demand and tight supply | More expensive servers with lots of RAM |
| DDR4 | Lower production priority | Older modules may not decrease in price |
| DDR5 | Pressure from servers and new platforms | Upgrading generations costs more |
| Enterprise SSD | Growing with data centers and AI | Tightened storage supply |
| Client NAND | Less constrained in the medium term | May loosen before DRAM |
There’s also an important difference between DRAM and NAND. TrendForce expects that NAND Flash capacity could see a more relaxed supply situation in the second half of 2027, thanks to new facilities and process migrations.
The scenario for DRAM is more complicated.
The newest server is not always the cheapest
This situation forces us to revisit a question that for a long time seemed to have an obvious answer.
Should we always use the latest available hardware generation?
From an absolute performance standpoint, probably yes. Economically, it depends much more on the workload.
A modern platform can offer better per-core performance, wider memory bandwidth, faster PCIe, and higher energy efficiency. But an application that mainly needs 512 GB of RAM and relatively little CPU can run perfectly fine on an older generation.
That detail significantly shifts the cost calculations.
Think of database servers, large caches, certain virtualization nodes, storage, legacy enterprise applications, or platforms that keep large datasets in memory.
For these, it’s less about squeezing the last percentage of performance and more about having enough RAM.
The current market can create seemingly contradictory situations: keeping DDR4 servers for more years may make more economic sense than immediately migrating to DDR5.
Even older platforms can still be useful for secondary loads, labs, storage, or services where capacity outweighs performance per watt.
That doesn’t automatically make old hardware a good buy. Consider factors like power consumption, spare parts availability, manufacturer support, performance, rack density, and operational risk.
But simply retiring a server because a new generation has been released doesn’t always make sense either.
The real cost appears when RAM is missing
There’s another risky temptation when memory becomes more expensive: sizing servers too close to their limits.
This can end up costing quite a lot more.
The latency difference between accessing DRAM and storage is measured in orders of magnitude. A modern NVMe SSD is extraordinarily fast compared to traditional disks, but still much slower than main memory.
When a database, virtual machine, or application starts constantly relying on swap, the issue is no longer about the cost of a RAM module.
It becomes a performance problem.
Therefore, reducing memory doesn’t just mean installing fewer DIMMs.
The right question is how much memory each workload truly needs and why.
Many infrastructures have virtual machines over-provisioned for years. A VM got 32 GB because it seemed reasonable in 2021, but often uses only nine GB today.
Multiplying that by hundreds of machines costs money.
Virtualization techniques like ballooning and dynamic allocation can help, but only if application behavior is understood. Containers can also reduce some overhead associated with running full operating systems for each service.
Databases require a different approach. Aggressively reducing their buffers can immediately shift pressure to storage and worsen latency.
Optimizing RAM means measuring first, then trimming appropriately.
Reduce, redesign, and reuse: the new 3Rs of infrastructure
The current landscape allows us to revisit the well-known three Rs—Reduce, Redesign, Reuse—and apply them, with some nuance, to data centers.
Reduce involves identifying where memory is being wasted. Rightsizing VMs, setting appropriate container limits, reviewing caches, compressing, and removing unnecessary services can free significant amounts of RAM.
Redesign involves questioning architecture. Maintaining ten servers at low utilization may be more expensive than consolidating workloads onto fewer, higher-capacity nodes.
Consolidation has limits. High availability, failure domains, maintenance, and disaster recovery require redundancy. A single massive server can’t replace a cluster outright.
However, many platforms remain where CPU and memory are underutilized simply because the original design was never revisited.
The third R is Reutilize.
Extending the lifespan of a server generation makes perfect sense when it continues to meet performance, security, and support requirements.
And this is probably one of the less discussed consequences of the memory shortage.
For years, the tech industry has associated modernization with replacement. The current pressure on DRAM calls for a distinction between the two.
Modernization can also mean better virtualization, workload consolidation, software updates, automation, or migrating only the applications that truly benefit from new hardware.
Hosting prices also absorb the increase
Infrastructure providers are not immune to this market dynamic.
A dedicated server, a private cloud platform, or a virtualization cluster all require CPU, RAM, SSD, network cards, and replacement components. When the acquisition cost of these increases, maintaining the same prices indefinitely becomes challenging.
This impact is especially noticeable in configurations with 384, 512, or 768 GB of RAM, which are common for virtualization platforms and databases.
Additionally, there’s the issue of replacement costs.
Providers can operate servers bought three years ago at reasonable prices, but when expanding or replacing them, current prices apply. The hardware’s historical cost no longer reflects the true cost of providing the service in subsequent years.
This situation can promote models where the lifecycle of well-functioning enterprise equipment is extended.
Such an approach isn’t unusual; it’s a rational response to a changing silicon market.
Memory is beginning to influence architecture decisions
The situation may still evolve. Manufacturers are expanding capacity, and new facilities scheduled for 2027 should gradually increase production, though TrendForce warns that a meaningful contribution might be delayed until 2028.
Moreover, DRAM and NAND are likely to diverge further. It’s not correct to assume that RAM and SSD prices will rise in tandem at the same pace indefinitely.
But the message for infrastructure designers is already clear.
For a long time, servers were sized starting from CPU and storage. Memory was a relatively cheap variable that could be scaled easily.
AI is challenging that assumption.
Perhaps in upcoming upgrade cycles, the first question shouldn’t be which processor to buy, but much more basic: how much memory does the application actually need, and what will it cost to maintain it over the next five years.
Frequently Asked Questions
Why are RAM prices rising so much?
The demand for AI servers is increasing the use of HBM and server memory, as manufacturers allocate more capacity to these products. TrendForce forecasts that the supply of DRAM will remain tight through 2027.
Is AI also making SSDs more expensive?
The expansion of data centers is elevating demand for enterprise SSDs, although the NAND market follows a different dynamic from DRAM. TrendForce expects capacity growth to alleviate NAND supply constraints during the second half of 2027.
Does it make sense to keep using DDR4 servers?
Yes, if their performance, power consumption, support, and reliability still meet your workload requirements. For applications where having large memory is more important than using the latest CPU, extending their lifespan can significantly reduce costs.
How can companies reduce server costs?
Start by measuring actual usage of RAM, CPU, and storage. Adjust VMs, consolidate workloads, use containers appropriately, and extend the life of still-supportable hardware to avoid purchasing capacity at the highest cycle costs.
