AI is not technology. It is sovereignty. (And almost no one is talking about it.)

Public conversation is still obsessed with models, GPUs, and productivity.

But the real strategic risk of AI has nothing to do with software.

It’s about who can turn it off.

A country that does not control its compute, energy, critical materials, cables, cloud, and cryptography does not own AI. It rents it. And that rental can be revoked.

This is no longer theoretical:

• China placed seven families of rare earth elements under export licensing - materials essential for magnets and chips.

• The United States banned Chinese magnets in defense and is financing massive critical‑materials projects.

• Saudi Arabia and the UAE are building 500 MW to 1 GW AI clusters.

• The European Union launched its own sovereign‑cloud architecture.

• The Department of Defense (DoD) accounts for 98.9% of projected U.S. federal AI spending in 2026.

The first wave of AI is not about productivity.

It’s about defense, control, and operational continuity.

And here’s the point almost everyone is missing:

The Sovereign AI Stack: five layers tied together by a single risk:

• Compute

• Power

• Materials

• Bandwidth

• Trust (cryptography, jurisdiction, control plane)

All five share the same vulnerability: the possibility of being shut down from abroad — the FOREIGN KILL SWITCH.

If any layer depends on a foreign actor, AI can be interrupted by:

• a president (we’re seeing this now)

• a regulator

• a court

• a cloud provider

• an export‑control authority

Markets are pricing infrastructure.

They are not pricing sovereignty.

That is the strategic gap for 2027–2032.

For Latin America, this raises an uncomfortable question:

Are we building capability… or dependency?

AI is not a strategy.

It is an accelerant. 

And where it accelerates you depends entirely on what lies underneath.