For decades, we believed the valuable asset was software.
Today, many assume the valuable asset is the model itself.
I believe both are becoming secondary.
The real strategic asset is the factory that continuously produces intelligence.
That factory is no longer made of steel, conveyor belts and assembly lines.
It is made of GPU clusters.
Inference infrastructure.
Evaluation systems.
Synthetic data pipelines.
Continuous training.
Safety engineering.
Operational feedback.
Deployment telemetry.
The model is a product.
The factory is the capability.
Industrial history has taught this lesson before.
The companies that transformed the world were rarely defined by a single product.
They were defined by production systems.
Toyota was never just a car.
TSMC is not simply a semiconductor.
Amazon is not merely a retailer.
Their competitive advantage lies in the systems that continuously produce value.
Artificial intelligence is entering the same phase.
Every inference request produces more than an answer.
It generates signals.
Usage patterns.
Evaluation data.
Operational feedback.
In some cases, entirely new training material.
Viewed individually, these outputs appear insignificant.
Viewed collectively, they become industrial knowledge.
The factory improves itself while it is operating.
This changes how we should think about intellectual property.
For decades, we protected source code.
Then we protected trained models.
Increasingly, the capability that deserves protection is neither.
It is the production system that continuously creates and refines intelligence.
The value no longer resides only inside the model.
It resides inside the operation.
This has important implications for digital sovereignty.
Sovereignty has never meant owning every product.
It has meant controlling the means of production.
During previous industrial revolutions that meant railways.
Steelworks.
Power plants.
Semiconductor fabrication.
Today it increasingly means compute infrastructure.
Inference platforms.
Evaluation systems.
Data governance.
Operational ownership.
A nation that depends entirely on external intelligence factories becomes strategically dependent, regardless of where the software itself was developed.
The lesson extends beyond governments.
Every organisation deploying AI should ask a different question.
Not:
"Which model should we buy?"
Instead:
"What capabilities are we building that remain ours?"
Models will continue to improve.
Benchmarks will continue to change.
Vendors will continue to compete.
Operational capability compounds.
The organisations that create lasting value will not necessarily own the largest models.
They will own the strongest production systems around them.
Just as factories transformed raw materials into products,
AI factories transform computation into knowledge.
Understanding that distinction may become one of the defining strategic questions of this decade.
We often hear that AI is the new electricity.
I think the comparison misses something important.
Electricity powers factories.
Artificial intelligence is becoming the factory itself.
And throughout history, societies that understood how to build, operate and protect their factories rarely depended on those that did not.