Image: Larry Page in his earlier days.

---

History Doesn't Repeat Itself, But it Often Rhymes

March 2026

If you want to understand what's going to happen in AI over the next decade, stop reading AI Twitter and start reading about the dot com bubble.

I think a lot of what happened during the search engine wars is about to happen again.

Google Wasn't Inevitable

Before Google won, the conventional wisdom was that search was a solved problem. AltaVista, Yahoo, Lycos, Excite; there were dozens of search engines, and most people thought they were good enough. Yahoo was the clear frontrunner with the most users and the most money.

Google wasn't even the first to use links as a ranking signal. But PageRank was fundamentally better, and anyone who used it could feel the difference in seconds. It wasn't a marginal improvement. It was a different experience entirely.

Now look at AI. OpenAI had the first-mover advantage. ChatGPT was the fastest-growing consumer product in history. Most people assumed the race was already over.

Then you actually use Claude, and you feel the same thing early Google users felt. Something is just different here.

The Seven Rhymes

Here's what actually made Google win, and why the same playbook is unfolding again in AI.

1. The Best Original Algorithm

Larry Page and Sergey Brin were PhD students who cared about one thing: making search actually work. PageRank wasn't built to capture a market. It was built because they were obsessed with the problem.

The parallel here is research culture. Anthropic was founded by people who left OpenAI specifically because they wanted to do the research differently. Constitutional AI, interpretability work; these came from people who are obsessed with making the technology work correctly, not from a marketing roadmap.

OpenAI, meanwhile, has spent the last two years shipping product after product at a pace that suggests they care more about the press cycle than the science.

2. The Best Business Model Execution

Google didn't invent search advertising. Overture (GoTo.com) did. But Google executed on it better than anyone. AdWords was cleaner and more self-serve. They took someone else's idea and built the definitive version.

The AI business model is still being figured out, but there's a clear split forming: API-first vs. consumer-first. Anthropic's enterprise and API business is growing while OpenAI chases consumer subscriptions, hardware partnerships, and a social media pivot. The companies building foundational infrastructure for other businesses to build on tend to win over the long term.

3. First Principles

Google didn't just have better algorithms. They thought about infrastructure completely differently. While competitors racked up massive server bills, Google built their own servers out of commodity hardware and wrote custom distributed systems (GFS, MapReduce, BigTable) that didn't exist before.

They didn't buy better versions of what everyone else was using. They invented new things because the existing solutions didn't make sense for their problem.

4. Top Talent

Early Google was ruthless about hiring. They wanted the best people in the world, and they got them. For years, Google had arguably the highest concentration of world-class engineers and researchers at any single company.

Anthropic's hiring bar is, from everything I can tell, the highest in the AI industry right now. Anthropic can hire pretty much anyone they want. They find the software companies with the best quality engineers and poach the top 10%. (This is a big problem at Stripe right now- Anthropic poaching is destroying orgs)

This matters because in AI, the difference between a good engineer and a great one isn't 2x, it's 10x. One key insight about training dynamics, architecture design, or product can be worth more than a thousand GPUs.

OpenAI, by contrast, has hemorrhaged senior researchers over the last two years.

5. Thinking Big

Google's early team was mostly young, relatively inexperienced people who hadn't been told what was impossible yet. They didn't know that building your own data centers was insane, or that a free search engine couldn't make money, or that organizing the world's information was too ambitious. So they just did it.

Inexperience is an underrated advantage. People who've spent 20 years in an industry know all the reasons something can't be done. People who haven't spent 20 years in the industry just try it.

Anthropic's founding team left one of the most well-resourced AI labs in the world because they thought things could be done better. You don't start a company to compete with the most hyped startup on the planet unless you believe you have a better idea.

6. Self-Reinforcing Data Network Effects

Once Google got popular, it got better, which made it more popular, which made it even better. This flywheel was nearly impossible to replicate because you couldn't get the data without the users, and you couldn't get the users without the data.

The same dynamic is forming in AI, but through a different mechanism. The companies with the best models attract the most sophisticated users and enterprise customers. Those customers generate the most valuable feedback and use cases. That feedback improves the model. Better model, better customers, better feedback, better model.

Anthropic seems to be actually improving based on user feedback, riding the flywheel.

7. Self-cannabalizing

In 1999, Larry Page and Sergey Brin tried to sell Google (then called BackRub) to Excite for $1 million. Excite's CEO, George Bell, said no. Their VC, Vinod Khosla, talked Page and Brin down to $750,000. Bell still said no.

The reason wasn't that BackRub was bad. It was that BackRub was too good. Excite's business model was built on banner ads and CPM pricing. They needed users to stay on their site, browse around, click on things. BackRub's search was so effective that users found what they were looking for and left. Better search meant fewer page views. Fewer page views meant less ad revenue. The better product would have destroyed their core KPI.

So Excite looked at a technology that was dramatically better for users and decided it was a threat to their business. They were right — but the real threat was that someone else would take it and eat their entire company.

The same trap existed for Google when AI started threatening search. Google Search makes money the exact same way Excite made money in 1999. Ads, impressions, clicks, time on site. AI answers questions directly. Users don't need to click through ten blue links. Every direct AI answer is a search that doesn't generate ad revenue. The rational move, the Excite move, would have been to sandbag AI integration to protect the cash cow.

To their credit, Google didn't make that mistake. They saw the Excite trap and chose to eat the short-term cost instead. They've shipped Gemini across their entire product suite, integrated AI Overviews into search, and accepted the revenue hit that comes with cannibalizing their own business model. The company that refuses to adopt the better technology because it hurts the current KPIs is the company that dies.

The ones who survive paradigm shifts are the ones willing to sacrifice today's metrics for tomorrow's relevance. The ones who can't are the Excites of the world, rational all the way to irrelevance.

What Happens Next

If the Google analogy holds, here's what's coming.

The market leader that everyone assumes will win (OpenAI) will slowly lose ground to the technically superior competitor that fewer people are paying attention to. It won't happen overnight. Yahoo was still the most visited website in America until 2008, nearly a decade after Google launched. But the underlying trajectory was set much earlier.

The winner will build an ecosystem. Google won with search, but they locked it in with Gmail, Maps, Chrome, Android, and Cloud. Whoever wins in AI will similarly expand into adjacent areas that reinforce their core advantage.

History doesn't repeat. But if you listen closely, you can hear it rhyming.