You know what I saw when I came back to Apple in 1997?

350 products. A pile of shit. Every product had a “reason to exist,” every product manager could tell you a story. But the result? The company was on life support.

The AI landscape in 2025 reminds me of that moment.

Let me start with what I see.

How crowded is the AI market right now? 153 companies, covering everything from infrastructure to applications. The top players: OpenAI, Google, Anthropic, Meta, and a Chinese company called DeepSeek that came out of nowhere with a model that rivals ChatGPT at a fraction of the cost. NVIDIA is selling shovels at the bottom, and everyone depends on them.

The capabilities of these models are converging fast — GPT-5, Claude 4.5, Gemini 2.5 — for most tasks you can’t even tell which is better.

This is a classic “feature race” trap.

Which brings me to the real question: Who’s actually making a product?

What do I see? Every company is benchmarking parameters, inference power, reasoning scores. That’s not wrong — but it’s necessary, not sufficient.

Google may have the strongest AI tech. But have you seen how messy their products are? Gemini, Bard, Workspace AI, NotebookLM… You open Google and you have no idea which door to walk through. That’s what happens when the toner-heads are in charge — engineers ship every feature, but nobody asks: What is the complete user experience?

OpenAI is doing a better job here. ChatGPT is one unified entry point, it has memory, now images, voice, Deep Research. It’s closer to the whole widget. But still a long way off.

I want to talk about DeepSeek separately.

A Chinese company built a model on par with OpenAI’s at a radically lower cost. A lot of people in Silicon Valley are scared by this.

I don’t think you should be scared. I think you should be ashamed.

If you’re spending tens of billions of dollars and a team with a fraction of that catches up — that doesn’t mean they’re too strong, it means you've failed to focus on what truly matters.

DeepSeek proved one thing: Model capability itself is not a moat. The real moat is the end-to-end experience, an ecosystem users can’t leave, the vertical integration of hardware, software, and services.

So who’s most likely to win?

My take: At this point, nobody has really won.

Everyone is selling better hammers, but nobody is redefining what it means to build a house.

Where’s the real opportunity?

Not smarter models — better products.

Someone will truly integrate AI into human workflows — not as a chat window, but as a completely trusted, seamless cognitive extension. That hasn’t been built yet.

Before the iPod, MP3 players existed. Before the iPhone, smartphones existed. Those products “worked,” but none of them were insanely great.

AI is at that exact moment now. It’s usable for everyone, but nothing makes you say — “Oh wow. Oh wow. Oh wow.”

That product hasn’t been made yet.