I Spent $30K on Local AI Hardware. Here's Why That Matters for Your Data.
Every security team I work with eventually asks some version of the same question. Where does our data go when you use AI tools. It's the right question, and most agencies don't have a real answer for it beyond "we're careful."
I have a better one.
The Two-Part Answer
Part one: every frontier AI tool I use professionally runs with data collection and training turned off. Not by default. I turned it off, on every provider, every time. That's a setting, and I checked it, and I keep checking it.
Part two: the most sensitive work never touches a third-party API at all. I run local AI models on my own hardware. No prompt leaves the building. No inputs get logged by someone else's infrastructure. No vendor terms of service to trust.
What I Actually Built
Over the last year I put together roughly $30,000 in local AI infrastructure, and I didn't buy one machine. I built a mixture: leading-edge Apple Silicon, leading-edge Nvidia hardware, and leading-edge AMD hardware, running side by side. Different chips are genuinely better at different jobs, and I wanted the best model, the best experience, and the best outcome for whoever I'm working with, not whatever happened to be sitting on my desk. Three separate local model systems, running in parallel, doing real work.
This isn't a hobby setup. It's the same discipline I bring to everything else: don't rent what you can own, don't outsource what you can control, and don't settle for one tool when the job calls for three.
The Conversation This Actually Solves
Here's the moment I built this for. You're the head of marketing, or you're the CEO, and you're the one who has to walk into a board meeting and say some version of "we should spend $200,000 with this guy."
Someone on that board is going to ask the obvious question. Who is this person, really. Is this a solo operator with a laptop, or is this someone who can actually be trusted with our growth and our data.
I want that question to answer itself before it's even asked. How many agencies do you know that run multiple custom hardware systems, each purpose-built for a specific kind of job, tuned for quality and speed at the same time. That's not a random agency. That's someone who built the infrastructure before you ever asked for it.
Why This Is Not a Nice-to-Have
For most businesses, "our AI vendor has good data practices" is a reasonable answer. For a security company, a fintech platform, or anyone whose entire product is built on trust, it isn't enough. Your buyers are trained to ask the follow-up question. Your board asks it. Your own customers ask it of you.
I'd rather answer it before anyone has to ask.
The Practical Reality
This setup isn't slower or more limited than renting compute from a frontier provider. Local models have gotten genuinely good, and for a growing share of the work, running it locally is now the better technical choice, not just the more private one. The privacy is the reason I built it. The performance is why I kept building it out.
If your team is asking how an outside operator handles AI and data, I'd rather show you the actual hardware than give you a policy document.
Want to see how this fits into your own security review? [Reach out](/#contact) and I'll walk you through it directly.