In the mid 90s, the Netscape browser introduced most of the world to the web and brought on the Internet era. Microsoft, which was the dominant tech company at the time, famously reacted by bunding its Internet Explorer browser into the Windows operating system and started taking share from Netscape.
Eventually, Internet Explorer became the dominant web browser and Netscape was sold to AOL.
Ironically, that battle for Internet dominance missed that the most important piece of software was the search engine, not the browser. And so the winnner ended up being an entirely different company - Google.
As Mark Twain famously told us:
History doesn't repeat itself, but it does rhyme
Google's purchase of DeepMind back in 2014 was the first shot fired in the foundation model battle. Google went on to develop the transformer architecture and it shared that idea with the world in a paper written in 2017. OpenAI took that architecture and started training large language models and introduced GPT1 in 2018.
What OpenAI was willing to do, that Google was not at that time, was to train its large language models on the entire Internet, regardless of who owned the content. I believe Google's lawyers and top executives, under massive scrutiny from DC and around the world, were not willing to go that far.
And so OpenAI developed a significant lead in large language models, leading to the launch of ChatGPT in late 2022. That moment was the equivalent of the Netscape browser and ChatGPT introduced usable artificial intelligence to the world.
And, like Microsoft thirty years before, Google "woke up" and eventually caught up. With the recent introduction of Gemini 3, most observers believe that Google has erased OpenAI's lead, at least for now.
And just like Microsoft, Google is also bunding Gemini in every consumer surface area it owns. This browser that I am typing into has the Gemini Icon in the upper right corner and I use it to draw the header images to my posts.
Is there a better AI tool to draw these images? Maybe. But it isn't in the upper right of my browser and so I use the one that is.
I am not saying that Google will dominate large language models and OpenAI will be forced to sell itself to the modern day equivalent of AOL, whatever that might be.
History doesn't repeat itself.
But it does rhyme.
I've long thought that eventually LLMs would turn into a two company battle, just as operating systems and phones have. And I have long thought that it was just a matter of time until Google emerged as that second company.
And so it has.
But the thing I haven't yet figured out is what is the search engine in the AI story. Is there an application in artificial intelligence that matters more than LLMs? And if so, what is it?
Wall Street is getting increasingly concerned that the current AI mania will burst and bring the entire market down with it. Silicon Valley brushes that concern off, and VCs and big tech companies continue to pour money into AI in search of big payoffs.
So who is right?
At times like this, I like to turn to the data and ignore the prognosticators.
Evan O'Donnell is a VC and blogger who took it upon himself to build a model that looks at the rate of growth of inference token usage and compares that to infrastructure investment and comes up with some answers. This post details that approach.
But what I like most is Evan's dashboard, which you can see here.
My only critique of this approach is that the data is not real-time. Not even close. When I asked Evan about that via email this past weekend, he said:
No material update on the token numbers.
As of Sept/Oct, token consumption is growing at ~13% monthly across providers (down from 30-40% earlier this year). I'm tracking everything
One of my favorite quotes, courtesy of William Gibson, is:
The future is already here — it's just not very evenly distributed
That's how it is with self-driving vehicles. They have arrived. But not everyone knows it.
I was thinking about that at dinner last night while talking to a longtime friend who had just bought an EV and was telling me how much he loves it. And I said, "But it can't drive itself." And he looked at me like I was joking.
I wasn't.
This year, 2025, has been a self-driving journey for the Gotham Gal and me. During our winter stay in Los Angeles, we started taking Waymos over Ubers. We became so comfortable in a car without a driver that we massively preferred it.
When we got back to NYC, we missed Waymos. Eventually, we got a new Tesla Model Y with the latest self-driving hardware and software in it, and now it drives us around NYC. One of us has to sit in the driver's seat, unfortunately, but otherwise it is a very similar experience.
My colleague Nikhil posted this on his return from SF to NYC last week:
every time I come off a week of taking waymos in SF:
1. it feels increasingly strange to return to a non-autonomous city (just as it felt weird to be in cities that didn't have uber yet in 2014-2016)
2. I come away feeling like we continue to under-discuss the second order effects of self-driving inevitability + ubiquity
I think the indifference in the air is largely a function of how gradual (relatively) the rollout of AVs has been and will continue to be
NYC is a tough place to drive in. There are pedestrians and bikes and scooters coming at you from every direction. When you make turns, you have to look everywhere to make sure you aren't going to hit someone. I can't look behind me. But my car can. And so I have found that our self-driving car is able to navigate the crowded and chaotic streets of NYC so much better than we can and almost certainly better than any human can.
In the mid 90s, the Netscape browser introduced most of the world to the web and brought on the Internet era. Microsoft, which was the dominant tech company at the time, famously reacted by bunding its Internet Explorer browser into the Windows operating system and started taking share from Netscape.
Eventually, Internet Explorer became the dominant web browser and Netscape was sold to AOL.
Ironically, that battle for Internet dominance missed that the most important piece of software was the search engine, not the browser. And so the winnner ended up being an entirely different company - Google.
