Foundry EU Data Zone premium doubles: the Swedish cost math

From today, 1 September 2026, an EU Data Zone deployment in Microsoft Foundry costs 20 percent more than the same model on Global Standard, up from 10 percent. Regional deployments outside the United States now carry a 25 to 50 percent premium over Global, with Sweden in the 30 percent band and West Europe and North Europe at 50 percent. Microsoft published the change on 9 July with a two-month notice period, and the Azure OpenAI pricing page has carried a banner about it since. Global pricing is unchanged. If you run models on Azure for a Swedish or EU organisation with a data residency requirement, the residency premium you pay just became a line item worth re-deriving, and the fine print about which customers pay it, and when, matters more than the headline percentages.

What Microsoft changed on 1 September

The announcement is short and the table below reproduces its core. Premiums are always expressed relative to Global pricing for the same model. Microsoft's stated reason is cost: running a model at high availability inside a single geography costs more than serving it from a shared global pool, and the new prices "reflect both that investment and the compliance value these options provide."

A note on the arithmetic, because Microsoft's own summary line reads "EU Data Zone increases 9% above Global" and that is easy to misread. The table in the same post lists the EU Data Zone at 20 percent above Global and calls the move a 9 percent increase. Those two statements only reconcile if the previous premium was 10 percent: 1.10 times 1.09 is 1.20. So the EU Data Zone premium has doubled from 10 to 20 percentage points over Global, and a customer already paying the EU Data Zone rate sees their unit price rise by roughly 9 percent. The US Data Zone stays at 10 percent, so EU residency now costs twice the US residency premium.

The fine print decides who actually pays

Two footnotes in the announcement carry more budget impact than the headline table.

Pay-as-you-go customers are grandfathered per model. For Standard deployments, the new premiums "apply only to models launched on or after September 1, 2026." If you deployed GPT-5.6 Terra to the EU Data Zone in July, you keep paying the 10 percent premium on Terra for as long as you run it. The 20 percent premium arrives the day you move that workload to a model launched from today onwards. The price increase is therefore coupled to your model upgrade cadence, not to the calendar.

Provisioned Throughput customers pay from today, on every model. The PTU increase "applies to all customers with EU Data Zone or Regional PTUs outside of the US." There is no per-model grandfathering. The relative premiums quoted are for monthly and yearly PTU reservations; hourly PTU rates relative to Global "might differ", so check the pricing page for your exact SKU rather than assuming the same 20 or 30 percent.

The regional table, read from Stockholm

Regional deployments pin processing to one Azure region. Microsoft's new regional price bands relative to Global, for the regions European teams care about:

Asterisked regions are newly available at this price with no prior rate. Two things stand out. First, Sweden Central sits in the cheapest European band. Pinning to Sweden costs 30 percent over Global while the two regions most Nordic teams historically defaulted to, West Europe and North Europe, cost 50 percent. If you inherited a West Europe deployment from 2023 and your residency requirement is "inside the EU", you are paying 20 points more than you need to. Second, the gap between the EU Data Zone at 20 percent and Sweden regional at 30 percent is now only 10 points, which makes single-region pinning cheaper relative to the zone than it used to be.

There is a catch that the pricing table does not show. Regional Standard availability lags badly. Microsoft's region availability page lists GPT-5.6 Sol, Terra and Luna in Sweden Central for Global Standard and for Data Zone Standard, but the regional Standard table for Sweden Central tops out at GPT-5.1. The deployment types documentation is explicit that new deployment types arrive in a fixed order, Global first, then Data Zone, then single region, and that single-region types "have no guaranteed availability date." For the current model generation, the 30 percent Sweden rate is mostly a PTU conversation. On pay-as-you-go, the EU Data Zone is the residency option that actually has the models.

Cost math on GPT-5.6

Global Standard list prices for GPT-5.6 have not changed: Sol at $5.00 input and $30.00 output per million tokens, Terra at $2.00 and $12.00, Luna at $0.20 and $1.20. The figures below apply Microsoft's stated premiums to those Global rates. They are derived estimates for planning; the Azure pricing page publishes the exact per-model rate for each deployment type and that is the number to put in your chargeback model.

