Nvidia is already around 10% below its peak. Google, Meta and Tesla are all under heavy pressure, with some sessions seeing declines of more than 2%. The S&P 500 is also feeling the impact.

The bigger story is what is happening to the AI trade — the enormous amount of investor money that has poured into companies connected to artificial intelligence and pushed their valuations to extraordinary levels.

That trade is now beginning to crack.

OpenAI and Anthropic are facing questions around their future IPO plans and valuations, while private valuations that were once approaching trillion-dollar territory are beginning to look increasingly fragile.

But what is causing this?

There are two layers to what is happening.

Neither is particularly subtle.

Layer One: China Has Changed the Cost of AI

Chinese AI models such as Kimi K3 and GLM-5.2 are now approaching the capabilities of some of the leading closed US AI models on important tests.

They are becoming competitive in areas such as coding, reasoning and handling complicated tasks that require the AI to work through a problem for a long time.

But the biggest difference may not be capability.

It is price.

Access to leading American AI models can cost individuals around $200 a year, while businesses can pay considerably more depending on their usage.

Chinese alternatives can cost a fraction of that — in some cases around 10% or less of the cost when measured by the amount of AI processing a customer uses.

And some of these Chinese models are also being released with their underlying technology, or “weights,” available for others to use.

In simple terms, the “weights” are a crucial part of what makes an AI model behave intelligently. If those weights are available, a company or government with sufficient computing infrastructure can download the model and operate it themselves rather than simply renting access to it through somebody else’s servers.

This is where the economics start getting uncomfortable for the US AI industry.

China has effectively done to frontier AI what it has done in several other industries — including fashion, phones and solar panels:

Take something expensive, produce a functionally similar version at dramatically lower cost, and destroy the premium.

The question for customers becomes very simple:

If I can get almost the same result for one-tenth of the price, why would I continue paying ten times more?

That is the first layer.

But the second layer is far more important.

Layer Two: The Real Kill Shot — Open Source

The real strategic advantage is not simply that Chinese AI can be cheaper.

It is that some of these models are open rather than locked inside a company’s private servers.

Think about the difference between renting a car and owning one.

With a closed AI model, you are essentially renting access.

You send your question or business problem to the company’s computers. The company processes it and sends the answer back.

With an open model, the situation can be very different.

A government or company with sufficient computing power can download the model, install it on its own servers and operate it itself.

That means the model can potentially sit inside the country’s own data centre or inside a company’s own private network.

India can do it.

Reliance can do it.

State governments can do it.

European research laboratories can do it.

Southeast Asian companies can do it.

African technology operators can do it.

The model can be installed behind the organization’s firewall — essentially a digital security barrier that prevents outside systems from freely accessing the company’s internal network.

Then the organization can take the model and fine-tune it.

Fine-tuning simply means teaching an existing AI model to perform better for a particular purpose by training it further using specific information.

An Indian company could train it for Indian customers.

An Indian government department could train it on its own requirements.

A bank could adapt it for banking.

A hospital could adapt it for its own internal operations.

A company could build its own AI assistant on top of it.

And the important part is that the sensitive information can remain inside the organization’s own infrastructure.

The intelligence stays inside.

The data stays inside.

The company controls the system.

The country controls the infrastructure.

There doesn’t necessarily have to be a constant connection back to the original Chinese company.

There is no requirement to keep sending every piece of information to an overseas AI provider.

And that is a massive change.

This Is Not About “Trusting China”

Some people will immediately ask:

“But why would India trust China with its AI?”

That is actually the wrong question.

India does not necessarily have to trust China.

If an open model can be downloaded and operated locally, India can take the technology and run it on Indian infrastructure.

The model can be examined, modified, secured and developed further inside India.

The same applies to other countries.

This is not necessarily about becoming dependent on China.

It is about breaking dependence on any single foreign provider.

That distinction is extremely important.

Because once an open model is available for download, the original creator has far less control over what happens next.

The model can be copied.

It can be modified.

It can be fine-tuned.

It can be combined with other technologies.

And eventually, countries can build their own versions.

Every Country Can Build on Top

Imagine India downloads an open Chinese model today.

Tomorrow, India can begin developing an Indian version.

It can add Indian languages.

Indian laws.

Indian government data.

Indian business knowledge.

Indian cultural context.

Indian security requirements.

Indian applications.

The original model becomes only the starting point.

Europe can do the same thing.

Southeast Asia can do the same thing.

Africa can do the same thing.

The Middle East can do the same thing.

Every country that wants affordable AI and greater control over its data has an incentive to follow this path.

And that is why this is much bigger than simply “China has made a cheaper ChatGPT.”

