I'm Uncle Haowai, and I'll teach you how to analyze tech companies during the AI transformation.

I. A Thought-Provoking Question

If you had to predict what the AI market will look like in one year (early 2027), what would you say?

You might think: Will OpenAI release GPT-6? Will Claude surpass GPT? Can Gemini make a comeback?

Honestly, I thought the same way at first. But then I realized I was asking the wrong question.

1.1. First Layer of Thinking: Product Competition Exists, But That's Not the Main Battlefield

"Which AI will win? ChatGPT, Claude, or Gemini?"

This question isn't wrong—there will be competition. ChatGPT has 600 million monthly active users, Claude dominates the enterprise market, and Gemini is integrated into 3 billion Gmail users.

But let's do the math:

Direct-Use Market:

• ChatGPT Plus: $20/month × 20 million users × 12 months = $4.8 billion/year

Embedded Market:

• Gmail, Siri, bank customer service, medical diagnostics... users don't even know whose AI they're using
• Gartner predicts: $200 billion AI software market in 2027, 80% of which is embedded

$4.8 billion vs $160 billion — a 33x difference.

It's like iPhone users buying iPhones, not Qualcomm chips, but Qualcomm is doing just fine.

So the real battlefield isn't "whose product is better," but "who can become the underlying supplier for more applications."

1.2. Second Layer of Thinking: So Where's the Competition?

Okay, if AI will "integrate into all industries," does that mean there's no competition? Everyone has a chance?

Wrong again.

Think about the power industry:

• End users don't care who generates electricity (user-level competition is weak)
• But power plants, grids, generator manufacturers... competition is fierce (infrastructure-level bloodbath)

AI is the same.

On the surface:

• ChatGPT: 600 million MAU
• Claude: thousands of enterprise clients
• Gemini: integrated into 3 billion Gmail users

Looks like a "three-way standoff," each with its territory.

But dig deeper:

• Whose chips does ChatGPT use for inference? → Mainly Nvidia
• Whose chips does Claude use? → Starting in 2025, switching to Google TPU (1 million unit order)
• Whose chips does Gemini use? → In-house TPU

This is the real battlefield.

1.3. Third Layer of Thinking: Three Questions That Will Determine the Future

Now the question becomes:

Question 1: Who Controls Computing Cost?

Assume an AI company processes 10 trillion inference queries per year:

• Using Nvidia GPUs: Annual cost $3 billion
• Using specialized AI chips (like TPUs): Annual cost $1 billion
• Saving $2 billion—that's not small change, that's survival capital

Question 2: Who Defines the "Communication Protocol" for AI Agents?

The future isn't one AI handling everything, but countless AI Agents working together:

• Restaurant booking AI needs to talk to restaurant AI
• Financial AI needs to talk to bank AI
• Procurement AI needs to talk to supplier AI

Whoever defines these "conversation rules" becomes the HTTP protocol setter of the AI era.

Question 3: Who Gets Data Others Can't?

Text data (web pages, books) can be crawled by anyone.

But real-world interaction data:

• Robot operational data in factories
• AI Agent actual transaction data
• Physical world sensor data

These are the moats.

1.4. Extended Thinking: Will AI Become "Invisible"?

Some say AI will eventually become like electricity—ubiquitous but never discussed.

I don't think so.

AI is more like nuclear power:

• Ubiquitous (nuclear accounts for 10% of global electricity generation)
• But forever controversial (safety, ethics, regulation... constant debate)

Why?

Because AI involves:

• Fairness issues (algorithmic bias)
• Employment issues (replacing human jobs)
• Power issues (whoever controls AI has pricing power)
• Identity issues (what is "uniquely human value")

So AI won't be "quietly transformative," but "ubiquitous and forever controversial."

1.5. Back to the Original Question: What Happens in One Year?

Now we can answer properly:

The AI Market in Early 2027:

Application Layer:

• No "one AI to rule them all"
• ChatGPT, Claude, Gemini each have their scenarios
• But overall scale explodes (users from 1 billion to 3 billion)

Infrastructure Layer:

• Nvidia's share drops from 90% to 60%
• Specialized AI chips (TPUs, custom chips) take 30%
• Price wars begin, inference costs drop 50%

Protocol Layer:

• Some AI Agent communication protocol becomes the de facto standard
• Companies controlling the standard see valuations double

Data Layer:

• Robots begin large-scale deployment
• Physical world data becomes the new moat

But these are just predictions. How do we verify? How do we quantify?

That's what comes next.

II. What Analytical Tools Are Needed to Understand These Trends?