Own Your AI is a nine-part Top Tool Stack series on running your own local AI instead of renting it from the giants. This is Part 2 of 9. See the whole series →
Last week we introduced the idea at the heart of this series: owning your AI instead of renting it. This week, the honest case for reducing your dependence on OpenAI, Anthropic and Google, including the parts that are inconvenient for the “just self-host everything” crowd. Because a good argument survives its own counter-argument, and we would rather you made this decision with clear eyes than with a grievance.
The five reasons to de-risk
1. The price only goes one way. The frontier labs are collectively losing enormous sums with the stated intention of earning it back later. “Later” means from you. Subscriptions creep, free tiers shrink, the good model moves behind the more expensive plan, and usage gets metered. None of that is a conspiracy; it is simply what happens when the introductory pricing on a capital-intensive product ends. Owning a capable local model caps that exposure.
2. Your data is the product, even when you pay. Cloud AI means your conversations, documents and half-formed ideas travel to someone else’s servers and live under someone else’s policy. Those policies change, retention periods get extended by litigation you are not party to, and “delete” does not always mean deleted. Running a model locally is the only version of privacy that does not depend on trusting a corporation’s current mood. Nothing leaves the machine.
3. Rug-pulls are routine. Cloud models get deprecated on the vendor’s schedule, not yours. The model you built a workflow around can be retired, quietly altered, or have its behaviour “improved” in ways that break what you had. An open model you have downloaded runs the same next year as it does today, because it lives on your disk, not on their roadmap.
4. Outages and rate limits. When the whole internet leans on a handful of endpoints, those endpoints go down, slow to a crawl at peak times, or refuse you when demand spikes. A local model has no queue and no maintenance window. It is there at 3am on launch day when the cloud is on its knees.
5. Lock-in and drift. The more of your workflow, memory and tooling lives inside one vendor’s walled garden, the harder it is to ever leave, and the more comfortable that vendor can get with a slowly worsening product. Open, local tooling keeps the exit door open, which is the single best pressure you can put on any service you use.
Now the part the self-host evangelists skip
Here is the honest counterweight, and it matters. Local models still lose to the frontier on the hardest problems. For deep reasoning, long complex coding, and genuinely novel analysis, a paid Claude, GPT or Gemini subscription is often simply better, sometimes by a wide margin, and pretending otherwise helps nobody. Owning is not free, either. You pay in hardware, electricity, setup time and maintenance, and you become your own IT and security department. For a lot of people, twenty dollars a month for a frontier model is the rational choice, full stop.
So this series is not an argument for deleting your subscriptions and retreating to a bunker. It is an argument for optionality. Run local models for the large share of everyday work they now handle perfectly well: summarising, drafting, answering questions about your own documents, routine coding help. Keep a cloud model on hand for the genuinely hard jobs if you want one. The person with both is more resilient, more private and, over time, cheaper than the person who rents everything.
The goal is not independence for its own sake. It is to stop being a captive customer of three companies whose interests and yours are only sometimes aligned. Next week we get practical: how a complete beginner runs their first local model in about twenty minutes, on hardware they already own.
The AI tool stack actually worth paying for
One email a week. The tools, models and moves that matter, minus the hype and the horseshit filter set to maximum. Free.