A few years ago, I was having dinner with the Americas VP of a European energy supermajor — one of those companies that extracts oil from war zones, negotiates with regimes that don’t appear on polite lists, and operates in places where “political risk” means your assets might get nationalized or your personnel kidnapped.
Seventy-plus countries. Active operations in Libya, Nigeria, Angola, Myanmar, Yemen. The kinds of places where security briefings come before breakfast.
Mid-conversation, he paused and told me: “Last week we were in a meeting with our global CEO, and I felt validated — because he finally said out loud what I’d been thinking for years: Brazil is very likely the most challenging country to do business where we operate today.”
Not the Niger Delta. Not post-sanctions Russia. Not the parts of the Middle East that are actively on fire.
Brazil.
I didn’t feel surprised. I felt seen.
Being Brazilian-born gives me a perspective that my American friends often struggle to internalize. When I share certain observations, the conversation almost always lands on the same realization: there are things you take for granted that simply aren’t.
I’m not interested in politics here. This isn’t opinion — it’s arithmetic.
In Brazil, roughly 40% of GDP flows through government expenditure. Not through industrialists. Not through entrepreneurs. Through the state. And if you count the growing annual deficit, that number creeps higher.
What does this mean for someone trying to build?
It means your largest potential customer is also your regulator, your tax authority, your licensing body, and your competitor’s sponsor. It means payment terms are dictated by political cycles. It means a contract signed in one administration may be “reviewed” in the next. It means the infrastructure you depend on — power, telecom, courts — operates under the same incentive structure.
This isn’t corruption in the Hollywood sense. It’s something more systemic: an economy where the center of gravity is not the market, but the machinery of government itself.
When Americans think about “doing business in a tough market,” they imagine red tape, maybe some bribes, perhaps unreliable partners. They don’t imagine an environment where the fundamental rules of enterprise — that you build something, sell it, get paid, and reinvest — are subordinate to a parallel economy that runs on different logic entirely.
It would be funny if it weren’t tragic.
I can’t count how many investment decisions I witnessed — mine and others’ — that ended with the same phrase: “Let’s see how it goes. Next year is an election year.”
This dominates all business expectations on the horizon. Not market conditions. Not competitive pressure. Not technology shifts. The electoral calendar.
And here’s the dark joke: between municipal, state, and federal elections, there is always an election year coming. The “wait and see” window never closes. The horizon never clears.
You learn to build anyway. But you build differently. You build defensively. You over-engineer for survival, not growth. You assume the rules will change, because they will. You keep options open that an American founder would consider paranoid.
Twenty years of that reshapes how you think about infrastructure, about resilience, about what “sovereignty” actually means when the ground never stops shifting.
So what is this “Sovereign AI” that everyone’s talking about?
I’ve been watching the discourse closely. I’ve seen the keynotes, the policy papers, the national strategies. I’ve also been watching what’s actually happening — and the gap is telling.
The United Arab Emirates isn’t writing white papers. They’re writing checks. Full throttle into AI infrastructure — data centers, sovereign wealth fund investments, partnerships structured for technology transfer, not just procurement. They looked at their oil-dependent GDP and decided that compute is the new reserve currency. Agree or disagree with the approach, but there’s no ambiguity about the intent.
Then there’s Taiwan.
Taiwan’s semiconductor dominance isn’t a trade policy. It’s a survival strategy. TSMC’s reported “silicon shield” — the understanding that their fabs could be rendered inoperable in an invasion scenario — isn’t a contingency plan. It’s a form of sovereignty more honest than any declaration. Taiwan’s existence as a nation-state is inseparable from the fact that the world economy cannot function without what they produce.
That’s not sovereignty by policy. That’s sovereignty by leverage.
I learned this early.
For years, I built edge computing infrastructure for companies operating outside the US — enterprises with national security sensitivity who couldn’t rely on standard SLAs. Not because the technology was inadequate. Because the Patriot Act explicitly permitted US agencies to access their data held by American companies. No warrant required that they’d ever see. No notification. No recourse.
This wasn’t paranoia. It was legal reality.
And before anyone dismisses this as anti-American sentiment from a resentful foreign competitor — I’m a US green card holder. I chose to build part of my life here. I’m saying this because it’s true, and because the selective amnesia in the current discourse is staggering.
Want to see the asymmetry in action? Look at GDPR.
GDPR didn’t emerge from some European instinct for privacy idealism. It was born from Snowden. From the revelation that US intelligence had direct access to data flowing through American platforms — including data belonging to European citizens, companies, and governments.
GDPR wasn’t a polite request. It was legislation that said: We need to make sure this doesn’t happen again.
Now look at the other side.