As Mark Twain famously told us:
History doesn't repeat itself, but it does rhyme
Google's purchase of DeepMind back in 2014 was the first shot fired in the foundation model battle. Google went on to develop the transformer architecture and it shared that idea with the world in a paper written in 2017. OpenAI took that architecture and started training large language models and introduced GPT1 in 2018.
What OpenAI was willing to do, that Google was not at that time, was to train its large language models on the entire Internet, regardless of who owned the content. I believe Google's lawyers and top executives, under massive scrutiny from DC and around the world, were not willing to go that far.
And so OpenAI developed a significant lead in large language models, leading to the launch of ChatGPT in late 2022. That moment was the equivalent of the Netscape browser and ChatGPT introduced usable artificial intelligence to the world.
And, like Microsoft thirty years before, Google "woke up" and eventually caught up. With the recent introduction of Gemini 3, most observers believe that Google has erased OpenAI's lead, at least for now.
And just like Microsoft, Google is also bunding Gemini in every consumer surface area it owns. This browser that I am typing into has the Gemini Icon in the upper right corner and I use it to draw the header images to my posts.
Is there a better AI tool to draw these images? Maybe. But it isn't in the upper right of my browser and so I use the one that is.
I am not saying that Google will dominate large language models and OpenAI will be forced to sell itself to the modern day equivalent of AOL, whatever that might be.
History doesn't repeat itself.
But it does rhyme.
I've long thought that eventually LLMs would turn into a two company battle, just as operating systems and phones have. And I have long thought that it was just a matter of time until Google emerged as that second company.
And so it has.
But the thing I haven't yet figured out is what is the search engine in the AI story. Is there an application in artificial intelligence that matters more than LLMs? And if so, what is it?
Wall Street is getting increasingly concerned that the current AI mania will burst and bring the entire market down with it. Silicon Valley brushes that concern off, and VCs and big tech companies continue to pour money into AI in search of big payoffs.
So who is right?
At times like this, I like to turn to the data and ignore the prognosticators.
Evan O'Donnell is a VC and blogger who took it upon himself to build a model that looks at the rate of growth of inference token usage and compares that to infrastructure investment and comes up with some answers. This post details that approach.
But what I like most is Evan's dashboard, which you can see here.
My only critique of this approach is that the data is not real-time. Not even close. When I asked Evan about that via email this past weekend, he said:
No material update on the token numbers.
As of Sept/Oct, token consumption is growing at ~13% monthly across providers (down from 30-40% earlier this year). I'm tracking everything
One of my favorite quotes, courtesy of William Gibson, is:
The future is already here — it's just not very evenly distributed
That's how it is with self-driving vehicles. They have arrived. But not everyone knows it.
I was thinking about that at dinner last night while talking to a longtime friend who had just bought an EV and was telling me how much he loves it. And I said, "But it can't drive itself." And he looked at me like I was joking.
I wasn't.
This year, 2025, has been a self-driving journey for the Gotham Gal and me. During our winter stay in Los Angeles, we started taking Waymos over Ubers. We became so comfortable in a car without a driver that we massively preferred it.
When we got back to NYC, we missed Waymos. Eventually, we got a new Tesla Model Y with the latest self-driving hardware and software in it, and now it drives us around NYC. One of us has to sit in the driver's seat, unfortunately, but otherwise it is a very similar experience.
My colleague Nikhil posted this on his return from SF to NYC last week:
every time I come off a week of taking waymos in SF:
1. it feels increasingly strange to return to a non-autonomous city (just as it felt weird to be in cities that didn't have uber yet in 2014-2016)
2. I come away feeling like we continue to under-discuss the second order effects of self-driving inevitability + ubiquity
I think the indifference in the air is largely a function of how gradual (relatively) the rollout of AVs has been and will continue to be
NYC is a tough place to drive in. There are pedestrians and bikes and scooters coming at you from every direction. When you make turns, you have to look everywhere to make sure you aren't going to hit someone. I can't look behind me. But my car can. And so I have found that our self-driving car is able to navigate the crowded and chaotic streets of NYC so much better than we can and almost certainly better than any human can.
So where does this leave us?
The current infrastructure spend rates are justified if the current rate of AI usage continues. If the growth rates start to decline, there could be trouble.
So we should all be watching the numbers when they come in over the next quarter.
Until then the debate will rage on.
That probably sounds crazy to many people reading this. A car is a better driver than a human?
Yes.
That's the reality of where are in 2025. Not everyone realizes it. But that is where we are.
And, as Nikhil points out, the downstream effects of this technology and behavior change are going to be profound.
So where does this leave us?
The current infrastructure spend rates are justified if the current rate of AI usage continues. If the growth rates start to decline, there could be trouble.
So we should all be watching the numbers when they come in over the next quarter.
Until then the debate will rage on.
That probably sounds crazy to many people reading this. A car is a better driver than a human?
Yes.
That's the reality of where are in 2025. Not everyone realizes it. But that is where we are.
And, as Nikhil points out, the downstream effects of this technology and behavior change are going to be profound.
Share Dialog
Share Dialog
Share Dialog
Share Dialog
Share Dialog
Share Dialog