Take a Terra workload at 500 million input and 100 million output tokens per month, a typical shape for a document-heavy internal assistant:

GPT-5.6 Terra, 500M input + 100M output per month
Global Standard:              500 x $2.00 + 100 x $12.00 = $2,200
EU Data Zone, old +10%:       $2,420   (still applies to Terra, launched 9 July)
EU Data Zone, new +20%:       $2,640   (applies to models launched from 1 Sept)
Sweden Central regional +30%: $2,860   (Terra not yet on regional Standard)
West Europe regional +50%:    $3,300

Residency premium, EU Data Zone:  $220/month old, $440/month new
Annualised difference old vs new: $2,640 per year on this one workload

The same shape for Luna at high volume, 2 billion input and 300 million output tokens per month, the profile of a classification or extraction pipeline:

GPT-5.6 Luna, 2B input + 300M output per month
Global Standard:              2,000 x $0.20 + 300 x $1.20 = $760
EU Data Zone, old +10%:       $836
EU Data Zone, new +20%:       $912
Sweden Central regional +30%: $988

In absolute terms these are modest sums for a single workload. Multiply across a portfolio of twenty assistants and pipelines and the doubling of the residency premium is a five-figure annual number, which is precisely the size at which someone in finance asks whether every workload in the EU Data Zone needs to be there.

One timing detail for Sol users: Microsoft's GPT-5.6 launch post states that metered Sol pricing drops to $4.00 input and $20.00 output from 1 September through at least 30 November 2026. Sol launched on 9 July, so an EU Data Zone Sol deployment keeps the 10 percent premium on top of the promotional rate. For the next three months, EU-resident Sol is cheaper than it was in August, not more expensive.

PTU reservations: the increase is immediate

Provisioned throughput bills per PTU per hour regardless of tokens, with monthly and yearly Azure Reservations discounting the effective rate. The announcement does not print absolute PTU rates, so the only defensible planning formula is relative: if your EU Data Zone PTU reservation was priced at 1.10 times the Global PTU rate, it is now priced at 1.20 times, a 9.1 percent increase on that line. A Sweden Central regional PTU reservation moves to 1.30 times Global. Whether an existing yearly reservation re-prices mid-term or at renewal is a contractual question for your account team; the announcement says the increase applies to all EU Data Zone and non-US regional PTU customers and does not carve out existing reservations.

The 3 August cost check on this blog already argued that the July GPT-5.6 price cuts pushed PTU break-even traffic up five-fold for Luna. Add a 9 percent increase on the PTU side for EU Data Zone reservations and the pay-as-you-go case for Luna and Terra in the EU zone gets stronger still. A Global PTU reservation, which is unaffected by today's change, is the other comparison to run if your residency requirement is about data at rest rather than processing location.

Global, Data Zone or regional: a decision guide

Microsoft's own guidance is to start with Global Standard and move only for a specific reason. The residency facts that should drive the choice, per the deployment types documentation:

• Every deployment type keeps data at rest in your designated Azure geography. The deployment type only governs where inference processing happens. Prompts and completions in flight are the thing you are paying the premium to keep in Europe.
• Global may process a request in any Azure region where the model is deployed, including the United States.
• EU Data Zone processes within the Azure EU Data Boundary. The deployment types page states this "can include European Free Trade Association (EFTA) countries and regions such as Norway and Switzerland in addition to EU member states", and the Data Zone Standard availability table for Europe does list Norway East and Switzerland North. The region availability page describes the same zone as "any EU member nation". If your DPIA was written on the narrower reading, reconcile it with the broader one before an auditor does.
• Regional processes in the single deployment region. It is the only option that lets you say "processed in Sweden" and it costs 30 percent over Global, where the model is available at all.

Mapped to common Swedish situations:

• Internal productivity tooling on non-personal or pseudonymised data. Global Standard. The premium buys nothing your risk assessment requires.
• Customer-facing or HR workloads with personal data under GDPR, no national-law residency clause. EU Data Zone at 20 percent. Confirm the EFTA reading is acceptable to your DPO; Norway is in the EEA and Switzerland holds an EU adequacy decision, so most assessments will accept it.
• Public sector, healthcare or workloads where a contract or procurement (upphandling) specifies processing in Sweden. Regional Sweden Central at 30 percent, on the models that are actually available there, or Regional Provisioned. Budget for older models or a longer wait on new generations.
• High, steady volume with a residency requirement. Compare EU Data Zone PTU at the new 1.20 multiplier against EU Data Zone Standard at your real cache hit rate. The July price cuts plus today's PTU increase have moved the break-even against you.

What to do this week

1. Inventory every deployment by SKU and model version. The SKU name tells you the deployment type; the model version tells you whether the pay-as-you-go grandfathering protects it.

az cognitiveservices account deployment list \
  --resource-group <rg> --name <foundry-account> \
  --query "[].{name:name, model:properties.model.name, version:properties.model.version, sku:sku.name}" \
  --output table

Anything with a sku of DataZoneStandard, DataZoneProvisionedManaged, Standard or ProvisionedManaged in a European region is in scope. GlobalStandard and GlobalProvisionedManaged are not.

