On July 28, 2026, Hostinger — one of the largest web hosts in the world — announced the biggest server order in its history: more than 3,000 servers, a new data centre in Düsseldorf, and expansions in Paris and Asheville. Buying servers is normal. What was unusual was how candid they were about why: overall server costs up roughly 5–6x year over year, some components 6–8x more expensive than in early 2025, and lead times stretched from a few weeks to around 30 weeks — with some manufacturers sold out entirely. The order is sized to cover their needs for a full year, because they no longer trust the market to sell them hardware when they need it.
That is not a procurement anecdote. It is the clearest public signal yet of a cost shock working its way through the entire hosting industry — one that will reach most customers as a price increase on their next renewal. Here is what is actually happening, with numbers you can verify.
The numbers behind the squeeze
The short version: server memory has become scarce and expensive, fast.
- DRAM contract prices rose 90–95% quarter-over-quarter going into January 2026 — close to doubling in ninety days.
- A 32GB DDR5 kit that cost about $95 in mid-2025 is projected to peak at $550–600 by Q2 2026 — roughly a 480% increase.
- Samsung and SK Hynix raised server DRAM prices 60–70% versus Q4 2025.
- Dell announced 15–20% server price increases back in December 2025 — before the worst of the surge.
- Memory now makes up as much as 25% of a server's bill of materials, so memory prices move whole-server prices.
- Gartner projects a ~130% memory cost surge, with the crunch extending into 2027. SemiAnalysis calls it a "once-in-four-decades" event.
Why this is happening (and why it will not resolve quickly)
This is not a factory fire or a shipping backlog. It is structural. The memory manufacturers — Samsung, SK Hynix, Micron — are reallocating fab capacity away from commodity DRAM toward HBM (high-bandwidth memory), the specialised stacked memory that AI accelerators consume in enormous quantities. HBM sells at far higher margins, so every wafer of capacity that can move, moves.
The result: the ordinary DDR5 that goes into web servers, database servers, and VPS nodes is being produced in smaller quantities exactly as AI data-centre buildouts consume record volumes of everything else in the supply chain. Demand up, supply deliberately down. That is why analysts expect this to run into 2027 rather than correct in a quarter or two.
How this reaches your hosting bill
Hosting is a capital-intensive business with thin margins at the commodity end. When a host's cost per server rises 5–6x on new purchases, there are only three places for that money to come from:
- Price increases — the most common response. Expect renewal prices and "new customer" prices to diverge sharply this year.
- Density increases — more customers packed per server, which shows up as degraded performance rather than a line item.
- Plan shrinkflation — same price, less RAM per plan, tighter limits, resources moved behind higher tiers.
Watch for all three. A host that says nothing about pricing while its hardware costs quintuple is answering the question somewhere else — usually in density or plan limits.
Not every host is exposed the same way
Here is the part of this story almost nobody is explaining: the squeeze hits hosts that must buy hardware, in proportion to how much they must buy.
Hostinger's 3,000-server order is rational precisely because they are growing fast on new-generation hardware (DDR5, latest AMD platforms) — they have no choice but to buy into the worst component market in four decades, and buying a year upfront beats 30-week lead times. Hyperscalers are in the same position at larger scale.
But a host running on already-purchased, fully-depreciated hardware is largely insulated this cycle. Older-generation DDR4 platforms are yesterday's stock to the supply chain — they are not the parts being fought over. A provider whose fleet is already racked, already paid for, and running well below capacity does not need to touch the 2026 hardware market at all. Its costs are not rising, because it is not buying.
That is DeployBase's position, and we can be specific about it: our production fleet runs on EU bare-metal (Hetzner, Finland) that we already operate, at utilisation levels that leave us years of headroom at current growth — as of today, our memory utilisation is in the single digits. We are not in the market for DDR5, we have no fixed-term contract expiring into a repriced market, and nothing about this cycle forces our costs up right now. That is a statement of fact with a date on it (July 2026), not a forever promise — an honest host should not make forever promises about a market this volatile, and we would encourage you to be sceptical of any host that does.
What to do if you run apps on someone else's infrastructure
- Check your renewal date and terms now. The gap between promotional and renewal pricing is about to widen industry-wide.
- Ask your host how the memory shortage affects their pricing. The quality of the answer tells you a lot. "It doesn't, because we are not buying hardware this cycle" and "we have secured supply through 2026" are both good answers. Silence is not.
- Benchmark your current performance. If your host absorbs costs through density, you will see it in response times before you see it in an email.
- Right-size before you migrate. RAM is the constrained resource; workloads tuned to use less of it are cheaper to host everywhere this cycle.
The bottom line
The 2026 memory shortage is real, structural, and industry-wide, and it will show up in hosting prices through 2027. The hosts most exposed are the ones buying new hardware at scale; the least exposed are the ones running paid-for hardware with headroom. When your renewal email arrives this year, you will know exactly why — and you will know which questions to ask.
