Hello everyone. Morning. Welcome to Cadence Live. We have a super exciting program this year. This year actually we went to two days given all the interest but to kick it off we have a phenomenal keynote session again with my kicking off with my good friend Jensen who's here, a person who needs no introduction, you know the godfather of AI himself. Please welcome Jensen to the stage.
Ladies and gentlemen, Anirudh. You want to start? Hey, this is M2000.
Oh, is that right? Yeah. Oh, we built this computer together. Yes. Three tons.
All right. Am I signing here? Yeah.
Remember Jensen signed another famous computer 2012, right?
Look at that. Let me sign my...
That computer is going to be worth a hundred times more in a few years. Okay.
All right. Great to see you guys. Which side am I? This side. This side. This side. Yeah.
Also, the other super exciting thing in '26 is now we have a lot more of our products running on the Millennium 2000 with Nvidia.
Well, we're going to have everything running. Exactly. Yeah. We're going to accelerate everything. I mean, this is such an extraordinary time. You guys know that Anirudh is the only CEO in the world that grew up in high performance computing, who grew up in algorithms, and who knew about the capabilities of CUDA long before. Yeah, it's really incredible partnership, great partnership, right? And it's really remarkable to see what Nvidia has become, you know, the most valuable company in the world. Of course, you're transforming multiple industries.
So last year, you know, we talked about at that time Agentic was just coming on, right? And of course you had a fabulous keynote last month at GTC. But as you look back, like the evolution of AI over the last 6 to 12 months, and what do you see in the next 6 to 12 months?
Yeah, in fact, if you just go back, it literally was only two years. In the two years time we went from generative AI, this whole thing started kicking off with generative AI, ChatGPT. Generative AI was such a big deal because it was able to, of course we solved perception long before that, but generative AI was a big deal because it was able to generate tokens, which is foundational to two things: the ability to think, right? You can't think if you can't create words, generate words in your mind. And then number two, the ability to interact with other tools, external tools. And so it set the foundation for reasoning, which was next. And then agentic AI, generative AI, ChatGPT really created the foundation of that. We're now at a point where agents are able to perceive, reason, and execute plans. The reason why this is such a big deal is because AI intelligence is interesting. People who know-it-alls, who understand-it-all, know-it-alls are interesting but not valuable. People who can do things are valuable. And so AI went from knowing everything, being able to spew out all kinds of knowledge and information, to now it's able to use tools, which is why it's such an incredible time for Cadence, it's such an incredible time for the industry because we can now have AI that use tools. And if you ever profile, we use agents, they're just all over our company. Our token use is going exponential. And if you profile what the agents are doing, the moment it reasons about what it has to do, it spins off tools, tools, tools, tools, tools. And we got all kinds of agents and we're working with you on chip design agents. And we're going to have all kinds of basic design agents, whether it's verification, which is the first one, ChipStack, and analog chip designers, backend chip designers. We're going to have all kinds of agents and the thing that's going to be incredible is the tool use of the EDA industry, the tool use of Cadence is just going to skyrocket.
Yeah. Yeah. That's actually a very good point because you know this worry last few months in the market that oh AI is going to replace tool use, right? So this...
Yeah, that's right. And there are, you know, of course software companies are now categorized, you need to categorize software companies a little bit better. There's some software companies that are focused on workflow automation. Workflow automation, of course, you could argue that a lot of that is going to be augmented by AI. But Cadence and many, many software companies, particularly engineering, EDA, the software is not workflow automation, it's fundamental tools. These algorithms are not going to be, you're not going to rediscover a better one. These are excellent algorithms that have been refined over time and they're fundamentally grounded and it's integrated into all of our companies and we can't have agents coming out with random other ones. We want them to use the tools, tools that we have. And so the tool use in this case, agentic systems, this agent revolution, this inflection point of inference is going to drive adoption, drive consumption of tools. We're, your industry and my industry is limited by one thing, which is the number of ASIC designers we have. In my case, the number of ASIC designers we have limits how many products we could create. In your case, the number of ASIC designers I have limits how many tools, tool use, tool licenses we need. Well, the number of ASIC designers is likely to increase by a thousand times in the near future. And so this is going to be an extraordinary time for tool industries like yours.
