Many friends recently have shared podcasts by Wang Yuquan about the “AI bubble bursting” thesis!

I watched both segments.
To be honest, I agree with his bubble-bursting thesis, because it’s a matter of inevitability following **the pattern**. But I don’t agree that 2029 should be treated as a precisely-timed marker.
Right now, it should be the early stage of AI from 2021 to 2027, similar to Crypto’s 2012 to 2016.
Then from 2028 to 2031, it could be a high-risk window where AI shifts from infrastructure frenzy to the acceptance and validation of commercial returns, similar to Crypto’s 2017 to 2021.
2029 is just the median point of this window. Whether it must “burst” here remains to be seen.
Some of his more interesting viewpoints:
1️⃣ Absolute early stage!
Wang Yuquan believes that we are in the absolute early stage of AI development. Major technological revolutions usually go through:
technology breakthrough → social panic → gradual acceptance → collective frenzy → bubble bursting → infrastructure maturation → long-term prosperity.
By his judgment, society is still debating things like “Will AI steal jobs?” “Will ordinary people be eliminated?”—and this still carries the characteristics of the panic phase.
The real peak of the bubble is when everyone is certain that AI can make money, all companies start packaging AI, and capital stops seriously calculating returns.
2️⃣ Why 2029?
He borrowed historical analogies from the Industrial Revolution and the auto industry.
After Ford’s assembly line appeared, it didn’t immediately lead to a society-wide productivity boom.
On the contrary, companies made large-scale investments in new capacity, but most industries hadn’t truly learned how to restructure organizations and production processes. In the end, severe mismatches between capacity and finance emerged.
Putting it in today’s terms, it’s a bit like:
Large models are like a new general-purpose technology;
data centers, chips, and electricity are like infrastructure;
AI programming and agents are like new ways of production;
but most companies haven’t yet completed the transformation of organizational structures and business processes.
So in the coming years, there may first be an investment frenzy.
When the market realizes, “We’ve put so much into compute, but enterprises’ profits and productivity haven’t caught up,” valuations and capital expenditures will be repriced. Then everything starts to break—cognitive misalignment leads to fluctuations, and finally the market reconstructs itself.
He believes all of this will most likely happen around 2029.
He often says: “Around 2029 could be a dangerous point for capital markets, but it may also be a golden starting point for AI application entrepreneurship.”
In fact, this isn’t an opinion he formed only recently. The material I found shows that as early as at the 2024 November 9 Frontier Conference, he also proposed that the first half of the digital revolution might face a ‘great canyon’ around 2029, because society might overestimate the changes AI will bring in the short term.
3️⃣ The biggest current contradiction is that capital expenditure growth is faster than commercial returns.
The following is not Wang Yuquan’s view, but my own thoughts—feel free to criticize.
I also share the same view as Wang Yuquan; the risk that’s visible to the naked eye is accumulating rapidly.
Reuters, based on LSEG data, analyzed that by 2027, the AI capital expenditure growth of companies such as Microsoft, Alphabet, Amazon, Meta, and Oracle may significantly exceed the growth of operating cash flow. In the calculations, for every $1 increase in operating cash flow, there may correspond about $1.57 of new investment.
This structure can’t be maintained forever.
What the capital markets are willing to accept now is logic like:
Build compute first, and revenue will come naturally later.
Once in the next two or three years it turns into:
Compute gets built, but revenue growth, profit margins, and productivity don’t keep up,
the market will shift from competing over “future space” to calculating the “investment payback period.” This is usually when the bubble begins to burst.
So the technical value of AI is real, but valuations can run ahead and discount many years early.
4️⃣ Some judgments.
Divide the coming years into three stages.
First stage: 2026 to 2027—Infrastructure continues to expand. Compute, power, data centers, chips, networks, and model training will still absorb a large amount of capital.
In this phase, selling shovels is the most profitable.
Second stage: 2027 to 2029—AI applications and agents enter large-scale validation/acceptance.
In this phase, truly big companies will emerge, and a large number of companies that look advanced but can’t charge fees will die.
Third stage: around 2029—Capital expenditures and cash flows collide in a positive direction.
If by then the industry still can only sustain growth through continuous financing and capital expenditures, and AI application revenues are insufficient to cover infrastructure costs, then a larger valuation compression could happen.