Why Fortune 500 Companies Are Quietly Switching to Open Models

Jason Calacanis (0:00)

All right, everybody, welcome back to This Week in AI. Really excited about today's guest, we have Jeffrey Morgan here, the co-founder and CEO of Ollama. Jeff, welcome to the program.

Jeffrey Morgan (0:11)

Great to be on.

Jason Calacanis (0:11)

You've listened to it for a little while, I understand you listened to This Week in Startups here. For a couple of years, you're on your second startup, yeah?

Jeffrey Morgan (0:19)

That's right, yeah. First startup started in 2013, and so it's a place just to hear from other founders.

Jason Calacanis (0:25)

Yeah, and now we have 18,000 podcasts and nobody's getting any work done. I apologize to everybody in technology for the number of VCs who now have podcasts. It's almost like you have to have a podcast if you're a VC because you have no other work to do every day. Thanks to our friends at PayPal, the exclusive sponsor of This Week in AI. Pay zero processing fees on your first 100,000 ineligible PayPal payment volume. Find out more at paypal.launch.co.

But you're crushing it with Ollama, right? Am I pronouncing it correct, Ollama?

Jeffrey Morgan (1:02)

Yeah, or Ollama. Ollama.

Jason Calacanis (1:05)

So Ollama is the llama and you added an O to it, Ollama.

Jeffrey Morgan (1:09)

A little bit. It's funny with Ollama, we have this character which is our logo and puts a face to the name for open models.

Ollama was a common name for open models. And yeah, it was local llama and Ollama was a product. The domain was available and we ended up being able to get the GitHub and the Twitter handle and we said, hey, you know, it's pretty easy. Most people know how to spell it once they hear it. So the name really stuck since.

Jason Calacanis (1:36)

If we've learned anything in the history of startups, it's that if you get a great domain name for your startup, it usually has two qualities. One, you can say to somebody over the phone and they can type it in and not misspell it. And two, it's available for eight bucks. These are the high art of branding when it comes to startups. Or you can just get the .a or the .co fake until you can buy it.

So tell everybody a little bit about Ollama and what you're doing. Obviously, folks know this is open source models. But you guys got started very early on this, I think two or three years ago, yeah?

Jeffrey Morgan (2:09)

Yeah, just over three years ago. And Ollama is the easiest way and the most popular way to build with open models. We launched it three years ago as an open source project. And there was a few initial models like the Ollama model, Microsoft had put one out, and then Ollama 2 came out. And that was the day we launched Ollama. And since then, it's grown to over nine million monthly developers. We launched Ollama's Cloud, so beyond an open source project to a Cloud. And that's really exploded this year with the introduction of these really powerful open models.

And it's grown over 200x since the start of the year.

Most interestingly, open models are very popular with individual devs because they're free, you can download and run them. But they're also super popular in the enterprise. And over 85% of the Fortune 500 have deployed Ollama in their networks. So it's been a pretty awesome three years.

Jason Calacanis (3:03)

Yeah. And the change of enterprises embracing it is relatively new. In the last six months, I think people started to get token shock.

You get this bill for the number of tokens you're using and you're like, hey, wait a second, I just spent a million dollars on tokens for this project in September and I didn't get a million dollars in output. I can't tell the difference between what my developers did that month with tokens or without tokens. So that actually, I think, got everybody open source model curious. Am I correct? Yeah.

Jeffrey Morgan (3:36)

Open models, they've always been an alternative and a popular way to access AI privately or if you're doing research, having the weights lets you fine tune and customize the model but it's really only in the last six months where, even three months, where taking an off the shelf open model and using it for these high volume expensive use cases has completely hit the market. And it really is driven by, I think, two things. One is the cloud code moment that happened over the holidays, which turned everybody into massive consumers of tokens.

And then the fact that we're working on harder and harder tasks and the cost per task and just the general cost per month of these businesses is skyrocketing. And so that's where open models come in. You can lower the costs. But that's really only happened in the last three months.