Connor Pu • Analysis • 2025.2

AI BUBBLE?
WHAT? WHY? HOW?

Two months ago, when I was writing the thesis for this speech, I said: "Build a new Netflix with AI in the next three years." Now, two months later, I have to say: I was wrong. AI cannot drive the creation of a new Netflix.

So, why it changed so quickly? Let me tell you why.

01. Stock Prices: Nvidia & Cisco

Before the internet bubble burst in 2000, Cisco was one of the most profitable companies. Nvidia, like Cisco, is a monopoly in the AI industry, selling essential tools. With over 80% market share, Nvidia has made huge profits in the semiconductor industry.

However, a report by Morgan Stanley states the semiconductor industry is cyclical, and revenue growth has likely peaked. This raises the question: Is AI a bubble? Will the AI bubble burst? Are we already in a 1999 situation?

So, let's compare the AI trends of 2025 with the internet trends of 2000. Since 1989, when the concept of the World Wide Web was first introduced, the internet was mainly used by businesses and governments. This is very similar to AI technology before 2022, such as AlphaGo.

By 2023 and 2024, with the launch of GPT-3, the cost of using AI decreased, commercial demand and capital inflow increased. Tech giants rushed in, pushing up prices of tech stocks.

Very similar to 1995, when companies like Netscape popularized the internet concept by building a web browser.

It changed as the Fed had to reduce interest rates in response to Russia's debt default and the Asian financial crisis; the economy continues to overheat, pushing the internet bubble to its peak. From 1998 to 1999, internet companies’ stock prices exploded, and the Nasdaq rose 100% in a year. New IPOs flooded the market.

We all know what happened next. In March 2000, the bubble burst. Nasdaq dropped 80%. Stocks of Amazon, Apple, and Microsoft fell by more than 90% and many of them went bankrupt.

Guys, it looks terrifying. But I can assure you. We are not in 1999. However, we are actually in 1995. A report from Bank of America proves it: both charts show right now, we are at the same levels as in 1995. Besides, AI application companies are not yet in full force.

02. Predictions: ARK Invest

The predictions from Cathie Wood, CEO and CIO of ARK Invest show some points.

Economic Growth: The IMF predicts global economic growth rate to be 3.1% by 2030—wrong, 10% said by Wood. Since the first industrial revolution raised the average growth rate to 0.6%, while the second industrial revolution enabled a 3% growth rate. Every technological revolution will lift the economy to a new level.
Specific Value: In specific areas, the value of AI is already clear. Tesla’s Full Self-Driving (FSD) AI has already outperformed its competitor Waymo. By June, it is expected to reach human-level performance. In 2024, efficiency of AI-driven drug development is eight times faster than in 2023.
Coding Capabilities: AI programming capabilities have also advanced. In one year, AI programming accuracy has jumped from 4% to 72%. The foundation of these is computational power; it has shown exponential growth in the past few decades, and it will continue to rise.

03. The Future of Human

Very similar, Yuval Harari, author of , also proposed three future predictions based on AI:

Humanity may lose economic and military value, since it fully replaces certain basic skills in both the economy and war, leading to the formation of a "useless class."
Loss of Decision Making: Humans may lose the right to make individual decisions. AI agents will have a significant influence. So, what is an AI agent? Let's use coffee-making as an example. A regular AI chat bot might help you choose coffee beans based on your preferences. However, an AI agent goes further: it can predict if you need a coffee by analyzing your daily habits and health. It will then make the coffee for you, but since it records you complained about the flavor yesterday, it will automatically create a new blend using your favorite ingredients.
Biological Differences: With advancements in biotechnology, gene editing, Neuralink and so on, a new biological class may form. For instance, Demis Hassabis, the Nobel Prize winner and the founder of AlphaFold, used AI to predict the 3D structures of 200 million proteins with high accuracy—a task that would have taken thousands of years.

05. DeepSeek's Impact

Now, let's take the elephant out of the room, talk about DeepSeek. There are several advantages of it:

  • A. Low costs: 5% of GPT-01, with $6 million in development, offering high quality.
  • B. Usage costs are only 2% of GPT-01.
  • C. China's chip manufacturing industry probably has fully integrated.
  • D. Open-source model, transforming how customers and businesses use AI.
  • E. New algorithms, such as reinforcement learning, improved efficiency drastically.

By the way, the low cost of DeepSeek has raised concerns on Wall Street. One partner at Sequoia Capital believes that AI applications won’t recover the over $1 trillion global investment in AI infrastructure, leading to a massive imbalance between investment and returns.

Another concern is whether AI will lead to mass unemployment. At the same time, Meta CEO Mark Zuckerberg declared that coding is dead. Scaleforce's CEO said they no longer hire programmers. Microsoft’s CEO suggested that SaaS/apps will no longer be developed.

So what will AI applications or companies look like, what will the new job be like in future? The analysis of a16z give us a hint. Not all companies will benefit equally from the economic growth driven by AI. First, companies that provide AI infrastructure will certainly benefit. However, the biggest winners will be companies focusing on AI applications.

Personal Experience

I am also running a B2B studio called X-tra AI studio. We generate AI videos for film production companies; some of our works have been recognized with awards, and I’m keeping on it, developing it as a new Netflix.

However, why did I say Netflix will not emerge at first?

  • First, since everyone will have the ability to create films with AI, it’s hard for companies and studios to build barriers.
  • Second, AI replacing the entire movie-making process will shrink capital, reduce GDP. Capital won't allow it.
  • Third, there’s already an oversupply in the film and video making industry. Higher productivity will only worsen the oversupply.

"But, the only thing I’m sure about is—so called just do it."

Even if I know that there probably won’t be a new Netflix with AI, even if I know I might fail. That doesn't matter. I am still trying, I am still learning, and still continuously reshaping myself. I also give this encouragement to you guys, hope to improve together. Thank you.