How Google KILLED ChatGPT in 2 years
Remember eighteen months ago?
“Google is dead.”
“ChatGPT killed search.”
“RIP the algorithm.”
Every headline. Every podcast. Every tech bro on Twitter kept saying the same thing: OpenAI won, Google lost, game over.
Well… the pendulum just swung back. Google Gemini hit 650 million users. Marc Benioff publicly dumped ChatGPT and switched Salesforce to Gemini. And OpenAI? They declared “Code Red” internally… the exact same panic mode Google entered when ChatGPT first launched.
The hunter became the hunted.
OpenAI might be about to teach us the most expensive lesson in Silicon Valley history: even if you have the best product and the biggest user base, you could still lose everything.
Defining Failure (It’s Not What You Think)
Let’s be clear: OpenAI has 800 million weekly active users. They’re not going bankrupt tomorrow. This isn’t Pets.com.
The failure we’re talking about is more interesting… and more dangerous. OpenAI might fail at being the enduring profit pool of AI, even if their technology stays world-class. They could keep building the best models, keep impressing people with demos, and still watch someone else capture all the money.
How? They lose default distribution… becoming the engine buried inside someone else’s product. They lose pricing power… because Google and Microsoft bundle AI as a free feature. Their margins get crushed by astronomical compute costs. And they end up owning less and less of the value chain, reduced to expensive middleware while platforms and chip makers take the profits.
We’ve seen this exact movie before.
The Google Lesson Nobody Learned
Three years ago, the tech world declared Google dead. ChatGPT launches, goes viral, and suddenly everyone’s an expert on Google’s demise. “Too slow. Too bureaucratic. Can’t innovate.”
Except Google didn’t die. They’re winning now. Bigger than ever.
The “Google is dead” narrative was never about capability… it was about distribution and switching costs. Google controls 13.7 billion daily searches, 3 billion Android devices, 2.5 billion YouTube users, and the dominant browser. When Gemini launched in Search, it reached two billion users immediately… not through marketing, but through integration.
You’ve probably experienced this yourself. You search something on Google now, and instead of clicking links, you just read the AI summary at the top. You didn’t switch to a new product. Google changed the product you were already using.
That’s the power of owning the default. And OpenAI doesn’t own one. ChatGPT is just another app… something you have to remember to open.
The Four Structural Threats
The Margin Trap. OpenAI burns $1.69 for every dollar they earn. Sora costs $15 million per day to run. And they can’t raise prices to fix it, because Google and Microsoft subsidize AI as a feature. AI is OpenAI’s only product. If they raise prices, users switch to the free alternative.
Distribution Bundling. When AI gets baked into Search, Windows, Gmail, iOS, and Slack, why open a separate app? Even when OpenAI powers the backend, the platform owner captures the margin. The pie gets sliced so many ways that OpenAI ends up with scraps.
Workflow Integration Beats Best Model. Look at Claude. They took coding not by being 10x better, but by embedding into VS Code, Cursor, and the tools developers already use. Users pick what’s in their workflow, not what tops the benchmark.
Partner Economics as a Cage. Microsoft owns 27% of the company, gets IP rights through 2032, takes 20% of all revenue, and locked OpenAI into $250 billion in Azure commitments. This partnership enabled OpenAI’s scale. But it also created structural dependency on a partner with competing AI interests.
The $6.5 Billion Hardware Bet
Sam Altman knows all of this. That’s why he spent $6.5 billion acquiring Jony Ive’s startup. The thesis: if OpenAI can create a new device category, a third surface alongside your phone and laptop, they own distribution instead of renting it.
But the graveyard of AI hardware is growing. Humane’s AI Pin, Rabbit R1, Meta’s original Ray-Ban glasses… all flopped or fizzled. Every “post-smartphone” device has failed.
Still, OpenAI has advantages the others didn’t: Jony Ive on design, Foxconn on manufacturing, and 800 million users already doing the thing they’d sell a better form factor for. Unlike Humane, they wouldn’t be selling a new behavior.
If this device flops, there’s no obvious Plan B for owning distribution. The fallback, an advertising model competing with Google’s 20 years of ad infrastructure is a weak hand to play.
What This Means If You’re Building a Startup
The takeaway isn’t that OpenAI is doomed. It’s that having the best tech alone doesn’t guarantee you capture the value. OpenAI literally invented the category. Two-year head start. Most recognized brand in AI. 800 million users. And they’re still getting squeezed by companies that control distribution.
If you’re building “AI for X”, sales, recruiting, legal, whatever, you need to answer one question: **what happens when the platform you’re built on decides to compete with you?**
The companies that are winning this game aren’t competing with OpenAI. They’re building where distribution advantages matter less than domain expertise:
Glean isn’t “AI search for enterprise.” They integrate with over 100 tools, own access to proprietary company data, and embed where employees already work. The AI is table stakes. The moat is integration depth and data access.
Cursor used someone else’s model but built a better IDE developers actually use daily. The AI is swappable… always has been, always will be. The user habits and workflow integration are the moat.
What This Means If You’re Investing
This might sound counterintuitive, but this is actually the best time to be a seed investor in AI.
When the giants are fighting each other at the top of the food chain, they stop building at the bottom. They start buying. Scale AI went from a YC startup to a $15 billion valuation. Bun got acquired by Anthropic less than two years out of YC. Deepgram bought OfOne 18 months after Demo Day. The talent war is compressing the best upstream… Google, OpenAI, Anthropic, and Microsoft all competing for the same 10,000 AI engineers. Hiring takes six months. Acquisitions take six weeks.
If you’re investing in ecosystems like YC that pre-filter for top-1% talent, you’re investing in the founders the larger companies will want to buy.
Three things I’m looking for:
Underserved verticals with real barriers. Healthcare, legal, manufacturing… industries with domain-specific data, regulatory moats, proprietary workflows, and compliance requirements that generic ChatGPT can’t touch. These companies aren’t competing with OpenAI; they’re using its API to solve problems in industries twenty years behind on technology.
Agentic AI that delivers outcomes, not just tools. Within those niches, I want companies building systems that execute end-to-end workflows, not just generate text.
Early revenue traction. The hype-driven era is over. I want to see how fast teams get to $100K+ ARR… not because the revenue matters at seed, but because it proves they can sell. It proves product-market fit. A product that commands a price tag early on is a product the market fundamentally needs, not something that just sounds good in a pitch deck.
Where This Leaves Us
OpenAI is fighting the fight of their lives. Google’s coming back with force. Anthropic is winning developers. Microsoft is hedging their bets.
The AI game is shifting from “best model” to “best leverage” in real time. And that shift creates opportunity everywhere — for founders building vertical AI companies, for investors backing the next wave, and for anyone paying attention to where the value is actually flowing instead of where the hype is pointing.
OpenAI taught the world that AI could be magical. Now they’re teaching us something harder: **Magic isn’t enough. You need distribution. You need margins.**
The pendulum keeps swinging. Make sure you’re watching.
Visit lobstercap.com for more
Is it already over?