Dedicated Servers Turkey | GPU & Data Center | SistemDC

Why Server Prices Are Rising: AI and the Memory Crunch

Why Server Prices Are Rising: AI, the Memory Crunch and Global Hosting Price Hikes (2026)

Why Server Prices Are Rising: AI, the Memory Crunch and Global Hosting Price Hikes (2026)

For the past year, the same question has been circulating on server, VPS and cloud invoices: why are prices rising so fast? The answer is not one provider's decision. It lies in a global surge in memory and storage demand created by AI models. As chip makers shift their factories toward the memory that AI hardware needs, the RAM and NVMe drives used in ordinary servers have become scarce and expensive. In this article we explain how the chain works, how memory prices moved, what providers such as Hetzner and OVHcloud and giants such as AWS and Azure have done, and how you can protect your own infrastructure budget in this period.

The figures in this article come from press coverage, analyst reports and providers' own announcements. Sources do not always agree, so some percentages are marked as approximate or as reported. This article is for information only and is not investment or purchasing advice.

In short: what happened?

Since late 2022, generative AI models have spread quickly. The huge GPU clusters built for training and inference consume high bandwidth memory (HBM) as well as large amounts of server DRAM and enterprise SSDs. Memory manufacturers moved their limited factory capacity toward these more profitable products. From late 2025, supply of general-purpose DRAM and NAND tightened and contract prices climbed quarter after quarter. Server makers, and then hosting and cloud providers, began passing that cost on in their price lists.

The key point is that this is not one company's opportunism. It is a cost shock that starts at the very beginning of the supply chain. Let us walk through it step by step.

The start of the chain: why do AI models need so much memory?

Large language models and image models are hungry for two things: compute and fast access to memory. The HBM that sits next to a GPU offers far higher bandwidth than classic DRAM, but for the same number of bits it consumes noticeably more silicon area. When manufacturers dedicate part of their fabs to HBM, there is less capacity left for the DDR4 and DDR5 memory used in ordinary servers and computers.

Demand has also been shaped by enormous agreements. In October 2025 it was announced that OpenAI had signed letters of intent with Samsung and SK hynix for its Stargate project; according to some reports, the volumes involved could equal roughly 40 percent of global DRAM output. In December 2025 it was reported that Google, Amazon, Microsoft and Meta had placed open-ended memory orders. Purchases of this kind pushed smaller buyers, such as ordinary server makers and hosting companies, to the back of the queue. Around the same time Micron said it would leave its consumer brand Crucial, which many read as another sign of supply being reprioritized.

Memory prices in numbers: from where to where?

Memory prices are not published as a single public index. Contract prices are set between manufacturers and buyers, and analyst firms publish estimates. Based on TrendForce forecasts and other reports, the overall picture looks like this:

PeriodDevelopment (general DRAM contract prices)
Q4 2025Quarterly increase reported at roughly 45 to 50 percent; NAND also saw sharp rises.
Q1 2026Prices nearly doubled in one quarter; estimates ranged from 55 to 110 percent depending on the category.
Q2 2026Quarterly increase of roughly 80 percent.
Q3 2026Quarterly increase of roughly 40 percent.
Q4 2026 (forecast)About a 20 percent increase expected; the pace is slowing but the direction is still up.

Gartner said in October 2026 that memory prices had more than doubled over the previous 12 months and that this "memory inflation" is expected to last through 2027. Some reports say DRAM revenue in the third quarter of 2026 rose to roughly four to five times the level of a year earlier, while shipped bits grew only about 4 percent. In other words, almost all of the increase comes from price, not volume.

The picture is not one-directional, though. In April 2026 DDR5 module prices fell by roughly 30 percent, but they stayed far above normal levels. TrendForce attributes the slower pace in the last quarter to a higher comparison base, weaker end-market purchasing power and price ceilings in long-term agreements. The fairest summary is "increases continue, but momentum is fading."

For more detail, see TrendForce's Q4 2026 DRAM assessment and its January 2026 report on the server priority shift.

What did server makers and the cloud giants do?

Memory used to be a small line in a server's bill of materials; today it carries a much larger share. Depending on the source, that share ranges from about a quarter to about half. Dell's chief operating officer said he had not seen memory costs rise this fast before. Dell, Lenovo and HPE were reported to have raised server prices by roughly 10 to 20 percent between late 2025 and 2026, and some vendors shortened their quote validity periods.

