Rent the Studio. Own the Masters.

I just spent $47 teaching a machine to write like me.

Not with me. Not for me. Like me. I took roughly 600 of my own blog posts — everything I've published over the years — and fine-tuned a small open model on them. The result is a version of an AI that's read all my writing and can draft in my voice. Not a chatbot that sounds vaguely professional. A me-shaped tool.

And before anyone asks: yes, I rented cloud compute to do it. A few hours on a GPU, less than the cost of dinner. The compute is a commodity. The part I won't rent is the part that's actually me.

The Trade You're Already Making

Every AI you use today — ChatGPT, Claude, Copilot, whatever's embedded in your workflow — was trained to sound like everyone. They're generic by design. Articulate, helpful, and utterly indistinguishable from each other in tone. You feed them your ideas, they hand back something that sounds like a slightly better version of a corporate memo.

That's the trade: convenience in exchange for sameness.

Here's what nobody's saying out loud: you're training these models every time you use them. Not just sending queries — feeding them examples of how you think, what you value, how you solve problems. And in most cases, you're handing that training data to someone else's platform, where it either disappears into their next model iteration or sits in a database you'll never touch again.

Musicians learned this the hard way.

The Ones Who Gave Up Their Masters

In the 1950s and 60s, record labels owned the master recordings. Artists — even the famous ones — signed contracts that gave away the tapes in exchange for studio access and distribution. The studio had the equipment. The label had the reach. The artist just wanted to make music.

Decades later, those same artists watched other people get rich off their work. The label could reissue, remix, license to films, sell to streaming platforms. The artist who actually created the music? They got whatever their original contract said, if anything.

The ones who kept their masters — or fought to buy them back — controlled their own catalogs. They decided when and how their work got used. The studio was rentable. The masters were the only thing that mattered.

We're watching the same pattern play out in AI, just faster.

What You Actually Own vs. What You Rent

Here's the distinction that matters:

Renting intelligence means you pay by the token, forever. You send a prompt, get a response, maybe refine it a few times. It's fast. It's cheap per query. And every single interaction evaporates unless you deliberately save it. You're borrowing capacity, not building an asset.

Owning the model means you have the weights — the actual trained parameters that encode patterns from your work. Once you've fine-tuned a model on your writing, your code, your decision-making frameworks, that model is the distilled version of how you think. You can run it locally. You can improve it. You can choose whether to share it or keep it private.

One is a subscription. The other is an asset.

I'm not saying subscriptions are bad. I rent cloud compute all day. Compute is infrastructure — you use it when you need it, you don't when you don't. But the model trained on my work? That's not infrastructure. That's the master tape.

What It Actually Takes (Less Than You Think)

The technical barrier to doing this has collapsed.

I used an open-source model (Llama, if you care), a dataset of my own blog posts in plain text, and a training script I found on GitHub. The cloud GPU rental cost $47. Training took a few hours. The result now runs on a $300 gaming card in my basement, and every draft after the initial training costs roughly nothing.

The training bill is a rounding error. The ongoing cost is zero.

Compare that to paying $20–$200/month for AI subscriptions, indefinitely, to access intelligence that was never trained on how you specifically think. You're renting generic smarts when you could own a version that's actually tuned to your voice, your domain, your patterns.

And yes, full transparency: I spent a day teaching a machine to write like me so I could write less. Its first halfway-decent draft was a LinkedIn post. I see where this is going.

The Uncomfortable Question

Here's what I keep coming back to:

If you can rent intelligence by the token forever, what's the one thing you'd actually want to own?

Not the compute. Not the interface. Not the brand name of the model.

The thing worth owning is the version that learned from your work — the one that encodes your expertise, your judgment, your voice. That's the only part that isn't a commodity.

And yet most people are handing that exact data to platforms they don't control, in exchange for convenience. I'm not saying that's wrong. I'm saying it's a trade, and most people haven't realized they're making it.

What This Means Monday Morning

You don't need to fine-tune a model this week. But you should be asking:

Where is your institutional knowledge going? If your team is using AI tools to draft memos, generate reports, or analyze data, are you capturing those patterns — or just renting intelligence that resets every session?
What would it cost to own instead of rent? For most professional use cases, the compute cost is trivial. The real question is whether you have the technical literacy in-house to even know what's possible.
What happens when the rental price changes? AI platforms are cheap right now because they're competing for market share. The ones who win will raise prices. If you've built your entire workflow on rented intelligence, you're in the same position as the artist who gave up the masters.