The bigger development is that China is helping make advanced AI portable.

Instead of AI being something you access only through a handful of American companies, it can increasingly become something that governments and companies can possess, operate and build upon themselves.

That changes the balance of power.

The US Restriction Strategy Has a Problem

The United States has spent years building some of the world’s most advanced AI systems.

Companies such as OpenAI and Anthropic have invested enormous amounts of money into developing these models.

America therefore has an obvious reason to protect its technological advantage.

One way to do that is to restrict access.

The US has already experimented with restrictions on foreign access to advanced AI models, including temporary blocks involving models such as Anthropic’s Fable 5.

The logic is straightforward:

If America’s most powerful AI remains available only to America and selected partners, America can maintain its advantage.

But here is the problem.

That strategy works only if there is no credible alternative.

The moment another country says:

“You cannot use the American model? Fine. Download ours.”

the power of the restriction starts to disappear.

Because the alternative doesn’t need permission.

It doesn’t need an American company’s approval.

It doesn’t necessarily need an American API.

It can run on your own computers.

It can sit behind your own firewall.

It can operate inside your own country.

And once it is there, you can build on it.

That means the US played the restriction card at precisely the moment when an alternative model of AI distribution was emerging.

And the moment the alternative can be downloaded and operated locally, the power of the restriction becomes much weaker.

This Is the Real Battle: Closed AI vs Open AI

This is why the real battle may no longer simply be:

America vs China.

The deeper battle could become:

Closed AI vs Open AI.

Closed AI means the company owns the model.

You pay for access.

You use its infrastructure.

You follow its rules.

The company can change the price.

The company can restrict access.

The company can decide which countries can use it.

Open AI models change that equation.

If the model is available for download, the customer has much greater control.

You can run it yourself.

You can modify it.

You can build on it.

You can keep your information within your own infrastructure.

And you don’t necessarily have to accept somebody else’s pricing forever.

That is potentially devastating to the traditional business model of expensive, proprietary AI.

Because the moat around closed AI becomes smaller.

The closed companies can still have the very best models.

They can still have better products.

They can still have stronger ecosystems.

But they increasingly have to justify why customers should pay the premium.

And This Is Where India’s Choice Becomes Critical

So where does India stand?

My answer would be:

India should not blindly choose Washington.

And it certainly should not become dependent on Beijing.

India should choose pragmatism.

Take advantage of whatever technology gives India the greatest capability at the lowest strategic cost.

If an open Chinese model is powerful and inexpensive, India should be willing to use it.

Run it on Indian data centres.

Develop Indian versions.

Build Indian applications on top.

Keep the data inside India.

Keep the value created by that AI inside India.

At the same time, India should continue its relationship with the United States for areas where America remains strategically important:

chips, capital, technology, research, businesses, global markets and geopolitics.

There is no reason India has to choose one side completely.

India can cooperate with America where America is strongest.

India can use open Chinese technology where it makes economic and technological sense.

And India can simultaneously build its own AI capabilities.

That is not weakness.

That is strategic independence.

The Naked Truth

The uncomfortable reality is this:

Closed, expensive and restricted AI models are going to face increasing pressure from open, cheap and locally controlled AI models.

The United States spent years building the technological lead.

Then it started thinking about how to restrict access to that lead.

China has taken a different approach.

It has undercut the price.

It has opened the gates.

And if these open models become sufficiently capable, countries around the world will have a powerful alternative.

The market is beginning to understand that possibility.

This doesn’t necessarily mean America is finished.

It doesn’t mean China has won.

And it doesn’t mean Nvidia, OpenAI, Anthropic or the other American AI companies suddenly become irrelevant.

But it does mean the rules of the game are changing.

The value of being the company that controls access to intelligence could fall if intelligence itself becomes widely available.

And that is the part investors need to think about.

India Does Not Need to Pick a Side in a Morality Play

India should not approach this as a question of:

“Are we with America or China?”

That is yesterday’s way of thinking.

The better question is:

“Which technology gives India the capability, affordability and sovereignty to build its own future?”

If an open model gives India that capability, India should use it.

If an American technology is superior in another area, India should use that too.

And wherever possible, India should build its own layer on top.

Because ultimately, the objective should not be to become dependent on American AI or Chinese AI.

The objective should be to make sure India can build Indian AI.

That is the real opportunity.

China has changed the economics.

Open source has changed the distribution model.

And together, they could change who controls the next generation of artificial intelligence.

India does not need to pick a side in somebody else’s morality play.

It needs to grab the capability that lets it build without asking anyone for permission.

That is the only winning move.

Comments

comments