The large cloud providers were affected too, though they have passed costs on in different ways:

  • AWS: Four separate updates were reported in 2026 for EC2 Capacity Blocks, the reserved GPU capacity used for machine learning: roughly 15 percent in January, roughly 10 percent for some configurations in April, roughly 20 percent in July and roughly 15 percent for most US configurations in October. AWS says these prices are updated regularly based on supply and demand.
  • Microsoft Azure: According to a notice reportedly sent to customers in October 2026, from 1 February 2027 some virtual machine and storage services will rise by roughly 8 to 17 percent depending on the region; figures of about 9 percent for West Europe and about 17 percent for North Europe have been shared. Some older VM series are expected to rise by about 25 percent worldwide. Microsoft points to changing regional market conditions and investment in new hardware.
  • Google Cloud: In the sources we reviewed we did not find an announcement of a general list-price increase on standard virtual machines driven by memory, although the view that the industry as a whole is moving in the same direction is widespread.

Hyperscale clouds have an advantage: long-term supply agreements and enormous purchasing volume. The fact that even they are moving their price lists shows how widespread the cost pressure is.

Hosting providers: Hetzner, OVHcloud and others

The hosting industry was hit directly, because its business model is built on buying hardware and renting it out monthly. The 2026 adjustments of two of Europe's best-known providers summarize the picture well.

Hetzner

In an adjustment announced in late February and effective 1 April, Hetzner raised cloud server prices in Germany and Finland by roughly 30 to 37 percent depending on the tier, with the US and Singapore reaching about 38 percent. Dedicated servers rose by roughly 3 to 21 percent depending on the model, and storage products by about 30 percent. The company pointed to steep price increases across the IT sector. A change to setup fees at the end of April explicitly referred to rising prices of hardware components such as RAM and NVMe SSDs. In June, the dedicated server range for new orders was reorganized and monthly fees went up, while existing rentals were left out. The size of the June change varies by source: some publications report an average increase of around one hundred percent on new orders, and much higher figures on some cloud products and memory add-ons. We could not verify those numbers, so we stress only the direction: three separate adjustments in 2026, each upward.

OVHcloud

In March OVHcloud announced an average increase of 9 to 11 percent for cloud infrastructure deployed between 2026 and 2028, an increase of 2 to 6 percent on older solutions, and caps of 10 percent for disks and 15 percent for RAM on option renewals. In the same post it projected that RAM and disk prices could rise by 15 to 300 percent compared with 2025, and that RAM could reach roughly two and a half to three times its September 2025 price by year end. The second wave announced in August was larger: from September, new-generation servers saw increases of roughly 40 to 59 percent, and the newest generation of the gaming range up to 87 percent. The company's CEO said the price paid for memory had risen about six times in the year to June 2026, could reach nine times by September and twelve times next year, and that NVMe drives had become about seven times and hard disks about three and a half times more expensive. Existing customers were given a much lower increase than new orders. From 1 October, storage and IPv4 addresses are shown as separate invoice lines on some cloud instances.

Others

Scaleway was reported to have raised compute instances by roughly 2 to 11 percent and block storage by roughly 10 percent in June, citing the hardware crisis in RAM and storage. Many global VPS and cloud providers are also reported to have adjusted their memory-heavy plans. In short, this move is not specific to one company or one country.

ProviderReported adjustmentStated reason
HetznerCloud roughly 30-37 percent and dedicated roughly 3-21 percent in April; further increases on new orders in JuneIT hardware prices, RAM and NVMe costs
OVHcloudAverage 9-11 percent in April; roughly 40-59 percent on new servers in September, up to 87 percent in the gaming rangeMemory and storage costs multiplying
ScalewayCompute roughly 2-11 percent and block storage roughly 10 percent in JuneHardware crisis in RAM and storage
AWSFour updates to GPU reservations during 2026, each roughly 10-20 percentSupply and demand based pricing
Microsoft AzureRoughly 8-17 percent by region from February 2027Regional market conditions and investment

Why is everyone moving in the same direction?

Hosting is a margin business: the provider buys hardware, amortizes it over years, and covers energy, network, staff and support from the rental fee. When hardware costs rise several times over, buying new servers and renting them at the old price either loses money or forces the provider to slow its investment. So providers are left with essentially three options: adjust prices, change the product structure (for example billing IPv4 or storage separately, or removing short-term plans), or slow down investment.

It is natural that a model that never adjusts prices looks attractive in the short run. Over the long run, however, it can create pressure to cut back on spare parts, hardware refresh, network capacity and support quality. That is why much of the industry treats passing on at least part of the cost as a sustainability requirement. What matters is moderation and transparency: it should be clear which items drive the increase, how existing customers are protected and what the price covers. The lower rate OVHcloud applied to existing customers, or Hetzner leaving existing rentals out of its new structure, are examples of that balance.

How long will this last?

There is no definitive answer, and the statements differ. TrendForce expects roughly a 20 percent increase in the last quarter, and some analyses mention a possible re-acceleration in the first quarter of 2027. Gartner says the effect will last through 2027. On the manufacturer side, Micron's top management said the shortage could extend into 2027 and ease in 2028, a Samsung executive said it could last until 2028, executives from the SK Group spoke of 2030, and an executive at a computer maker said prices could start to fall from late 2027. New fab investments have been announced, but they take several years to reach production.

