If your firm’s documents go through an AI, two questions decide how much of that work is yours. Whose machine does it run on? And who decides what it does? The first is about ownership, the second about agency.
On 19 November both were put in front of a room of investors and founders at The Bridge33, an invite-only evening on digital money in Marbella, now in its second edition. Three things from it are worth an hour of anyone’s attention:
- A lawyer whose clients are wealthy individuals and family offices now calls the location of AI an asset-protection question, and said so in public.
- The hardware to act on that belongs to the people who showed it. One rig never left its owner’s house, and took his orders from the room.
- Owning the machine does not finish the job. What it sends out still has to be measured. That part is ours.
One of us was there, with a Mac mini.
The organiser’s argument
Vicente Ortiz founded The Bridge33. His day job is a law firm that helps wealthy newcomers settle in Spain, Portugal and Luxembourg: tax, crypto, estates. On the day of the event he posted his reason for giving floor space to hardware at an evening about prediction markets, stablecoins and tokenisation:
Everyone is talking about AI models. Almost nobody is asking where their data sleeps at night.
Blockchain taught us: not your keys, not your coins. AI is about to teach us the next one: not your hardware, not your data.
For a family office, a fund or a founder, data sovereignty is becoming an asset-protection question, the same way custody did in crypto.
The comparison works for this audience because they have already lived it once. Anyone who held coins on an exchange that later froze withdrawals learned that “mine” depends on who holds the keys. A document sent to a rented model sits on somebody else’s disks and in somebody else’s logs, under somebody else’s retention rules. For a holiday itinerary that is fine. For a client’s tax position it is a question a lawyer gets asked.
He closed with one for the reader:
In five years, will you own the machine your AI runs on, or rent it?
The table
The lab was called Garage · Bring Your Own Stack. The rule was that people bring the machines they run AI on, so that others can see and touch them. No slides.
The organisers announced Guti with a desk robot that works without a connection, and an open NVIDIA rig. The rig stayed at his house. What we watched him do was send it commands from the room. He calls it Jarvis. It is the second time in a month we have heard that name in Andalucía: at Malaga-AI in October a developer told a room to build their own.
Our Mac mini was on the table, switched on, and running two of our Telegram bots. One is CasaSol Guide, a property guide for the Costa del Sol. A question to it travels through Telegram, which is a cloud service. The answer is computed on the box.
At our table, one visitor’s advice is worth passing on: for a business, none of this should stay abstract. Show a real use case, and show how it solves the problem. Ownership and agency are abstract words too, and we take the point.
Agency
The second question is who decides.
The evening is itself the example. A lawyer decided that a room of investors should see where AI physically runs, and three panels, four partners and sponsors, and a hardware table were there on the night. The invitation to bring a machine reached us the evening before.
The two machines described above were the same thing at a smaller scale. Neither is a subscription. Each has an owner who chose the hardware and put the software on it. When a model runs on a box you own, you are the one who gives the orders: which model, which data, what it is allowed to touch, when it is switched off.
That is what agency means here: you decide what the machine does, and you need nobody’s permission to do it.
What owning the machine does not settle
Three cautions, from our own record.
A box on your desk can still talk to the internet. We measured ours. The models sent nothing out. The software that runs them was calling its vendor every four hours, and it stopped only when we found the setting and checked again. The measurement is here. If nobody has watched the wire, “it’s local” is a belief.
Rented models are better at most things. One of the evening’s panel partners was Claude, a model you rent, and that is no contradiction. Our own results say the same: local hardware earns its place for data that cannot go out, and for narrow jobs cut into small steps. It is a constraint we accept, and we publish what it costs, failures included.
The weak point is usually a person. One panellist, speaking about security, said most attacks now arrive through employees, and that everyone in the room could be hacked. Hardware you own does not change that.
So our answer to the organiser’s question is both: rent for work that others may see, own the machine for work they may not, and measure what it sends.
Thanks to Vicente and the Bridge33 team for the table.
The quoted lines are from the organiser’s public post of 19 November 2026. Remarks from the panels and from the table are from notes taken on the night and are not quotations. The organisers said beforehand that a write-up was welcome. They have not reviewed this text.