The data centre question, answered without pretending we are not part of it.
Almost every business tool now runs on hardware somebody else built in a building near a substation. That is now a contested arrangement, and this page is about the part of the contest FoundrySuite is on the right side of — and the part where the honest answer is less comfortable.
What is genuinely different
Not "we are greener" — that framing is available to anyone who buys offsets, and it is not a software property.
The real difference is that adopting FoundrySuite requires no new infrastructure to be built, by anyone. No campus, no power contract, no water study, no planning application, no interconnection queue, no local planning committee that has to be persuaded. The machine is one you already own, in a building you already occupy.
That is a real and quantifiable footprint difference rather than a slogan, and it is the one that matters when a county, a planning authority or a customer asks what your software costs the neighbourhood.
What is not different, stated plainly
- It uses electricity. A 3-billion-parameter model running on a CPU draws power, and someone is paying for it. The honest comparison is small local draw against a facility built for the purpose — not against zero.
- Running it locally is slower. That is the trade you are making: no new build, no per-query bill, no queue, at the cost of speed.
- It only helps if your hardware can carry the model. A machine that cannot run it well will be an unhappy experience, and we would rather tell you that before you buy than after.
- Pointing it at a remote endpoint undoes the whole argument. Then your prompts go to someone else's data centre and you are back where you started. FoundrySuite says so on screen when it does.
The other half: what you were buying from a cloud vendor
For most small businesses the deciding factor was never the carbon. It was the standing arrangement that came attached. The money half of the same argument is on the pricing page, and it is not repeated here.
| Cloud-hosted AI | FoundrySuite | |
|---|---|---|
| Your data | On someone else's infrastructure, in a region they chose | On the disk you installed it to |
| Who can compel access | The vendor, and whoever they answer to | Nobody. There is no copy |
| AI cost | Metered, and rising — the meter can change | Zero, and it cannot rise |
| If the vendor changes terms | Your business changes with it | Your copy does not change |
| New infrastructure required | Yes, by the provider | None, anywhere |
| Needs a connection to work | Yes, for everything | Everything except researching a company online |
| Data protection law that applies | GDPR, transfers, sub-processors, residency | Your own file, on your own disk |
The last row is the one people underestimate. Running on your own hardware removes a category of legal obligation rather than satisfying it with paperwork.
What we are not claiming
Every vendor in this space is asked to claim that their product is the responsible one, and most of the claims are unfalsifiable. So, the falsifiable version:
If you run FoundrySuite, you do not add load to a data centre. If you stop, nothing is switched off, because nothing was switched on. Your electricity bill changes by a laptop's worth, which you can measure on the meter and which is not zero.
Any vendor who tells you their cloud product is impact-free is describing an accounting entry. This one is describing a power draw, and you get to see which.
If this matters to you, so does the other thing
Data protection law bites far more often than energy regulation does for a small business. A processor agreement, a transfer assessment and a sub-processor list are annual obligations that arrive whether or not anyone is thinking about substation capacity.
A database on your own disk has none of them. That is a quieter win than a carbon claim, and it is the one your compliance spreadsheet will actually record.