So when planning, it is safer to prepare for a scenario of high but fluctuating prices rather than an optimistic return to the old levels.

Explore dedicated servers in Turkey

In Istanbul, in a facility built to Tier 3 standards, with dedicated hardware and remote management (IPMI/KVM) options.

See the Turkey dedicated server plans

8 practical steps to protect your budget in this period

  1. Size resources to real usage. Look at the last 30 days of CPU, RAM and disk graphs. Moving a server that constantly runs below 20 percent to a smaller plan is the fastest saving.
  2. Shut down idle resources. Test, staging and old project servers are the items most often forgotten on invoices.
  3. Calculate the total cost. When comparing prices, add items that may be billed separately, such as IPv4 addresses, extra storage, backups and traffic; some providers now show these separately. For a method, see our server cost and TCO guide.
  4. Consider the billing term. Longer billing terms can give budget predictability while prices keep moving, but do not commit to more than you need.
  5. Choose between dedicated and VDS correctly. If your workload is spiky, a VDS usually fits; if it uses steady, heavy memory, a dedicated server is usually more efficient. Read the VPS versus dedicated server comparison for details.
  6. Pick the location by where your users are. If your audience is in Turkey, a local location simplifies both latency and questions about where data sits. Our 12 things to check when buying a VDS list can help you decide.
  7. Optimize memory use. Database caching, application memory leaks and unnecessary services are often a cheaper fix than a plan upgrade.
  8. Check price lists regularly. In this period, prices for new orders and renewal prices can diverge; the most reliable way to see current prices is to check the relevant category page.

Where does SistemDC stand in this picture?

SistemDC has been operating since 2012 and runs its own autonomous system (AS211804). We deliver server and rack services from our Istanbul facility, which is built to Tier 3 standards. Because hardware comes from the global supply chain, we want to say openly that we also keep our plan structure and prices up to date in light of the change in memory and storage costs; the aim is to maintain service quality, spare parts stock and 24/7 support in a sustainable way. To see which plan offers which hardware at the current price, you can review the live listings on the dedicated server and VDS server category pages.

Many options such as RAM, disks, bandwidth and extra IPs can also be chosen separately at order time, so you do not pay for resources you do not need. For pre-sales questions you can write to us through our contact page.

Conclusion: a short-term swing or a new lasting balance?

The memory demand created by AI models has changed the cost structure of the server market. The fact that providers of very different sizes, from Hetzner to OVHcloud and from AWS to Azure, are moving in a similar direction shows this is not a single company's choice. The speed of increases may be slowing, but the expectation that high levels will persist for some time is dominant. In this environment, the winners will be users who use resources deliberately, calculate total cost correctly and choose an infrastructure provider not only by the sticker price but also by transparency, support and continuity.

Size your VDS needs correctly

In Istanbul with KVM virtualization and a 1 Gbps port, with plans that scale to your needs.

Explore the VDS plans

Related Guides

Frequently Asked Questions

Why are server prices rising?

The main driver is the sharp rise in memory and storage costs. AI data centers are buying a large share of DRAM, HBM and NVMe output, so supply of the RAM and drives used in ordinary servers has tightened, and manufacturers and providers have to pass part of that cost on.

Did Hetzner and OVHcloud really raise prices?

Yes. Hetzner made several adjustments in 2026: cloud servers rose by roughly 30 to 37 percent in April, followed by further increases on new orders. OVHcloud raised prices in March and again with a second wave announced in August, and said its memory costs had multiplied.

How are AI and memory prices connected?

AI training and inference need very high memory bandwidth. HBM and server-grade DRAM use the same silicon capacity as ordinary DRAM, so when manufacturers shift capacity to the more profitable AI products, general-purpose memory becomes scarce and more expensive.

How long will these increases last?

There is no firm date. TrendForce expects increases to continue at a slower pace in the last quarter of 2026, and Gartner says the effect will persist through 2027. Some manufacturer executives expect supply to ease from 2028, while others expect a longer squeeze.

Do all providers have to raise prices?

It is possible to hold prices while hardware costs rise, but over the long run that risks cutting back on investment, spare parts and support quality. That is why many providers treat passing on at least part of the cost as a sustainability requirement.

What can I do to limit the impact?

Size resources to real usage, shut down idle servers, consider longer billing terms where they fit, and when comparing offers include separately billed items such as IPv4 addresses, storage and traffic in the total cost.

Are servers in Turkey affected by this increase?

Hardware comes from the global supply chain, so data centers in Turkey face the same memory and storage costs. The most reliable way to see current plans and prices is to check the live listings on the SistemDC category pages.

Server packages covered in this article

+902243340242