Google's $100B Mistake

In partnership with

Hey there,

Imagine inventing the wheel and then watching someone else build a car empire with it.

That’s basically what happened with Google and AI.

Back in 2017, Google invented transformers, the very architecture powering ChatGPT today. But at that time, they didn't take it seriously enough to invest big 

Why? They were scared the chatbot would "say dumb things."

OpenAI didn’t wait. They shipped anyway.

Now OpenAI is valued around $157 billion, and Google is trying to catch up using technology it originally created.

That difference, moving fast vs playing safe, shows up everywhere in AI right now.

Speed First, Rules Later

Speaking of moving fast vs playing safe, last week, Trump signed an executive order blocking US states from setting their own AI rules. 

Reason? Too many laws will slow innovation.

On the surface, Big Tech loved this. Fewer rules means faster shipping.

But when Fortune spoke to CEOs across healthcare, finance, education, and utilities, the reaction was more mixed.

Most of them said the same thing:

They don’t want 50 different state laws. But they also don’t want no rules at all.

Now that pause creates something worse. Uncertainty.

And uncertainty matters when AI systems are getting bigger, faster, and more expensive.

The Model Race Gets Costly

At the same time regulation was being pushed aside, the model race heated up.

Google launched Gemini 3 and claimed it had overtaken OpenAI.

OpenAI responded almost immediately with GPT-5.2:

• Much larger context window

• Stronger math and coding scores

But every improvement comes with a cost.

And this is where it stops being about benchmarks.

The Money Problem Underneath

Deutsche Bank estimates OpenAI could lose $143 billion between 2024 and 2029 before turning profitable.

Even today:

• Around $10B in revenue

• Around $5B lost every year

So while models get better, the business behind them gets harder to sustain.

That’s where open source models are winning.

Why Open Models Matter More Than Ever

While OpenAI and Google spend billions competing, open-source models quietly improved.

Models like DeepSeek R1, Qwen 3, and Kimi K2 now perform close to GPT-4 on many tasks.

They’re cheaper.
They’re customizable.
And you’re not locked into one company’s pricing or rules.

This lines up with what Yann LeCun keeps saying: we confuse good language with real intelligence. Big models sound smart, but that doesn’t mean they’re the best solution for every problem.

The Oracle Warning Sign

Speaking of costs spiraling, Oracle just became the first major AI implosion casualty.

They raised $145 billion in AI-linked bonds this year alone. That's more than the last three years combined. GPU costs are through the roof. Memory prices are insane. And their revenue? Not meeting expectations.

The stock collapsed.

This isn't just Oracle's problem. This is a warning sign for anyone in the AI space: The economics need to make sense.

Google had the technology but was too cautious. Oracle had the money but wasn't cautious enough.

What This Means For Your Business

Here's my take after watching all this unfold:

1. Stop betting everything on one API. Whether it's OpenAI, Anthropic, or Google, diversify. The open-source models are good enough now for most use cases. Test them. Learn them. Build on them.

2. The real opportunity isn't in the models. It's in solving specific problems that require domain expertise plus AI. Everyone has access to similar models now. Your competitive advantage is understanding your users better than anyone else.

The tools are democratized. The question is: What are you building with them?

As I watch companies raise billions for AI infrastructure while open-source models close the gap, I keep thinking about Google's mistake.

They had transformers in 2017. They could have owned this space. But they were paralyzed by fear of looking stupid.

Don't make their mistake by obsessing over having the "best". Make sure you're solving real problems instead.

P.S. If you're interested in getting deeper into open-source models and learning what's actually working, join my AI community where we discuss this stuff daily.

- Aashish

You can (easily) launch a newsletter too

This newsletter you couldn’t wait to open? It runs on beehiiv — the absolute best platform for email newsletters.

Our editor makes your content look like Picasso in the inbox. Your website? Beautiful and ready to capture subscribers on day one.

And when it’s time to monetize, you don’t need to duct-tape a dozen tools together. Paid subscriptions, referrals, and a (super easy-to-use) global ad network — it’s all built in.

beehiiv isn’t just the best choice. It’s the only choice that makes sense.

Start free today. No credit card required