BrokeIt - Daily AI News · All episodes: ↗

PC Local AI, Singapore's Sovereign Push, Inference Data Risk

2026-11-22 · 6 min

Listen · Apple Podcasts Listen · Spotify

Stories covered

Transcript

Intro

Ivy: Mistral AI is betting its whole Asia expansion on something called 'sovereign AI'.

Marcus: I'm Marcus.

Ivy: And I'm Ivy.

Marcus: And you're listening to Broke It for Sunday, November 22nd.

Ivy: We've got three stories today where the AI hype gets a dose of reality.

Marcus: First, a top European AI company makes a huge play in Asia. But are they building bridges, or just higher walls?

Ivy: Next, a new model claims you can finally run powerful AI on your home computer. Key word: 'claims'.

Marcus: And finally, a not-so-gentle reminder of what happens to your data when you use third-party AI. Spoiler: it's not good.

Mistral Bets on Sovereign AI with Singapore HQ

Marcus: Alright, first up: this 'sovereign AI' thing. French AI darling Mistral just announced a new HQ in Singapore to cover the Asia-Pacific market.

Ivy: And their big buzzword is 'sovereign AI'—the idea of making custom models for specific regions, governments, and big companies.

Marcus: It's brilliant! Models that get local context, languages, and rules. This is how you expand responsibly.

Ivy: Is it responsible? Or is it just marketing spin for fragmentation? Sounds like we're just building a bunch of smaller, walled-off AI gardens.

Marcus: I call it specialization. If a government in Asia wants an AI that keeps all its data inside its borders, Mistral can now sell it to them. That's a huge market.

Ivy: A market built on compliance, not performance. Are we really giving up the power of huge, global models just to check a box for some bureaucrat?

Marcus: But performance isn't just one number! A model that aces Thai legal docs is higher performance for a Thai law firm, even if it's not great at writing English poetry.

Ivy: Fair point. But this could easily create data silos and little nationalist AI kingdoms. That risk is very real.

Marcus: So for developers or businesses in Asia, this means you're about to get a powerful new AI option that's actually built for you.

Ivy: It also means you have to figure out which AI works where and under which rules. It's getting complicated.

Marcus: A trade-off I'd say is worth it. More players, more options, more competition. That's a win for everyone.

Inkling 2 Mini: This Is the Local AI That Finally Runs on a Normal PC

Ivy: Alright, our next story brings things from the national level down to your actual desktop. There’s a new model blowing up on Reddit called Inkling 2 Mini.

Marcus: Yes! This is what I've been waiting for. The promise is a truly useful AI that runs locally, on your own machine, no cloud needed.

Ivy: The headline is 'finally runs on a normal PC.' I'm going to need a definition of 'normal'.

Marcus: Okay, fine, it won't run on a ten-year-old laptop. Looks like you need a consumer GPU with around 12 to 16 gigs of VRAM.

Ivy: So, a high-end gaming PC. That's 'normal' for a very small, very dedicated slice of the population, Marcus.

Marcus: But it's a consumer card! Not some thirty-thousand-dollar server chip. This is a huge step for accessibility. A lot of people have these machines!

Ivy: And what about that other phrase: 'surprisingly powerful'? How powerful are we talking? Is it writing my code, or is it just a glorified Clippy?

Marcus: It's way beyond Clippy. People are using it for coding help, summarizing docs, drafting emails—all completely offline. The privacy win alone is massive.

Ivy: So, what this means for you is pretty simple: if you have one of those gaming PCs, you might be able to ditch the cloud for a lot of your daily AI tasks.

Marcus: It's the ultimate 'sovereign AI,' Ivy. The AI is sovereign to you.

Ivy: That's... an annoyingly good point.

Data Processing Incidents on Inference Providers

Marcus: And our last story is a perfect, if painful, example of why you'd want a local model. OpenRouter just published a post about some... 'data processing incidents'.

Ivy: 'Data processing incidents.' That's the most polished, corporate-speak I've ever heard for 'we screwed up and lost your data'.

Marcus: To be fair, it's not clear whose fault it is. OpenRouter just sends your prompts to other AI services. They're saying the 'incidents' happened at those third parties.

Ivy: That's the whole problem! Your data is bouncing between who knows how many systems. It only takes one weak link. And of course, the blog post is conveniently vague about what data was lost and how many people were affected.

Marcus: Look, at least they announced it. That's more than some companies would do. It's a wakeup call for the whole industry.

Ivy: It's a wakeup call for users. And what it means for you is brutally simple: don't paste sensitive information into a web-based AI. Ever. Full stop.

Marcus: Your private notes, your company's secret plans, your embarrassing search history...

Ivy: Just assume it could end up in a blog post titled 'a data processing incident.' Just don't do it.

Marcus: Alright, let's recap.

Ivy: Mistral is pushing 'sovereign AI' in Asia, for better or worse. And Inkling 2 Mini might actually make local AI a reality, if you've got a decent gaming PC.

Marcus: And we all got a harsh reminder: when you use a cloud AI, you're trusting them with your data. And sometimes, that trust is broken.

Ivy: Before we go... did you see that new AI-powered browser extension that replaces all online ads with pictures of capybaras?

Marcus: No! Is that real? It just finds the ad and puts in a capybara?

Ivy: Yep. It's the most useful thing I've seen AI do all year.

Marcus: That's our show! Thank you for listening to Broke It.

Ivy: We're back tomorrow. Until then, try not to have any of your own... 'data processing incidents'.

This show is made with AI: the hosts’ voices are synthetic and the scripts are AI-assisted. Every story links to its original source.