2. Separate PTU from Standard. Pull the reservation list for EU Data Zone and regional PTUs and price the 1.20 and 1.30 multipliers into this month's forecast. Then decide per reservation: keep, shrink with spillover to Standard for bursts, or convert to Standard entirely.

3. Put a residency gate on model upgrades. For Standard deployments, the 20 percent premium is triggered by moving to a model launched from 1 September onwards. Add the residency cost delta to whatever review already approves a model change, so the next "let's move everything to the new model" decision includes the number.

4. Retire West Europe and North Europe regional deployments that do not need to be there. At 50 percent over Global, an Ireland or Netherlands regional deployment held only for "inside the EU" is the most expensive way to meet that requirement. EU Data Zone at 20 percent or Sweden Central at 30 percent both satisfy it more cheaply.

5. Enforce the decision with Azure Policy. The deployment types documentation includes a policy pattern for restricting SKUs on Microsoft.CognitiveServices/accounts/deployments. Use it in both directions: deny GlobalStandard in subscriptions that hold regulated data, and deny the regional SKUs in subscriptions that do not, so the premium is only paid where a documented requirement exists.

{
  "mode": "All",
  "policyRule": {
    "if": {
      "allOf": [
        { "field": "type", "equals": "Microsoft.CognitiveServices/accounts/deployments" },
        { "field": "Microsoft.CognitiveServices/accounts/deployments/sku.name", "equals": "GlobalStandard" }
      ]
    },
    "then": { "effect": "deny" }
  }
}

6. Update chargeback rates and tell the budget owners. If internal teams are billed per token, the EU Data Zone rate card for new models changes today and the PTU rate card changes for everyone. A one-line note now avoids a surprised finance team in October.

The Swedish and EU angle

Data residency in Sweden is rarely a single legal requirement. It is a stack: GDPR transfer rules, sector guidance, IMY's expectations for public bodies, procurement terms that name a country, and internal risk appetite shaped by years of Schrems litigation. Microsoft's announcement is careful to say that you "remain responsible for determining whether a given configuration meets your regulatory obligations." The pricing change does not alter what any of those layers require. What it changes is the price of satisfying each layer with a deployment type, and it makes that price visible in a way it was not when the premium was a footnote on a sales call.

That visibility is useful. A 20 percent EU Data Zone premium and a 30 percent Sweden premium are concrete numbers to put next to each workload's actual legal basis. Some workloads will turn out to be in the EU Data Zone by habit rather than requirement, and those can move to Global with a documented risk decision. Others will turn out to be on Global by oversight, and those should have been paying the premium all along. Both corrections are easier to make when the cost of residency is an explicit line rather than an ambient assumption.

For public-sector buyers the regional table is also a procurement input. A tender that specifies "processing in Sweden" now implies a 30 percent premium over Global and a model catalogue that lags by a generation on Standard deployments. A tender that specifies "processing within the EU/EEA" implies 20 percent and the full catalogue. Writing the requirement precisely, rather than reaching for the strictest phrasing, is worth a tenth of the model budget.

Finally, the direction of travel. Microsoft has raised EU residency prices while holding US residency flat and launching an APAC zone at the EU rate. The stated rationale is infrastructure cost at high availability inside a geography. Whatever the cause, Swedish enterprises should plan on residency carrying a persistent 20 to 30 percent premium in their Azure AI budgets rather than treating today's change as a one-off, and should revisit the sovereign options covered on this blog in July for the workloads where the requirement is strict enough to justify a different architecture entirely.

Conclusion

Today's change doubles the EU Data Zone premium from 10 to 20 percent over Global, sets Sweden Central regional at 30 percent, and leaves Global untouched. Pay-as-you-go customers only pay the new EU Data Zone rate when they adopt a model launched from 1 September; PTU customers in the EU Data Zone or a non-US region pay from today on every model. Inventory your deployments by SKU, re-price the PTU reservations first, gate model upgrades on the residency delta, and move anything sitting in West Europe or North Europe regional for no stated reason. The residency premium is now a number. Treat it like one.

Sources

• Microsoft Foundry Blog: Microsoft Foundry Model Deployment Pricing Update (9 July 2026)
• Microsoft Learn: Understanding deployment types in Microsoft Foundry Models
• Microsoft Learn: Region availability for Foundry Models sold by Azure
• Azure Blog: GPT-5.6 now available in Microsoft Foundry (pricing and Sol promotional rate)
• Azure OpenAI Service pricing page