Hosting price hikes in 2026: every announced increase, and when it hits you

What changed, and when it takes effect

The single most useful thing to know about 2026 hosting pricing is that there was no single increase. Hetzner adjusted prices 3 times between 2026-04-01 and 2026-06-15. OVHcloud did it 2 times, and its second wave does not reach existing customers until 2026-10-01. Anyone quoting one percentage for the year is quoting one wave.

Two Hetzner announcements state an increase and name the cause but publish no percentage. Those rows read not published rather than borrowing a number from trade press that Hetzner never stated.

The chart carries the argument on its own. The 9 to 11 percent figure that circulated most widely in 2026 is the second smallest move on the board. It was accurate for the announcement it described and badly misleading as a summary of the year.

Why memory is the cause

Providers have been unusually direct about this. Hetzner attributes its June adjustment to the massive increase in procurement costs and, in its April statement on setup fees, names RAM and NVMe SSDs specifically. OVHcloud attributes its increases to a memory market disrupted since autumn 2025 by an industrial shift toward GPUs.

The most concrete evidence on record is OVHcloud founder Octave Klaba quoting what his own company pays. Those are one company purchasing figures rather than a market index, which is both their strength and their limit: they are real transacted costs for a buyer at scale, and they are not necessarily what anyone else pays.

How a memory price becomes your hosting bill

A 6x component cost does not produce a 6x bill, and the gap between those two numbers is where most confusion about this topic lives. Memory is one line in a server that also contains CPUs, motherboards, network cards, chassis, storage, power, cooling, bandwidth, floor space and the provider margin. Klaba puts the increase on CPUs, motherboards and network cards at 15 to 20 percent, well below memory. So a multiple on one input becomes a much smaller multiple on the finished product.

That arithmetic explains the shape of the data above rather than just its level. The plans that rose furthest are the ones where memory is the largest share of what you are renting. Hetzner Arm CAX plans rose about 31 to 33 percent while its x86 CPX and CCX plans rose two to six times that. OVHcloud increases scale with hardware generation, because the newer generations carry DDR5 and the older ones do not. Legacy pre-2025 equipment moved 2 to 6 percent, and OVHcloud left the Kimsufi, So you Start and Rise ranges alone entirely.

The cleanest demonstration is the memory option line. When a provider sells you RAM as a separate item rather than folding it into a plan price, the component cost passes through almost undiluted: OVHcloud memory options on new orders rose 68 to 652 percent, against 16 to 100 percent for the whole servers those options go into. Same shortage, same company, same month. The difference is how much of the other stuff is in the price.

Read the channel before you compare two numbers

Hosting price reporting is unusually easy to get wrong, because one announcement can contain four different kinds of price at once, and they are not comparable with each other:

• List price for new orders. What a pricing page shows today. This is what almost every published percentage refers to.
• Billed price for existing customers. Often unchanged at first. Hetzner left existing contracts on their terms in June 2026. OVHcloud reaches existing customers at renewal from 1 October 2026.
• Option or add-on price. Memory or storage sold as a line item. Moves furthest, because it is closest to the raw component.
• One-off setup fee. Not a recurring rate at all. Hetzner adjusted these separately in April 2026 and reduced them for most dedicated servers in June while raising monthly prices.

Every row in the table above names its channel for that reason. A headline that compares an option price to a plan price will produce a number that is arithmetically correct and completely meaningless.

What to do about it

The single highest-leverage action is to check how much memory you are actually paying for. Because the increases track memory share rather than compute, right-sizing RAM is worth far more in 2026 than right-sizing vCPU. A plan one tier down on memory may now cost meaningfully less than it did relative to the tier above.

Beyond that: renew before a stated deadline where a provider offers it, since OVHcloud let customers renewing before 1 April 2026 hold rates for up to two years. Check whether an Arm instance fits your workload, given Hetzner Arm plans rose a fraction of what its x86 plans did. Treat legacy hardware ranges as a feature rather than a compromise while the newer generations carry the memory premium. And if you run your own hardware, the timing question is the same one covered inthe buy or wait check on the RAM tracker.

What happens through 2028

Nobody on record forecasts relief inside the next eighteen months. Klaba has said the situation is exceptional and will last until 2028, and projects the price OVHcloud pays for RAM rising to 12x the June 2025 level in early 2027. Since that is above the 6x already passed through into the 2026 increases, the reasonable planning assumption is that 2026 pricing is a floor rather than a peak, and that further waves land in 2027.

The supply-side reason is covered inthe memory capex buildout: the fabs being funded now deliver in 2028 and 2029, not into this shortage. For the underlying chip market driving all of it, seewhy RAM got so expensive and the live readings on the RAM price tracker.

How this tracker stays honest

Every row is recorded from the provider own announcement or its published price table, and carries the date it was verified. Where a provider publishes before and after prices but no percentage, as Hetzner does, the change is computed here and labelled as our arithmetic over their numbers rather than presented as a vendor figure. Where a provider announces an increase without a percentage, no percentage is recorded.

Vendor announcements are treated as vendor statements throughout, never as independent findings. A provider explanation of why it raised prices is evidence of what it says, and the component multiples above are one company purchasing data rather than a market index. Percentages are stored and shown as the ranges the sources published. The full dataset is available as machine-readable JSON.