It would be easier for us to run everything ourselves. One installation, every client on it, your data sent to our servers. That is how most AI companies work, and it is a sound business.
We do not do it, and the reason is not caution. It is a view about who the intelligence belongs to.
01Intelligence should come to your data
The data that matters in a company is the data it can least afford to send away. Supplier prices. Payment records. Designs. Contracts. The honest way to work with that data is to bring the system to it. Sovereignty is a design choice we make at the start, not a setting we add later.
02A promise is not a control
A contract can say your data will not be used, seen or kept. I do not doubt the people who sign such contracts. A promise still depends on someone else's systems, staff and future. A system that runs inside your walls does not need the promise. There is nothing to leak because nothing left.
03What you build should stay yours
When a system has learned your materials, your vendors and your documents, something valuable exists that did not exist before. I think it belongs to you. If it lives on my servers, you are renting your own knowledge back from me. If it lives on yours, it is an asset you own.
04The objections I hear
- "Local models are weaker." For open-ended writing, the largest hosted models lead. For reading a description, flagging a payment or finding a clause, well-chosen local models are enough and more consistent.
- "It needs enormous hardware." Most of the work is rules and search. The model is used sparingly, and the hardware follows from that.
- "We cannot run it ourselves." You can, after a proper hand-over. Making sure of that is part of the job.
05Where I draw the line
This is not a rule against the cloud. A website should be hosted where it serves visitors best. And where a client decides a hosted model suits a task, it can be switched on, deliberately, with a record. What I will not do is make an outside service a requirement for intelligence that reads your private data.
In short
- Intelligence should be brought to a company's data, not the reverse.
- A system that runs inside your walls replaces a promise with a fact.
- What a system learns from your data should be an asset you own.
- Outside services can be a choice; they should never be a requirement.
Questions
Why does Quantum Beetle deploy on the client's own servers?
Because, in Achintya Verma's words, intelligence should come to your data, and what is built from that data should belong to the client.
Is he against cloud AI services?
No. He treats them as an explicit choice a client can make for a given task, not as a requirement.
Does this cost more?
It changes where the cost falls: more at the start, little that grows with use. For work that runs over every record, every day, that usually favours running it yourself.
Sounds like your problem?
Tell us about it. We'll say honestly whether the swarm can help, and what it would take.
Read next
- AI should sit inside the business, not on top of itConviction
- One brain, multiple limbs: why I built the company this wayArchitecture
- The layer behind the dashboardProduct thinking
- One company should own the resultHow we work
- A small business deserves the same engineering as an enterpriseWho we build for
- What I ask before we take on a projectJudgement