Blog
Why your AI should run in your cloud
Where an AI system runs decides who holds your clients' data, which processors you answer for, and what you keep if you change supplier.
Shivang Patel, Mirai360 · 19 July 2026
Many established companies already rent computing from a cloud provider. It is the account that runs the website, the document store and the line-of-business systems, under access controls your IT team manages. Running an AI system in your own cloud means building it inside that account, rather than sending your documents out to someone else's service.
Follow one document to see the difference. Say a client's signed contract needs summarising. With a shared AI service, the contract travels to the provider's systems, is processed under the provider's terms, and the summary travels back. In your own cloud, the contract, the stored copy, the logs and the access rules all sit in an account you control.
The language model that does the reading may still be an outside service, and that is the part to pin down: which model providers see which data, and on what terms. We agree that with you and write it into the data agreement before work starts.
Why location matters
Four consequences follow from where the processing happens.
The first concerns promises to clients. Many companies hold data under explicit promises in engagement letters, non-disclosure agreements, supplier terms or professional duties of confidentiality. Sending a client's material to an outside AI service can break a promise about who holds it, even if nothing goes wrong. With the system in your own account, you can tell a client exactly where their documents are kept and who can open them.
The second is UK GDPR, which requires a written contract with every company that processes personal data on your behalf. Building in your own account keeps the stored data under the agreement you already have with your cloud provider. Any outside model provider the system calls is a processor too, and so are we whenever we handle personal data, so each needs its own contract. Treat this as a starting point rather than legal advice, and ask your data protection officer what applies to you.
The third is retention and training. A common worry with consumer AI tools is whether what you send is kept, or used to train someone else's model. When the system runs in your account, how long anything is kept is a setting you control. For any outside model provider, get its answer on both points in writing before your data reaches it.
The fourth is changing supplier. A system builds up history over time, including its instructions, its tests, its logs and the corrections your people made. When that history sits in your account, changing supplier means changing who maintains the software, rather than asking for your own records back. You own the code, the instructions, the evaluations and the data written for your engagement.
What running it yourself costs
Running in your own cloud needs a cloud account, a person accountable for it, and a team responsible for updates and monitoring. A company that already runs its systems in the cloud has the first two, and the third is part of what we do.
We build on de-identified samples, and your own staff keep access to live data. If we need to see live data, for example to trace a fault on a real case, including from our team in India, your company first approves a written agreement covering that processing and the transfer abroad, on the advice of your data protection officer.
A managed service, where the supplier runs the system on its own infrastructure, leaves retention, access and training under the supplier's controls. The contract then carries the whole weight, so it needs to state each of those three explicitly. Which route suits you depends on the data you hold and what you have promised about it.
The logs are yours too
A full record of what the system read and did serves both a privacy review and your own operational checks. When that record sits in your account, you can audit the system without asking a supplier's permission. The logs hold personal data themselves, so they need their own retention period, and your data protection officer will probably want a data protection impact assessment before launch.
When a larger customer's procurement team sends a security questionnaire, the same record lets you say where your AI processing happens and show who accessed what.
Five questions for any supplier
Put these in writing and keep the answers:
- Where exactly is our data processed, and can all of it run inside our own cloud account?
- Which outside model providers see our data, do they keep any of it, and for how long?
- Is any of our data used to train a model we do not control?
- Who can access our data, and is every access logged?
- If we leave, what do we take with us, and in what form?
Any answer that comes back vague goes into the contract negotiation as a clause to be written before you sign.
If you are weighing an AI system for a task that handles client data, take the ten-question survey to see whether AI fits that task. Find your bottleneck