Earlier this month, Azure customers across Europe received a Service Health notice. From February 1, 2027, most virtual machine series and Azure Storage, including Managed Disks, cost 17% more in North Europe (Dublin) and 9% more in West Europe (Amsterdam).
Take an organization spending €20 million a year on Azure compute and storage in Dublin. It pays roughly €3.6 million more per year, over €10 million across three years, without changing anything. The same estate in Amsterdam pays about €2.3 million more per year. In less than four months, it has to decide what stays, what gets modernized, what moves, and what it commits to.
Many European organizations are in this position because of choices made years ago. Dublin and Amsterdam became the default European regions because new Azure services usually arrived there first. Capacity limits seemed a distant, theoretical concern.
They are no longer theoretical. The cloud is not a magical thing in the sky. It is someone else’s computer, in their data center, and that data center faces the same physical limits as any other building.
What Microsoft announced
Four documented changes take effect on February 1, 2027.
What changes in Amsterdam and Dublin?
| Change | Amsterdam | Dublin |
|---|---|---|
| Most VM series and Azure Storage, incl. Managed Disks | +9% | +17% |
| Legacy v1 and v2 VM series, in every region | +25% | +25% |
| Exempt Intel v7 sizes available in the region | No | No |
| Reservations bought on or after February 1 | Not exchangeable | Not exchangeable |
Source: Microsoft Azure Service Health notice, October 2026.
Summary of the announcement
- v1 and v2 are excluded from the regional increase because they take the larger, global 25% instead.
- Microsoft now classifies VM series as current (v6, v7), extended (v4, v5) and end of life (v1 to v3). Dv3 and Ev3 retire on November 15, 2029.
- The exempt Intel v7 sizes currently run only in Sweden Central and Germany West Central within the EU.
Microsoft’s stated reasons are market conditions, consistency across regions, and investment in capacity and new hardware. Everything that follows about why is my own analysis.
A perfect storm
As I see it, four pressures meet in Dublin and Amsterdam, and each one amplifies the next.
1. Land and grid are fixed
Anyone who has run an on-premises data center knows the limits: floor space, power, cooling. Cloud providers face the same ones at far larger scale, and must always keep headroom for demand spikes. In Noord-Holland, TenneT’s high-voltage grid reached maximum capacity in October 2023. Large consumers face waiting lists of at least ten years, and Amsterdam’s data center vacancy is 2.7%, the tightest in Europe. In Dublin, EirGrid effectively stopped connecting new data centers from 2022. New sites may connect again since January 2026, but only with their own power plant or battery, sized to full demand.
2. Running costs keep rising
Data centers already use more than 22% of Ireland’s electricity, and Irish power costs are among the highest in Europe. Dutch household network charges rise 19% on average in 2027, and 23% in the Liander area that covers Amsterdam. Heavy users are expected to pay more from 2029.
3. Buying more capacity is getting harder
Even with space and power, filling a building is now slower and more expensive. AI data centers are absorbing memory production, and DRAM contract prices rose about 95% in a single quarter in early 2026. Microsoft’s CFO attributed roughly €22 billion of a record €170 billion 2026 capital budget to higher component prices. She also expected capacity constraints on CPUs, GPUs and storage through at least 2026.
4. Old hardware occupies the scarce floor
Older generations do less work per rack and per watt. Where space and power are cheap, that hardly matters. Where both are sold out and new servers are scarce, every rack of decade-old hardware blocks capacity that cannot be added any other way.
Put together, Microsoft cannot build its way out of these two regions, and cannot easily buy its way out either. The remaining lever is to free the floor it already has and steer new growth elsewhere. Migration is hard and costly, so few customers move unprompted. I read the price increase as that prompt: pay more for location, and more again for age.
That changes the customer’s options. Staying in Dublin or Amsterdam remains possible. Staying and growing there is not realistic: new capacity, quota and the newest VM sizes will arrive in other regions first. This is not a one-off adjustment. It is the direction for years to come.
