THE CHIP WAR ISN'T REALLY ABOUT CHIPS

Open ChatGPT. Generate an image. Translate a sentence.

Ask an AI to write code.

Create a video.

None of it looks particularly geopolitical.

It feels like software.

But underneath the software is hardware.

And underneath that hardware is an extraordinary industrial system involving semiconductor designers, chip manufacturers, equipment companies, materials suppliers, energy infrastructure and data centres.

That is why the U.S.–China technology conflict is much bigger than a dispute over individual chips.

The chip is simply the visible object.

The real battle is over computing capacity.

**FOLLOW THE CHIP BACKWARDS**

Take an advanced AI processor.

It begins as a design.

Then the design needs to be manufactured.

That requires extremely sophisticated semiconductor equipment.

The manufacturing process requires specialised materials. The resulting wafer has to be processed, tested and packaged.

The finished processor goes into servers. Those servers go into data centres.

The data centres need enormous amounts of electricity and cooling.

Then software turns all of that hardware into usable computing.

So the real chain looks something like:

AI application ↓ AI model ↓ software stack ↓ data centre ↓ AI processors ↓ advanced packaging ↓ semiconductor fabrication ↓ manufacturing equipment ↓ materials ↓ energy

Now the geopolitical picture looks very different.

**THE UNITED STATES HAS CHOKE POINTS**

America has enormous advantages in semiconductor design, AI software and technology companies.

Companies such as Nvidia have become central to the AI computing boom.

The U.S. government has also used export controls to restrict China's access to certain advanced computing technology and semiconductor manufacturing capabilities.

The policy is explicitly connected to national-security concerns.

U.S. export-control rules cover advanced computing items and semiconductor manufacturing equipment under specified conditions.

This creates a powerful form of leverage.

If a country cannot easily obtain the most advanced computing hardware, building frontier AI becomes harder.

But there is an obvious problem.

Restrictions also create incentives for the restricted country to build its own alternative.

And China has been doing exactly that.

**CHINA'S ANSWER: BUILD THE STACK**

Huawei has increasingly positioned itself around domestic AI computing.

Reuters reported in September that Huawei's AI computing demand in China was exceeding its current production capacity.

The company is accelerating development of new Ascend processors and is working on systems that connect huge numbers of processors together.

Alibaba is also pursuing a vertically integrated AI strategy.

In September, it announced a new AI chip and plans for a much larger AI model, while describing an ambition spanning models, semiconductors and data centres.

This is important because China doesn't necessarily need to reproduce every American technology component exactly.

It can attempt to construct an alternative architecture.

That is a completely different strategy.

**THIS IS WHY "WHO IS WINNING AI?" IS A BAD QUESTION**

Suppose America has the strongest AI models. China has enormous manufacturing capacity.

America has stronger semiconductor design.

China has greater control over portions of the critical-mineral processing chain. America has Nvidia.

China has Huawei and a growing domestic chip ecosystem.

Taiwan has leading-edge manufacturing. Japan has important industrial and materials capabilities.

South Korea has major memory-chip manufacturers.

Europe has crucial semiconductor equipment expertise.

India is attempting to build more manufacturing capacity.

So who is "winning"?

The answer depends on what you're measuring.

Model performance? Manufacturing? Energy? Chip design? Supply-chain control? AI deployment? Robotics? Data centres?

The question itself hides the complexity.

Reuters recently described the U.S. and Chinese AI strategies as fundamentally different, with China's approach placing substantial emphasis on embedding AI throughout society while dealing with constraints created by chip access and export controls.