Are Big Tech Companies Starting to Regret Investing in AI?

A few years ago, Big Tech’s view of AI went roughly like this:

Large models keep improving → users adopt them at scale → enterprises buy them at scale → AI becomes the next internet → whoever has the most computing power wins.

So everyone began buying GPUs, building data centers, and training large models like crazy.

The problem is that after several years, everyone has realized that the earlier steps did indeed happen—but the final step, making money, has not been nearly as easy as expected.

And that creates an awkward situation.

AI is now in a very strange position.

On the one hand, no one dares to say that AI is useless.

Programmers use it. Designers use it. Customer service teams use it. Office workers use it. Search engines are integrating AI. Smartphones are integrating AI. Enterprises are building agents.

Many companies are even seeing explosive growth in both AI users and token consumption.

But on the other hand:

The more users use AI, the faster providers sometimes burn through cash.

Traditional internet products have one particularly attractive characteristic.

For example, when WeChat gains one more user, Tencent’s marginal cost barely increases. When Google handles one more search, the cost is also very low.

That makes it natural for internet businesses to follow this pattern:

More users → more revenue → higher profit margins.

AI does not work quite the same way.

When you ask ChatGPT a complex question, GPUs genuinely have to perform the computation in the background.

More users mean more tokens, and more tokens mean higher inference costs.

This has created a situation rarely seen in the traditional internet industry:

A product becomes wildly popular, and the company starts worrying about the cost.

Worse still, inference is only part of the story.

Before that come GPUs, servers, high-bandwidth memory, networking equipment, data centers, electricity, cooling systems, and model training.

Companies such as Microsoft, Google, Meta, and Amazon are now spending extraordinary amounts every year on AI infrastructure.

That is why Wall Street has started asking a question it previously hesitated to raise:

You are spending all this money—but when will you earn it back?

This is the real source of the claim that “Big Tech is starting to regret getting into AI.”

But there is a major misconception here.

If Big Tech genuinely believed that AI was a scam, the solution would be simple:

Cut capital expenditures.

Buy fewer GPUs. Build fewer data centers. Hire fewer AI specialists.

Yet the reality is exactly the opposite.

Companies may be talking more about “return on investment,” but their AI spending remains enormous.

Why?

Because this is no longer a question of whether they want to invest in AI.

The reality is:

No one dares not to.

Can Microsoft afford to stop?

If Microsoft pulls back, what happens if Google’s Gemini succeeds in building the next generation of search, productivity tools, and AI agents?

Can Google afford to stop?

Even less so.

After all, generative AI’s earliest and most direct threat was to Google Search.

Can Meta afford to stop?

If AI agents truly become the gateway to the next generation of the internet—and Meta lacks its own models and computing infrastructure—it may end up repeating the mobile internet era, when it was constrained by Apple and Google.

The same logic applies in China.

What Alibaba, Tencent, ByteDance, and Baidu truly fear is not earning a few billion dollars less from AI this year.

What they fear is this:

What if AI really is the next computing platform—and they fail to secure a seat at the table?

As a result, the entire industry is now trapped in a classic prisoner’s dilemma.

Suppose no one invests in AI.

Great.

Collectively, companies save hundreds of billions of dollars a year, and their profits look fantastic.

But if just one company secretly invests hundreds of billions and actually manages to build an operating-system-level AI agent, everyone else could lose control of the gateway to the next decade.

So the rational choice for every company becomes:

I do not know whether AI will ultimately be worth this much money.
But I cannot afford to bet that it will not be.

And so everyone keeps spending.

This is also why I believe it is still too early to say that the AI bubble is about to burst.

AI may very well satisfy two conditions at the same time:

1. AI is a genuine technological revolution.
2. The AI industry is currently experiencing a massive investment bubble.

These two statements are not contradictory at all.

The dot-com bubble of 2000 is the best example.

The market was not wrong to believe that the internet would change the world.

Quite the opposite.

The internet changed the world even more profoundly than the wildest predictions of that era.

What the market got wrong was this:

The internet is important → therefore, every internet company is worth an enormous amount of money.

That middle step was wrong.

AI may follow the same pattern.

Over the next decade, AI may genuinely transform software development, customer service, search, advertising, gaming, education, healthcare, and office work.

But that does not mean:

• Every AI company today is worth tens of billions of dollars.
• Every data center being built today will operate at full capacity.
• Every GPU being purchased today will generate enough profit to justify its cost.

So the real question is not whether Big Tech will abandon AI.

At this point, the answer is already fairly clear:

It will not.

The real question is something else:

Will Big Tech discover, over the next few years, that the $1 trillion worth of AI infrastructure it has built can ultimately generate only a few hundred billion dollars in revenue?

If that happens, AI will not disappear.

ChatGPT will not disappear.

Claude, Gemini, and Doubao will not disappear either.

Programmers will continue using AI to write code.

Enterprises will continue deploying AI agents.

But the capital markets will go through an extremely painful repricing.

And when that happens, we may witness a fascinating scene:

The technological revolution was real.

AI really did change the world.

But many of the people who invested in AI still lost their shirts.

That is the real danger we should be watching in this AI cycle.