Aaron Levie

It's actually not even necessarily the case that the that the customer has to pick the open source vendor, they might buy it through an abstraction layer that that is letting them get the benefit of open source, but but still buy through a proprietary.

I believe... In AI? No, no, no. I believe open source causes pricing to always be extremely low.

No, not necessarily. You don't? Okay, explain. Because only meta has the scale to be able to provide... I think what Aaron is saying here, let me maybe try to frame it.

And it'll go to the cost of the compute, to be clear, with a little bit of margin for the work of- The energy, the physicality of that compute.

Now, it's important to step back and say, I still think you could probably have the entirety of the model providers make 10 times more revenue than they do today because we're just literally in like the first percent of the total TAM.

So it'd be a mistake to think that that has some kind of downward pressure in terms of the long-term economics of these businesses, especially because I think OpenAI's revenue stream is increasingly looking like a SaaS revenue business as opposed to just the API kind of token pricing. So none of this is to provide any sort of color on what would you bet on today.

I agree with Chamath that you're going to have maybe not 30 providers, but let's say you at least have five to 10 good choices all competing heavily for the next breakthrough. Like literally this morning, Google had a breakthrough in sort of this reasoning-oriented model from their Gemini family. They're 2.0, yeah. Yeah.

And so what's incredibly great is it's sort of the best time ever to be building software, assuming that you have a play in the market that lets you remain differentiated. And the key there is just do enough on top of the AI model that there's enough value there.

No, not at all.

Yeah, yeah. So I think there's... I just don't agree with the TAM compression because I think there's another kind of counter event that's happening that AI is really going after, like services. And so that then conversely expands the TAM to software where IT budgets weren't usually applied to those types of things, but we can get into that in a second. But I think we've already seen...

I think we've already seen this, though, and it hasn't exactly played out as you're saying. So Zoho is this really interesting business. It's probably a couple billion in revenue at this point. And it's basically a suite of extremely low-cost, affordable software products by category.

That's not been the reason people don't switch, though.

Yeah, I guess the counter, and maybe you'd look at an ERP system or a CRM system or something else, that is sort of- Those things are totally screwed. No, but this is my point, the opposite. The last thing you want to touch is the system that is powering your supply chain.

Yep.

Maybe, so Chamath, just to clarify, are you, there's two different ways to approach the lowering the cost of developing software. One is that it just creates more competitors in each of these categories, which then lowers my price because now there's some downward pressure. Dave's bringing up a different example, which is I'm going to build my own software at effectively the price of zero.

The challenge on this, especially the second one, is most organizations don't want to be in the business of having to think about building their own software. They just want it done for them. It's not a core part of their... Your business would be very unique relative to the broad economy, which is like, I want a place to put my CRM records. I want a place to just have my HR get managed.

And I think the downward pricing pressure due to software cost lowering makes total sense. I don't think it's a 10x factor. But, you know, I think that we've always had a long tail of applications that enterprises build. You did it in Microsoft Access. Now you do it in Retool. So the next era of that will be obviously AI built and there'll be 10 times the amount of that software.

But it's not obvious to me why that would go after the kind of core, the core systems of running a business because just most companies are like not looking to reinvent the wheel of that.

In life sciences- If you want anything in a clinical system, I don't need to tell you guys this, but you have to do a QA test on every single change that ever happens and be able to prove that you tested every single thing.

So the idea of a probabilistic AI system generating the kind of code for you in a clinical trial workflow, I just think, yeah, it's going to take a lot of change from regulators.