The US data zone costs 10 percent more. The EU one is not on the price list yet.

What the post says

The post presents GPT-6 Astra as a frontier model built for work: it reasons through a problem in several steps, makes a plan, uses tools, and drives the computer itself in workflows that have no dedicated API. There are two deployment types, Standard (pay per token) and Provisioned Throughput (reserved capacity), and two geographies, Global and US Data Zone.

The four Standard price rows, input and output, in US dollars per million tokens (cached input and cached write prices left aside):

For Provisioned Throughput the post gives no dollar price, only that the US Data Zone version carries a 10 percent premium over Global. It says nothing about Europe, an EU data zone, data residency, or when further zones will follow.

We have no quarrel with the numbers. The question is what they mean for a company whose contracts require EU processing.

On 3 September there was no EU processing option to select

Microsoft's own documentation on Foundry deployment types, updated 12 August 2026, draws the line. Global: prompts and responses may be processed in any Azure region. Data Zone: processing stays inside the chosen zone, the United States, the EU, or Asia Pacific. Data at rest stays in the selected geography under every type. What moves is where inference happens.

For GPT-6 Astra on 3 September, two of those were for sale. Global: processed anywhere. US Data Zone: processed in the United States, at a 10 percent premium. If you needed EU processing, there was nothing to select.

Many companies have a data processing agreement or an internal policy that keeps personal data processing in the EU. If you have treated that as a property of the platform, launch day showed that the newest model lacks it, and that the nearest option, the US zone, costs 10 percent more. Data residency became a priced line item with its own release order.

The delay is by design, and Microsoft documents the order

Here we have to be precise, because the easy reading is wrong. The same documentation states that new deployment types arrive in a fixed sequence: Global first, then Data Zone, then geography-based, with no guaranteed date for the last. An EU Data Zone exists as a deployment type, under the Azure EU Data Boundary, which can also cover EFTA countries such as Norway and Switzerland. Which deployment types a model supports varies from model to model.

So the missing EU row is almost certainly scheduling, and we do not read an exclusion policy into it. We expect it within weeks, probably at a premium similar to the US one. Reading this post as Microsoft shutting Europe out is wrong, and within weeks that will be plain to see.

The narrower claim holds, and the documentation is what supports it: the newest capability reaches Europe later and costs more, and the next model will follow the same order. If your procurement assumes day-one EU processing for the latest model, that assumption is false today.

What we think, including the parts that cut against us

Four things need saying before we talk about our own answer.

First: we cannot deliver GPT-6 Astra. Only Microsoft and OpenAI can. We run open-weight models on EU GPUs: a different model, closer to you. Bringing the same model closer is not something we can do either. If your workflow needs capability that today exists only in frontier models, our answer does not replace Foundry. For document search, extraction, summarisation and internal Q&A, an open-weight model is often enough. Whether it is enough for you, we cannot tell in advance, which is why our pilot is paid and runs on one of your real workflows.

Second: the delay is scheduling, as set out above. We do not argue that Europe is left out, only that it gets the model later and pays more. That is a smaller claim, and it is true. If you are reading this weeks later, the EU row may already be on the price list. Our headline will go out of date; the pattern will not.

Third: 10 percent on its own is small. On a Foundry bill of a few hundred dollars a month it is a rounding error. It becomes material when the premium and the delay both go into the contract and return with every new model.

Fourth: we have an interest. We sell exactly what we are arguing should be the default. That is why we spelled out the part of Microsoft's documentation that works against us.

What we do instead, and what you give up for it

Pricing follows a different logic too. We charge a fixed monthly fee for one operated confidential workflow, with no per-token billing. When we put a better open-weight model into service, neither the bill nor the processing location moves. ATAILA Cloud and ATAILA Enterprise differ in whose hardware they run on. The data path is the same.

Here is the trade: in Foundry, Microsoft's capacity planning decides when a route appears. With us there is one route, and its location is known. You get fewer models to choose from. In return, the processing location is written into the contract, and no release schedule rewrites it.

Where we stand

We have no argument with Microsoft's announcement: a new model, two deployment types, two geographies, a clean price list. The documentation is honest too: Global first, Data Zone next, geography-based last, with no date.

Our conclusion is different. If you want the newest model and your contracts require EU processing, in this order you will always wait and pay more. If the wait fits your business, Foundry is a good choice, and the strongest model is there. If your workflow cannot wait and the data cannot leave, you need a different model, closer, and you need to know what you are giving up for it.

One testable sentence: if your contract requires EU processing, you should be able to say in one sentence which model runs under it today, and what happens to it when the vendor ships a new one.