Swallow the Frog

There’s a number from Sequoia Capital that should make every C-Suite executive choke on their morning flat white. The gap between what the AI industry is spending on infrastructure and the revenue it’s actually generating? $600 billion. By 2030, that requirement climbs to $2 trillion. And that changes everything — including your job security.

The Snickers Problem

You know that feeling when you open a Snickers and it’s suspiciously small? That’s not your imagination. It’s shrinkflation — manufacturers quietly reducing what you get while charging you the same (or more).

AI is doing exactly the same thing.

OpenAI, Anthropic, Google — they’re all staring at the same fiscal cliff. Keeping up with demand costs extraordinary amounts of money. The model that seemed magically affordable eighteen months ago is quietly becoming more expensive. Features are disappearing behind higher-tier paywalls. Usage limits are tightening. The era of cheap, unlimited AI was always going to be a land grab. Now the bill is coming due.

The Awkward Conclusion Nobody Wants to Say

Here’s the thing nobody in the AI industry wants to say out loud: as AI infrastructure costs spiral, human labour starts to look like the budget option.

That’s a strange sentence to write. But follow the logic.

If a GPU cluster costs more per task than a person, the C-Suite does what C-Suites always do — they optimise for cost. The AI that was supposed to make people redundant suddenly needs to justify its own price tag. The math stops working in AI’s favour.

Congratulations. You’re officially more cost-effective than a data centre. 🥂

So What Do You Actually Do With This?

The answer isn’t to ignore AI — that ship has sailed. It’s to stop renting it and start owning it.

Right now, I’m running AI on a 2017 MacBook Pro. It is not writing Nobel Prize-winning literature. But it’s doing the small, repetitive jobs I don’t want to do — drafting, summarising, sorting, classifying — for the cost of a wall socket.

No subscription. No per-token pricing. No data leaving the building.

Think of it like the move to buy local fruit and veg. The supermarket chain (OpenAI, Google, Anthropic) is convenient and scalable. But the local market — Ollama, Qwen, Mistral — gives you something the chains can’t: ownership, privacy, and zero ongoing cost.

Local AI in Practice

Running a local model isn’t the technical nightmare it used to sound like. Tools like Ollama have made it genuinely accessible — install it, pull a model, and you’re running a private AI assistant on your own hardware within minutes.

Is it as powerful as GPT-4o on raw capability? On some tasks, no. But for the jobs that eat your team’s time — summarising meeting notes, drafting first versions of emails, categorising data, generating report templates — it does the job. Quietly. Privately. For free.

At Swallow The Frog, we’ve been running local AI implementations for clients in sectors where cloud AI simply isn’t an option: finance, healthcare, education. The student data doesn’t leave the school. The client information doesn’t touch a third-party server. The compliance officer doesn’t have a heart attack.

That’s not a limitation. That’s the whole point.

The Practical Upshot

If you’re currently paying for AI subscriptions you’re not fully using — or you’re avoiding AI entirely because you can’t send sensitive data to a cloud provider — there’s a third path.

Stop renting. Start owning.

The infrastructure cost crisis at the top of the AI market is your opportunity at the bottom. While the big players figure out how to monetise their way out of a $2 trillion hole, you can be quietly running your own models, on your own hardware, with zero ongoing cost.

Your job is safe. Probably. But more importantly, your data is private. Definitely.

---

Curious what local AI could look like for your business? Book a free 30-minute Frog-Check and we’ll tell you exactly where to start.