Frankfurt is showing the same signs
While you solve your Dublin and Amsterdam problem, look at Frankfurt. Germany West Central is one of only two EU regions that offer the exempt v7 sizes today, which makes it an obvious destination. The metro around it is already under the same pressure.
- Data center vacancy in Frankfurt fell to 3.8% in September 2026, from 6.1% in mid-2025.
- Amprion, the transmission operator for the Rhine-Main area, puts new connections above 50 MW at some substations into 2029 or 2030.
- Mainova, the local utility, does not expect large new connections to be possible again until the mid-2030s.
- Germany’s Energy Efficiency Act adds obligations: new capacity must reuse 10% of its waste heat from 2026, and 20% by 2028.
Community tracking does not report Germany West Central as capacity-constrained today, and I don’t expect a price change there tomorrow. But Dublin and Amsterdam once looked the same. I’d expect any metropolitan region with a congested grid to become a candidate for the same pricing logic, sooner or later.
So while you decide on Dublin and Amsterdam, turn the decision into a playbook: how you measure exposure, how you choose placement, how you price a move, and who signs off. When the next notice arrives, for Frankfurt or anywhere else, you run a known process instead of starting from scratch.
The cost of doing nothing
Doing nothing is also a decision, and its price is simple arithmetic. Here are two illustrative estates of the same size and shape, one in Dublin and one in Amsterdam: €20 million a year each, mostly on v3 to v5, with about 15% still on v2, typically network appliances and old database servers.
What does doing nothing cost?
| Workload | Spend today | Increase | Extra / year | Extra / 3 years |
|---|---|---|---|---|
| v3 to v5 VMs and managed disks | €17.0M | 17% | €2.89M | €8.67M |
| v1 and v2 VMs | €3.0M | 25% | €0.75M | €2.25M |
| Total | €20.0M | €3.64M | €10.92M |
What does doing nothing cost?
| Workload | Spend today | Increase | Extra / year | Extra / 3 years |
|---|---|---|---|---|
| v3 to v5 VMs and managed disks | €17.0M | 9% | €1.53M | €4.59M |
| v1 and v2 VMs | €3.0M | 25% | €0.75M | €2.25M |
| Total | €20.0M | €2.28M | €6.84M |
These figures are illustrative, at list-price increases. Your actual exposure depends on your mix of VM series, storage meters, discounts and commitments.
How much of that you actually pay depends on how you buy:
- Pay-as-you-go, EA and MCA rates take the full increase on February 1.
- Savings plans do not protect you. The hourly commitment is fixed, but the rate it buys is not, so it covers fewer hours.
- Reservations bought before February 1 hold today’s price for their term, but only for v4 and newer.
Amsterdam’s smaller regional increase still adds €2.28 million a year. Note that the v1 and v2 share costs the same in both regions: location is priced regionally, age is priced globally.
Moving the data costs less than you think
The usual objection to moving is the data. The transfer itself costs less than most people expect. Here is one petabyte from West Europe (Amsterdam) to Sweden Central, at Microsoft’s list price, converted to euros.
What does moving the data cost?
| Item | Value | Basis |
|---|---|---|
| Volume | 1 PB (1,000,000 GB) | assumed |
| Transfer between Azure regions within Europe | €0.018 / GB | observed |
| One-time transfer cost | €18,000 | calculated |
| Time at a sustained 10 Gbps | About 9 days | calculated |
| Same volume leaving Azure to the internet | €0.045—0.078 / GB | observed |
Microsoft list prices, converted at about $1.12 per euro (October 14, 2026). Excludes engineering, testing and parallel running.
Illustrative data transfer costs
That €18,000 covers bandwidth only. A real migration also costs engineering, testing, possible downtime, and a few weeks of running both environments in parallel. AI tooling now greatly accelerates and simplifies much of that work, from dependency mapping to configuration changes and test generation, but it does not replace it completely. It is a line item in the business case, not a show stopper.
Set that against the cost of staying: €3.64 million a year in the Dublin example, €2.28 million in Amsterdam. The move is paid once. The increase is paid every year.
And of course, data and the compute that reads it move together, in a planned order. Split them across regions and every read crosses the border: a data platform pulling 200 TB a day would pay about €3,600 a day, roughly €1.3 million a year, for nothing.