TikTok is being forced to divest its US operations or face a ban. The stated reason? A foreign adversary might access American user data. Huawei is effectively banned from US telecom infrastructure. The concern? Backdoors. Espionage. National security.
These concerns are legitimate. I’m not disputing them.
But let’s be honest: the United States has been doing to others exactly what it fears others will do to it. The difference is that when America does it, it’s called “national security.” When others try to protect themselves from it, it’s called “protectionism” or “fragmenting the internet.”
This isn’t whataboutism. It’s the foundation you need to understand if you want to have an honest conversation about Sovereign AI.
So — what is Sovereign AI?
First, let’s acknowledge what it’s up against.
If you’re an American company trying to export a simple WiFi access point, you need to select a specific part number that excludes certain cryptographic capabilities. Not for export to Iran. Not to Russia. To allies. Commercial partners. Countries where Americans vacation and do business without a second thought.
That’s for a router.
Now consider AI. GPUs. Training infrastructure. Model weights. Inference optimization. The entire stack is saturated with dual-use implications — and nobody has figured out the compliance framework.
Right now, our government is in a real-time confrontation with China over NVIDIA chips. Should compliance be verified at the point of shipping? At arrival? What about chips resold through third countries? What about cloud compute sold by the hour — does sovereignty apply to the silicon, the data center, or the API endpoint?
Nobody knows.
The rules changed last quarter. They may change again next quarter. Export licenses granted under one administration are reviewed under the next. Companies are building strategies on sand.
The geopolitical map isn’t just shifting — it’s being redrawn while we’re standing on it.
Want proof? Look at NVIDIA’s numbers.
China’s share of NVIDIA’s revenue dropped from 21% to 13% between fiscal 2023 and fiscal 2025. Jensen Huang said it publicly: revenue from China fell to half of pre-export control levels. A $5.5 billion charge in Q1 2025 for H20 chips they could no longer ship. Another $2.5 billion in revenue they simply weren’t allowed to collect. An estimated $8 billion loss projected for the following quarter.And here’s the part that should terrify anyone trying to plan a Sovereign AI strategy: the rules kept changing.
The Biden administration imposed controls in 2022. Tightened them in 2023. NVIDIA designed the H20 specifically to comply — a deliberately hobbled chip, slower interconnects, reduced bandwidth. Legal. Compliant. Then in April 2025, the Trump administration told them they’d need a license anyway. By December, Trump reversed course — H200 exports to China were suddenly permitted again, but with a 25% surcharge and the U.S. government taking a cut.
In the span of eight months, companies watched the rules toggle from “banned” to “compliant” to “banned again” to “allowed with tribute.”
This is the environment in which nations are supposed to build sovereign AI infrastructure?
I started writing this article with certainty. Twenty years of building infrastructure in impossible conditions — I thought I had a clear framework to offer.
I don’t.
The more I’ve written, the more I’ve realized: there are too many variables, moving too fast, with too many actors rewriting the rules in real time. Export controls that toggle quarterly. Alliances that shift with elections. Supply chains that route through Singapore one year and get investigated the next. Technology that obsoletes faster than policy can adapt.
What I can say is this: we’re looking at a market that’s likely worth half a trillion dollars within the decade. And right now, it’s in the grab-the-land phase.
No clear boundaries. No settled rules. No established winners.
Just nations, corporations, and sovereign wealth funds planting flags as fast as they can — some with genuine capability, most with PowerPoint decks and press releases.
I’ve seen this before. Not at this scale, but the pattern is familiar. When the rules aren’t written yet, the people who build real things — infrastructure that actually runs, talent that actually stays, systems that actually survive — end up owning the territory. Everyone else ends up with expensive lessons.
In my opinion, there is no clear answer — because it hasn’t been written yet.
This isn’t a policy problem with a policy solution. It’s not a technology problem with a technology solution. It’s a moment of genuine historical discontinuity, and the playbooks don’t exist because the game has never been played.
Those who realize that the map is not the territory — who understand that the frameworks, the forecasts, the five-year strategies are all approximations of a reality that’s being invented in real time — they’re the ones who will navigate through.
And those who can actually build in the fog? Who can maintain uptime when the rules change, keep talent when the incentives shift, deliver capability when the supply chains fracture?
They’ll end up as the Founding Fathers of what comes next.
Not the consultants. Not the policy advisors. Not the keynote speakers.
The builders.
What’s coming isn’t just another technology cycle. It’s likely the most significant advance since the first Industrial Revolution — or, perhaps, in all of human history.
I spent twenty years building infrastructure in one of the hardest markets on Earth. I thought that prepared me to have answers.
It didn’t. It prepared me to operate without them.