Shall you home brew your AI?
On the owning vs. renting tradeoffs
Let me take you back to where this started: our work Slack channel called #ai-news was blowing up. Too many links. Too many hot takes. So we created a quieter space: #ai-discourse-australia. Lately, there’s been a lot of talk in Australia about AI approaches and local companies launching “sovereign AI” projects.
The government is trying to decide: should we make strict rules for AI, or let it grow freely so it helps productivity? I posted some thoughts on LinkedIn, and it kicked off great conversations.
Why does this matter? If you’re leading a company, shaping policy, or just trying to use AI responsibly, this debate affects you.
Should we build our own AI systems? Or keep using tools built elsewhere?
It’s a hard question.
SIGNAL
Australia’s Computer Society wants the country to invest $2–4 billion in sovereign AI. Meanwhile, Switzerland launched its own large language model, Apertus. It’s open-source and trained using green electricity, with sustainable cooling systems that use Lake Lugano. It also follows Europe’s strict privacy laws and the EU AI Act. Microsoft, on the other hand, is setting up 25,000 advanced AI chips in France. To give you a sense of comparison, Apertus used 4,000.
While Australia debates, tech giants are spending $100 billion every 6 months on AI infrastructure. And by 2027, at least 25 countries are expected to launch their own AI models.
If a country has the resources and strong leadership, building local AI might make sense. But here’s the key question:
Will these efforts create useful, competitive systems or just expensive, feel-good projects?
The market usually doesn’t give out points for good intentions.
STORY
A few weeks ago, I posted a simple cheat sheet on LinkedIn highlighting the most common trade-offs associated with “sovereign AI.”
I was surprised by the incredibly thoughtful and illuminating discussions that followed.
I spoke with experts who’ve built and studied business cases for sovereign cloud systems. They said it always goes the same way: governments pay for it, companies say they want it, but no one wants to pay the necessary premium when it’s built.
Their point was clear: if a cloud system costs 25% more, people won’t use it.
But maybe comparing cloud to AI may not tell the whole story.
There’s more to sovereign AI than just beating ChatGPT. It’s a coordinated effort to ensure that your country’s core systems—such as healthcare or the courts—don’t suddenly change because a foreign company updates its model.
AI is unlike any software or infrastructure. It is a complex supply chain. It reflects values and assumptions and has a profound impact on society. When Singapore created SEA-LION to avoid Western bias, or Switzerland made sure Apertus respected European law, it wasn’t about catching up. It was about control and fostering innovation that doesn’t depend on others’ worldviews.
Still, there are tradeoffs:
Innovation lag – you’re 1–2 years behind global leaders
Security risks – being local doesn’t always mean safer
Opportunity costs – focusing on rules might slow real progress
So the question isn’t: is sovereign AI cheaper?
Maybe the better question is: Is depending on others more expensive in the long run?
Let’s look at what to do next.
THE HUMAN OVERRIDE
8 Things to Consider Before Investing in Sovereign AI
Whether you’re a company leader or part of a government team, here’s how to make wise decisions.
0. Know the Costs
Local AI can cost 60–200% more
Expect delays: 1–2 years behind big players
Most people won’t pay a 25% premium
Being local doesn’t automatically mean safer
1. Start with the Data
Think about laws first:
Banking? Think privacy.
Health? Think patient protection.
Government? Think surveillance.
Switzerland made Apertus work with just public data. That’s enough in many cases.
2. Don’t Overbuild
Training huge models like GPT-4 can cost $100 million. Most needs are fine with:
8 billion parameters: for business use
70 billion: for more complex tasks
3. Run the Payment Test
Ask early:
Will users pay more?
Wait longer for updates?
Trade some performance for control?
Usually, the answer is no.
4. Build Partnerships
Going solo is tough. Look at these examples built by coordinated efforts between academia, industry, and, in some cases, governments:
JAIS (Arabic language model)
Gaia-X (European data system)
Apertus (Switzerland)
5. Make Rules Work for You
Turn regulation into a strength:
Built-in privacy (GDPR)
Local data storage
Transparent training data
6. Focus Your Efforts
Don’t try to match OpenAI’s speed. Instead:
Go open-source
Specialize in a topic
Make it easy to connect with global tools
7. Local ≠ Automatically Safe
Sovereign AI can mean:
Slower updates
More risk of cyber attacks
Weaker monitoring
So plan for security from the start. Collaborate with local and global AI security companies and red teamers.
8. Measure the Right Stuff
Look beyond just tech profits:
Are you growing talent?
Keeping more value in-country?
Building trade strength?
Creating local innovation?
Reality Check:
Start small. Pick real problems. Use shared infrastructure. Sovereignty isn’t about doing everything alone—it’s about doing the things others won’t.
SPARK
What happens when you rely on someone else’s AI?
Think back to 1973, when oil exports were suddenly cut. Countries learned that relying too much on others for energy could be dangerous.
Today, AI is becoming just as essential.
Countries like China, Switzerland, and many in Europe are building their own models. Even small players are finding ways to do it by using open-source tools like Hugging Face and new techniques like fine-tuning.
Australia is having this debate now. Should we buy cheaper tools from tech giants or build systems we control? There’s no one-size-fits-all answer. But thanks to open-source platforms and smarter strategies, local AI is more doable than ever.
Dr. Alberto Chierici helps leaders calibrate their trust in AI and founders to build responsibly. Former Tesla, 3x founder, PhD in Computer Science. Thanks for reading. If this helped you, share it with someone making decisions about AI.