Sarel Esterhuizen

Software leader by day, code monkey by night. Distracted by drumming, botany, mycology, brewing, distilling, & rocketry. Amsterdam based but Africa in my blood.

[Sarel Esterhuizen @Esterhuizen]

TLDR: Watch till end 🤕

Wanted to mess around with the latest AI video and voice models, so I built a proof of concept pipeline for something I’ve wanted to exist for a long time.

What happens when you stop asking historians what happened…

…and start asking history what it thinks of the world we’ve built?

Here’s a first look.

Coming soon (maybe).

[Sarel Esterhuizen @Esterhuizen]

AI demand can be real while AI investments are still overpriced.

Railways changed the world and bankrupted plenty of investors.

The same may happen with data centres: foundational infrastructure, weak individual economics and a lot of capital discovering too late that strategic importance does not guarantee returns.

[Sarel Esterhuizen @Esterhuizen]

AI creates a strange loop.

It requires enormous amounts of reliable electricity.

It also gives attackers better tools to disrupt the grid supplying it.

The strategic advantage will not belong merely to countries with more power. It will belong to those that can digitise infrastructure without making it fragile.

[Sarel Esterhuizen @Esterhuizen]

Iran’s leverage is not that it can defeat the US.

It is that it can make escalation intolerably expensive for everyone around it.

That gives Gulf states unusual influence over Washington.

Shared vulnerability can restrain a superpower more effectively than formal diplomacy.

[Sarel Esterhuizen @Esterhuizen]

Japan did not dominate the consumer internet.

But AI is becoming an industrial buildout: chips, machinery, materials, robotics and power systems.

That changes the map.

The countries that looked slow in the software era may prove indispensable in the physical one.

[Sarel Esterhuizen @Esterhuizen]

The West keeps telling China to consume more and manufacture less.

Beijing keeps choosing factories.

That may be economically unbalanced. It is not necessarily irrational.

Consumption raises living standards. Industrial capacity creates leverage, resilience and options in a crisis. China has decided which matters more.

[Sarel Esterhuizen @Esterhuizen]

Europe’s AI Act becomes real today.

The usual argument is regulation versus innovation. The harder question is whether legal power can be converted into industrial power.

Europe can govern AI inside its market. But can it build the chips, energy, compute and companies the market depends on?

[Sarel Esterhuizen @Esterhuizen]

Iran does not need to defeat the US militarily to gain leverage.

It needs to make escalation expensive for everyone else.

Hormuz turns geography into coercion: allies, importers and markets become indirect pressure on Washington.

Interdependence is not the opposite of power. It is often its transmission system.

[Sarel Esterhuizen @Esterhuizen]

AI is no longer just consuming chips. It is beginning to decide what else gets built.

Apple’s supply warning is a glimpse of the next phase: data centres competing with phones, cars and industry for the same physical capacity.

The AI race is becoming an allocation problem.

[Sarel Esterhuizen @Esterhuizen]

Amazon plans to spend $220bn this year while still expecting AI capacity shortages.

At some point, “Big Tech” stops being a useful description.

These firms increasingly resemble industrial states: securing energy, designing chips, allocating compute and deciding who gets access to intelligence.

[Sarel Esterhuizen @Esterhuizen]

The important detail in the latest AI hacking incident is not that the model became “rogue.”

It is that humans believed they had built a sandbox… and had not.

AI safety will increasingly look less like philosophy and more like permissions, isolation, audit trails and competent operations.

[Noam Brown @polynoamial]

An internal version of Astra, @OpenAI’s next major model family, solved 10 major open problems in mathematics, quantum complexity, and theoretical computer science.

We believe it will be a major step for scientific reasoning. https://t.co/iP6cyheZ7i

[Sarel Esterhuizen @Esterhuizen]

Europe’s AI problem is ultimately execution:

Faster permits.
Cheaper energy.
Concentrated bets.
Real customers.
Leaders willing to close weak programmes.

Industrial policy without selection eventually becomes subsidy theatre.

[Sarel Esterhuizen @Esterhuizen]

Europe’s AI debate is framed as regulation versus innovation. Wrong battlefield. The real contest is whether states can convert capital into chips, power, data centres and deployed systems. Rules shape markets. Industrial capacity decides who leads them.

[Sarel Esterhuizen @Esterhuizen]

The useful historical parallel is not the early internet.

It is electrification.

The winners were not merely the countries that understood the technology. They reorganised infrastructure, capital, regulation and industry around it.

AI will demand something similar.