Decide placement first, then commit
The order matters. A commitment made before you know what should stay can lock in the wrong answer for three years.
- Establish your exposure. Inventory VMs by series, region and billing meter. Microsoft lists storage by service type rather than SKU, so check meters, not assumptions.
- Decide placement intent per workload. Stay, modernize in place, or move. Judge each option on price, technical fit and growth together. Technical fit covers v6 and v7 availability, vendor support for appliances, latency and data residency. Growth asks whether the region can give you the quota and capacity you need next year, not only today.
- Rightsize and switch off. The increase is a percentage. Every size you drop and every idle VM you stop makes it smaller, and shrinks what you would commit to.
- Get off v1 and v2 first. They take the 25%, can no longer be reserved, and Dv2 and Dsv2 retire on May 1, 2028 anyway.
- Know which moves are a resize and which are a replatform. Moving within v3 to v5 is usually an instance change: a resize and a restart, with care around local temp disks. Moving to v6 or v7 is a replatform. Disks are presented over NVMe instead of SCSI, networking needs the MANA driver, and the VM must boot as Generation 2. That means new images, driver updates and testing, and some appliance vendors may not support it yet. The exempt v7 sizes therefore cost both a replatform and a region move. Price the two paths separately.
- Price the full move before you make it. Include transfer, engineering, testing, downtime and parallel running. Quota is per VM family, per region, per subscription: check it before your landing zone is ready, not after.
- Commit only to what stays. For stable workloads that remain in Dublin or Amsterdam, buy or exchange reservations before February 1. They keep today’s price for one or three years and one final exchange. Reservations bought later cannot be exchanged at all.
One caution about the obvious exit. If many customers move to the same uncongested region, it will not stay uncongested for long. The lasting fix, I’d argue, is not a new region. It is portability: infrastructure as code, landing zones ready in more than one region, and data placed so that moving it is a decision, not a project.
Ten minutes to a decision-ready comparison
Seven steps, less than four months, and every step depends on numbers. This is the decision we built CaeliQ for.
Once your billing data, contracts and landing zones are ingested, CaeliQ produces the report in ten minutes. It shows your exposure to the February changes, the cost of doing nothing, the options available to you, and the risk of each. Technical fit and growth sit in the same comparison as price, and every figure is labeled as observed, calculated or assumed.
Speed matters because the analysis is not where the time should go. A comparison in ten minutes rather than weeks leaves time to validate assumptions with your engineers, negotiate with Microsoft or your partner, and plan execution before February 1.
The same platform answers the questions this change will raise next: where a new workload should land, and whether another provider fits it better. Azure’s price change is today’s trigger. Placement is the ongoing decision.
Your business should decide
Microsoft has made a business decision about its data centers. Your response should be a business decision too, made on your own numbers rather than on a provider’s price list.
If you run a significant Azure estate in Dublin or Amsterdam, I’d like to talk with you about the approach to take. We can ground that conversation in your own numbers: ten minutes after we ingest your billing data, the report shows your exposure, the cost of doing nothing, your options and the risk of each. More at www.caeliq.io.
Sources
- Microsoft Q&A: Azure VM and Storage price update, February 2027
- Microsoft: Azure bandwidth pricing
- Alan Kinane: Microsoft puts dates and prices on the problem
- nOps: Azure VM price increase, February 2027
- Tom’s Hardware: Skyrocketing component prices push Big Tech capex to record $725 billion
- TenneT: no extra grid space in large part of Noord-Holland for the next decade
- Datacentres.com: Amsterdam data center market
- Datacentres.com: Frankfurt data center market
- Igor’s Lab: Frankfurt at its limit, power shortages slow the data center boom
- Aidan Finn: EU Azure regions capacity, June 2026
- DutchNews: energy network charges to rise 19% next year
- IT Brief: Ireland’s new rules for data center power use
- DCD: Dublin data centers and the grid moratorium
- Friends of the Earth Ireland: the cost of data centers
US dollar figures from sources are converted to euros at about $1.12 per euro.