The 10 AI Slides From Mary Meeker That Actually Matter For Builders

Mary Meeker dropped 340 slides on AI. But only 10 matter if you're building something real. Here's the shortcut to what you need to know and how to act on it.

Quick insights 🔎

Here’s what caught my radar this week

📊 $142 Billion AI Reality Check
2,000 companies fought for 50 spots on Forbes' new AI list. The winners are companies whose AI actually completes your job, not just chats about it.

💰 The Anti-VC Playbook
Y-Combinator runs like a software company, not a fund. Their bizarre approach created $800B in value. One ex-employee spills what really happens.

🎵 ChatGPT's Weirdest Bug Yet
Paid users discovered ChatGPT's new voice mode randomly generates background music and gibberish. OpenAI calls it a "hallucination" but won't fix it.

---

This week's mental model 🧠

The Knowledge Trap

You read everything. You build nothing.

The guy who shipped a crappy MVP is now onboarding your future customers.

---

Last week, Mary Meeker dropped 340 slides on AI. Brilliant work but buried in that mountain of data is one simple truth.

The fairytale 🧚

To compete in the AI gold rush, you need to understand everything about AI… every trend, every model, every benchmark, every tool.

The reality ✅

Mary Meeker's exhaustive research actually proves the opposite. The winners in AI aren't the ones studying everything. They're the ones who understood 10% and are using that 10% to solve real customer problems.

They picked one boring problem in one boring industry. They're using basic AI to solve it and they're charging money for it.

People laughed at these AI wrapper companies previously, but actually they might be the winners in the long run.

It took me a week but I went through all 340 slides (yes, really, my eyes hurt) and kept coming back to these 10.

The 10 slides that will change how you build:

1. AI just became dirt cheap (slide 137)

Training an AI model is still expensive. But the good news is using AI is getting cheaper and cheaper. That’s great for startups focused on AI-powered apps, not building the models themselves.

2. All AI models are basically the same (slide 142)

At the start of 2024, there was a big gap between models.

By early 2025, all 3 are scoring within 23 points of each other. That means the "best" AI model is only 2-3% better than the others. They all do the same thing.

Therefore, use whatever model is fastest and cheapest and switch whenever you find a better deal.

To benefit from this, I recommend you build your app to work with any model.

3. The vertical AI gold rush is real (slides 233-243)

• Cursor (helps programmers code): $300M revenue in 2 years

• Harvey (helps lawyers write): $70M revenue in 15 months

• Abridge (helps doctors take notes): $117M revenue in 5 months

"AI for everyone" is yesterdays news.

"AI for dentists" will print money.

Pick one job in one industry.

Build AI that does one specific task and dominate that niche before expanding.

4. OpenAI is losing money (Slide 173)

OpenAI made $3.7B but spent $5B on compute alone. They're bleeding cash.

Even the AI leaders aren't profitable. "Build it and they'll come" doesn't work in AI.

Charge from day one and price for profit, not growth. Don't try to spend your way to product-market fit.

5. Open source won (slide 268)

Meta's free AI model Llama was downloaded 1.2 billion times. It's almost as good as paid options. This means you don't need proprietary AI anymore. The free stuff is good enough.

Use open source models. Spend your money on customer acquisition, not AI licenses.

I even wrote about this here: Why Renting AI Is The New Fairytale

6. Developers already use AI (slide 147)

63% of developers use AI daily. Last year it was 44%. Next year it'll be 80%.

If your developers aren't using AI, they're working at half speed.

Make AI tools mandatory.

7. China's building fast and cheap (slide 286)

Chinese AI models match US performance at 1/10th the cost.

They're moving scary fast.

Your technical moat is temporary. Someone will clone your AI features for pennies. So build moats that aren’t technical. Customer relationships, data, brand, network effects.

8. Big tech spent $212 Billion on infrastructure (slide 97)

Amazon, Google, Microsoft etc. spent $212B building AI infrastructure in 2024.

They built the highways and you should build the gas stations.

Don't build infrastructure. Use theirs. Focus 100% on solving customer problems.

9. AI jobs are exploding (slide 332)

AI job postings up 448%. Traditional IT jobs down 9%. The skill shift is happening.

Everyone's hiring AI people. You won’t win this talent war on salary alone. Train your existing team. Hire for potential, not experience. Build a learning culture.

10. ChatGPT hit 800M users in 17 months (slides 55-56)

ChatGPT reached 800M weekly active users faster than any product in history. 90% of users are outside North America.

Speed and global reach matter more than perfect AI. ChatGPT won by being first and everywhere.

Launch fast, go global immediately.

But not before you are ready: Why Startups Fail at Global Expansion

Where to start

1. Pick one vertical from slides 233-243

2. Find 10 people who do that job

3. Watch them work

4. Build AI that saves them 2-3 hours.

Tools

For the tools you need, read these posts:

1. Vibe Coding Sounds Like a Joke Until It Ships Faster Than Your Team

2. Vibe Marketing Is Already Here

3. Why Renting AI Is The New Fairytale

North Star

The 330 slides I didn't mention?

They're about energy consumption, robot statistics, and philosophical debates. Interesting for academics. Useless for builders.

Mary Meeker's 340 slides prove one thing. The people winning are the ones solving real problems that customers actually pay for.

The best AI company is the one that picked the right problem.

What problem are you solving?

---

Vibe Marketing Is Already Here

---

Until next week, keep building, no fairytales required.

Martin, Chief Ranter at Uncharted

This week’s track

LCD Soundsystem - "Losing My Edge"