"Processed in Spain" is not a sticker: it means that when we run out of room, we tell you so

September 22, 2026

Almost every voice provider says some version of "we process in Europe". It is easy to say and hard to check, so it's worth looking at the question that does separate them:

What happens when their capacity fills up?

There are two possible answers. One is to say there's no room. The other is to forward the request to a third party's API — usually American — and return the result as if they had done it themselves. The second is invisible to the customer, cheaper to operate, and turns the earlier promise into paper.

We do the first. If there is no free capacity at that moment, you get a 503 and you retry. It is the worse experience and the only one compatible with what we claim.

What it means exactly

Three concrete statements, which is what goes into a procurement file:

Audio is processed on our own machines, in Spain. Not on capacity rented from a cloud provider who in turn rents it from someone else. We bought the hardware, we hold it and we operate it.

There is no sub-processor outside the EU in the processing chain. There is no third-party API behind ours. The engines that turn your audio into text, or your text into speech, are the ones we publish as open source, running on our machines.

There is no international data transfer to declare in your record of processing activities, because there isn't one.

Why this shows up in procurement and not in the demo

In the demo everything works the same. The difference appears three months later, in the file, when someone in compliance asks about the sub-processor annex and the standard contractual clauses. If the answer is "none outside the EU", the file moves. If it is a list of three providers across two jurisdictions, it stops and waits for a legal opinion.

That wait costs more money than the per-minute price difference between us and anyone else.

What it does NOT mean

Because telling only the good half is useless:

• It does not mean we are immune to an outage. It means that when we go down, we go down and we say so, instead of covering the gap with somebody else's infrastructure.
• It does not mean infinite capacity. Owning machines is exactly the opposite: capacity is what it is, which is why the 503 exists.
• It does not mean we keep your audio. The opposite: it is processed and discarded. The one exception is the voice cache, which lasts an hour, costs 10% and can be switched off on every request.

How to check without trusting us

The engines are open source and published. You can read them, stand them up on your own machine and compare what comes out with what our API returns. If what you receive ever stops resembling what that code produces, something is off — and that is precisely the check a provider who forwards to third parties cannot offer you.

And for the audio we generate, the watermark carries a fixed signature: if it isn't there, it didn't come from here.

Anything to add or correct? Write to support@uttera.ai. If you correct us, we edit the post and credit you.

← All posts