The Economic Case for Generative AI

Tara Z. Manicsic

All right, I like the delayed applause too, like you just realized what was happening. Thank you all for joining us today. Thank you, Martin, very much for joining us. I'm very happy to be here. I've been very excited to chat with you about this, especially because you have stated that the current state of generative AI is, and I want to get this quote right, "the start of a new super cycle on par with the advent of the microchip or the internet," so the third epic of compute, which is a very exciting and bold statement.

Martin Casado

Yes.

Tara Z. Manicsic

But I would love to walk everybody through the journey of how you got to that epiphany or realization.

Martin Casado

Yes. Is this just VC bravado, or is there something material there?

Tara Z. Manicsic

Well, and that's the thing. You have a really good amount of insights and data, and you've actually thought about this. So you didn't just have AI give you a bunch of buzzwords.

Martin Casado

We'll see what the crowd thinks.

Tara Z. Manicsic

So with all these seasons, can you give us a bit of the history of the seasons you went through to get to this point?

Martin Casado

So here's the basic argument. I probably took my first AI course in the late 1990s, around 1997. We've been doing AI, if you define AI as using data to create a model for some sort of prediction, since the '50s. And if you look at every single curve, they're all up and to the right. We talk about winters and summers, but the reality is they keep getting better. We keep solving problems, using them for 20 years now. AI has been better than we are for many things, like handwriting recognition. We've monetized it really well. Every time you do a search, it's using AI, and personalization is using AI. So all of this great stuff has happened, and yet we've never seen a platform shift like we saw with mobile or the cloud. Google, Facebook, Netflix can make money from it, but we've never seen an industry transformation. So the question we should ask ourselves is, why haven't we seen that? The answer is the economics suck. They're terrible.

One of the best examples of that is autonomous vehicles. We've poured $75 billion into it, and right now the unit economics are worse than Uber. If you're Google, GM, or a startup, you're going to run out of money, and even if you do, the business won't work. So the question is, why is this time different? The answer is the economics are better. For example, with image creation, the model inference cost is about one-hundredth of a penny, compared to a designer costing a hundred bucks. The difference is four to five orders of magnitude. We have many examples of this, in LLMs, images, code generation. In every case, there's a 4 to 5 orders of magnitude difference in economics, and every time we've seen that in the past, the industry changes. We saw this with microchips when the compute cost went to zero, we saw it with the internet when distribution cost went to zero, and now we're seeing it with AI.

Tara Z. Manicsic

I really appreciate what you're saying about the different aspects of where we can utilize compute power. When you and Marc Andreessen were talking about his article on why AI will save the world, one of the things you brought up was the huge amount of powerful compute and a ton of data. Is there another aspect that makes this time different for generative AI?

Martin Casado

It's a good question. I think it's worth being specific about what this new AI is and what it's unlocked. We like to call it the 3C's - companionship, creativity, and co-pilot. Companionship is a real use case with AI like waifu or virtual girlfriends. The retention curves on these companies are unbelievable. Creativity is another aspect. The cost of creation is going to zero, and anything digitally created is going to change. Co-pilot, like code co-pilots, is also emerging. We've hit an algorithmic and compute point that allows us to enter these new areas that computers couldn't handle before, like chatbots. We've reached an algorithmic point, a compute point, and a funding point where computers can do entirely new things, creating a real industry.

Tara Z. Manicsic

When it comes to interacting with businesses like banks, do you think people will prefer using AI companions they've already established rapport with, or will they go to traditional chatbots that may not have the same personal touch?

Martin Casado

I think these new behaviors are very powerful, and they start with individuals like you and me. The Enterprise usually lags behind but eventually figures it out. Companionship with AI, like AI friends, is a growing trend. Creativity is changing the landscape, where almost anything digitally created can be transformed. Co-pilot AI, for tasks like coding, is also making an impact. The Enterprise is taking time to adapt, but these new behaviors are taking root, and the market will dictate their growth.

Tara Z. Manicsic

Regarding the concerns around data privacy and AI, you tend to come from an optimistic and data-based perspective. What are your thoughts on the risks and concerns around AI and data privacy?

Martin Casado

I'm the opposite of a Doomer. The Enterprise always worries and tends to get it wrong. The market wins, and it's driven by our behavior. The Enterprise will figure it out eventually. There are concerns about regulations and privacy, but the market and the behavior of individuals will determine the direction. The battles are important for the future of tech, and the industry is mobilizing, with a focus on open source. We need to defend the ability to keep open source open.

Tara Z. Manicsic

You mentioned in one of your talks that this new wave of generative AI will give rise to new iconic companies. Can you provide one reasonable prediction and one really out-there prediction for the future?

Martin Casado

The three use cases that are working — companionship, creativity, and co-pilot — are going to expand. We'll have AI friends as part of our social fabric, and creativity will be transformed, especially in areas like video games. The Enterprise might take longer to adapt, but these new use cases are already showing productive growth. As for a wild prediction, I work on a video game called AI Town, and it's a 3D RPG simulation with AI characters. I'm convinced that the future will be wilder and weirder than we can imagine. Compute changes, and the possibilities are endless. It's an epochal moment, and I'm here for the wild and weird.

Tara Z. Manicsic

Thank you so much for taking the time, and I look forward to the exciting future with AI.