As they seem to say (a lot) these days, Hawk Tuah. Bear with me, I’m spitballing here.
+++
Feel free to ewww and ahhh as I spew technical specs. In 1979, coaxial cable strung on polls had 216 megahertz of bandwidth.
Those wires could deliver about 28 channels to your home. As limited as that seems, there still weren’t enough channels to fill the space. It took years for the Closet Conjecture to run its course. OK, the idea that we fill the space and time we have is really Parkinson’s Law. But nobody remembers it as that.
Ignoring HBO which started in 1972 (on bicycles, a neat story for another day), and, superstations — like TBS — bouncing local programming off a satellite to a national audience, cable programming didn’t really start until ESPN ran SportsCenter for the first time on September 7, 1979. CNN followed in 1980. The Financial News Network (today, you know it as CNBC) and MTV in 1981. Commerce arrived with the Home Shopping Network in 1985.
As more channels emerged to fill those 216 MHz, more people paid for cable. Those people wanted more channels. Cable expanded its bandwidth to serve them. 300MHz. Then, 400, 500, 600, 750, and, by the early 2000s, 870MHz. Each jump in bandwidth meant cable could offer subscribers more channels. And, more services. The cable plant went from coaxial wire to highspeed fiber. And, it went from being strung on polls to embedded in the ground. Those lines are the pipes that deliver internet to our homes today. Even after all the cord cutting, we paid cable providers $108B last year.
This isn’t about cable.
The New York Times went online in 1996. Those were the heady dot-com days. Amazon, Sportsline, and Real Networks went public in 1997. MarketWatch went public in 1999.
There’s a parallel here. We expanded cable and networks popped up. A highspeed (ish) information superhighway emerged and websites emerged with it. In both cases, the same information and commerce services launched. In nearly the same order.
News. Sports. Finance. Music. Shopping.
I don’t see any of this as a coincidence.
Hawk tuah. Here comes the spitball.
LLMs take mammoth piles of what we’ve written, spoken, and recorded and they structure it. The key here (at least to me) is the concatenation “we’ve.” We have. Past tense. They’re laying down pipes for others to travel.
LLMs are the next information superhighway. Smart services that relate benign babble to people are what’s coming down the pike. Generative AI is the foundational layer for a staggering array of Relationship AI smart services.
We’re not talking about the bandwidth of coax or the speed of fiber. The speed wars have been fought. We know the winners. Depending on your area: Comcast/Xfinity. Charter. Spectrum. Cox.
This time there won’t be regional winners. A handful of LLMs plus or minus two are battling it out to see who gets to challenge the Klingons for interstellar domination.
If I bet money, it would be on OpenAI. It has the most users. The biggest name. And, it’s done the most deals with publishers. Google (Gemini), Meta (Llama), and Grok (Twitter/X) have an advantage of making money on the content hundreds of millions of users give them for free. Perplexity has no data and no deal. So it has the steepest hill. Amazon is a new entrant.
Whoever wins will be the pipe that provides the underlying smarts to the apps that service people. It’s their road and that’s their toll.
This is not about ChatGPT. OpenAI’s ChatGPT isn’t the application for us. It’s a proof of concept to show other builders what they can do. Today, ChatGPT has 3 million users paying $20 per month. That’s $60M a month. Not nearly enough to run cities of Nvidia chips powered by dedicated nuclear reactors. If you multiple that by 100, that’s $6B per month. Which is $72B a year. Which sounds like an awful lot. Until you realize that Google does 5x that.
Then, you realize 100x 3 million is pretty much everyone in America. And, not everyone in America is going to pay $20 per month to chat with OpenAI or have it code something. With that business model, their upside isn’t close to Google.
OpenAI is trying to win a much bigger game. It wants to be Comcast/Xfinity, Charter, Spectrum, Cox, AND every other cable company rolled into one. Every smart app for every person will run on it.
Now do the math. I’d use multiple smart apps — each powered by OpenAI. So would you. So would everyone else. Now, OpenAI isn’t getting $20 from me. It’s getting money from each smart app I use.
And, given the history of pipes and services, we can guess the first services: News, sports, finance, music, and shopping.
Creating the nuances that that relate past tense into thematic, current smarts for every vertical is too much work. Managing relationships with every user is too much work. LLMs are in the superhighway business; not in the relationship business.
The pipe is called Generative AI. I’m going to call these smart services Relationship AI. Who knows, maybe that will stick. And I’ll be like that guy, whats-his-name who invented the hashtag. Chris Messina. More likely it will be like when the Labatt Brewing Company and others paid $7M to buy a baseball team for Toronto. And, named it the Blue Jays, hoping fans would nickname it Blues after Labatt’s #1 beer, Labatt’s Blue. Nope. They’re the Jays.
I don’t usually do this. No here anyway, but I made this slide. The kind of slide you might see in a consulting deck. Or in hallowed feeds of LinkedIn well-liked by consulting lemmings. I know argh. Holy shirts and pants what was I thinking? Sorry. In the moment, it seemed like a good idea. But, hey, feel free to share on LinkedIn and have your friends like it.
If you need a catchy sentence, to toss around at your next VC cocktail soiree, may I suggest, “Generating AI is nice. Relating AI is nicer. LLMs will structure what we know. Relationship AIs will be the smart services we’ll use. Sports, news, finance, music, and shopping will be the first Relationship AIs.”