How big is the bubble in the AI market?

Author: Honest and Steady Mr. Mai Source: X, @Michael_Liu93

A few days ago, I discussed a topic with a friend: “How big is the AI market bubble?” Let me share our discussion and do some calculations on big numbers with everyone.

First, roughly estimate the market size. Taking ChatGPT, which has the most users, as an example: currently, there are about 800 million monthly active users worldwide (already penetrating 10% of the global population). Among them, paid users number between 35 million and 40 million. Taking 37.5 million as the median, most of these users pay around $20 per month. Assuming an average monthly customer price of $25, ChatGPT’s annual revenue from the consumer side (C-end) is approximately $11.25 billion.

Suppose these 37.5 million customers each pay for at least 3 AI services, then the total C-end market size would be about $33.75 billion. (I think an average of 3 services is already a very optimistic assumption. I personally consider myself a heavy AI user, paying for about 3-4 AI products, roughly $100 per month.)

So, what is the market value of the companies serving this AI C-end market?

Assuming we exclude the primary market (roughly estimated at around 1 trillion dollars), and only consider the seven major tech giants in the secondary market—the Magnificent Seven: Nvidia at $4.4 trillion, Apple at $3.9 trillion, Microsoft at $3.7 trillion, Amazon at $2.4 trillion, Google at $2.4 trillion, Facebook at $1.7 trillion, and Tesla at $1.3 trillion.

Except for Nvidia’s $4.4 trillion market cap, which is more purely AI-focused, the valuation of the other companies is more diversified across various business segments. Assuming we only assign 25% of their valuation as AI-related (I believe the actual market premium for AI is much higher than 25%, but for conservative bubble estimation, we take 25%).

The total AI-related market cap would then be approximately $8.25 trillion. (Of course, many AI companies are not included here—AMD, Palantir, Qualcomm, Oracle, etc.)

Dividing $8.25 trillion by the $33.75 billion market size gives a P/S ratio of about 244x.

How should we interpret this 244x P/S figure? We can look back to March 24, 2000, the peak day of the dot-com bubble, to see what the P/S valuations of leading internet companies were at that time.

Amazon: 19x

Cisco: 35x

Qualcomm: 22x

Microsoft: 26x

IBM: 3x

Oracle: 27x

Intel: 16x

After the dot-com bubble burst, most of these companies, even giants like Amazon and Microsoft, took 10-15 years to recover to their bubble peak levels in 2000.

Of course, the 244x figure doesn’t necessarily mean the AI sector is definitely a bubble, because currently, most AI revenue doesn’t come from this $33.75 billion C-end market but from “spending on infrastructure and hardware.” Ultimately, companies buy hardware and build infrastructure to serve the C-end market, which at present isn’t as large as many imagine.

The awkward problem AI faces now is: “Even if AI is quite useful, it remains a ‘production tool,’ not ‘productivity’ itself, so AI service companies can only earn money as ‘tools,’” while the capital market values them based on the ‘productivity’ story.

What does this mean? If you ask a consumer or business owner to spend $50-100 per month to equip employees with AI, they are likely willing to do so. This AI might indeed help save the cost of one or two entry-level employees. But if you ask them to spend $2000-3000, they are unlikely to be willing.

So even if today’s C-end market penetration increases tenfold to 375 million people (more than the US population of 350 million), a sector valuation of 24.4x P/S is still not cheap in any industry.

But who knows? Maybe one day, AGI will truly emerge, and AI will become productivity itself. Then, these AI companies won’t be so expensive.

Just like looking back to 2000—were Amazon, Apple, and Microsoft expensive? Compared to now, they are very cheap. But the premise is, if you bought in 2000, you’d have to wait 10-15 years to see your investment pay off.

And we don’t know when AGI will arrive, or if it will arrive at all.

Let’s take it one step at a time.