By the way, you had a great analogy. Remember we were talking a few weeks ago and you had this robot analogy coming to your house. I love that analogy that you know if you have a robot come to your house and it has to like warm up a food, like do you do it with your hands or you use a microwave that's already there?
Yeah. Tool use. Yeah. The ultimate expression of agentic systems. We have agents which are digital robots. We're going to have physical robots that we're working on. And suppose we just created the world's most amazing physical robot and they come to our house. Are we going to create a physical robot that has microwaves beaming out of its fingers, you know, or its hands turns into a 100-pound hammer? Or they're just going to pick up a hammer, you know, use the microwave, use the dishwasher. They're going to use tool use. And so tool use, agentic systems is defined by the word tool use. They're going to use existing tools and we need to be grounded on those tools because in the end I got to know whether the answer is right and I know how to use Cadence. And so if it uses Cadence and I use Cadence and everything is put back into Cadence, then I can come back and I can verify it, I could collaborate with it, I can fine-tune it, you know, so on and so forth.
And Jensen, it's great to work with you. Of course you are always ahead. You know, the one thing remarkable about you is not just AI outside, but you know, you're driving the industry, but also how much you apply inside to improve your own. And of course we have worked with Nvidia for a long time to always try to improve the design process. And so like we announced ChipStack as you may know like a couple of months ago with Nvidia. And Jensen, you're saying the first use case was RTL and verification because that's kind of a natural thing. And this agent, this agent knows PCI Express protocol, which is extremely complex by the way, and memory interface protocols and CXL protocols and basically the APIs of chips. This agent specializes in the language of chips and understands all the complicated protocols and it's built on top of that understanding so that it could do verification with us.
Yeah, it's really incredible. Nothing is more frustrating than a PCI Express bug. Yeah, exactly. You know, because verification that one forever. Do you guys agree? You know, one thing remarkable about chip design is how much effort we put in verification, you know, because of course software industry there's 100 to one. Yeah. It takes one person to design it, you know, 100 engineers to fix it. And that's something that the software industry doesn't do as much, right? Like the pure software. But in chip design, you know, with RTL design, the verification part is super critical.
Yeah. So, of course, we are very glad to work with you and your team on ChipStack. But now, in the last year, we have a bunch of, we have a bunch of engineers. They have all these different names. And then we have a bunch of engineers, the names Cadence.
Yeah, exactly. And they're on Slack channels with us. Slack channels with us. Virtual engineers. Virtual engineers. And I love that. I love that all the email addresses that start with Cadence, you know, they do exactly what I want. Yeah. Exactly.
But Jensen, even the real engineers from Nvidia, I work with them 24/7. Just in the future, my Slack channel will be 99% Cadence.
Yeah. Thank you. Yeah, I got Cadence agents talking to other Cadence agents.
By the way, even the real Cadence employees do what you want, not just the...
Thank you. Thank you. Thank you.
Now on the, I know recently, more recently, you know, we have two new super agents, you know, that we're working with Nvidia. Of course the back end, because you know what I keep hearing from customers, you know, like especially Nvidia, you have this amazing Cadence now every year and the chips get bigger so the amount of PD work goes up in the back end. Yeah. And then the other big thing is analog. You know, analog design has always been very difficult, very difficult to automate. So we have this new Virtuoso agent, V-Stack, that we work with your team. So I'm super excited to what happen in these two new areas.
Apart from those, those are really great, great choices. The challenge with physical design, you know, there's an impression for people that are outside the industry that we design the chip and then at some point you have a net list and you take that net list and you do place and route. Well, these chips are so large, you know, if you think about chips the way you think about cities, you know, each one of our chips is like all the cities in the world combined. And you're doing city planning while the buildings are being created, while everything is being designed. And so we plan physical design from practically the beginning of the design process because that's how long it takes. You're in the physical design process for two years. And the reason for that is because of course the amount of technology, the amount of components that fit into the physical design that you have to place is large. But you have a fundamental bin packing problem and you're trying to optimize for size, you're optimizing for manufacturability, you're optimizing for timing. Meanwhile, everything's being designed at the same time. It's an incredibly challenging process. And so now we have agentic systems that help us basically work 24/7, but you're parallelized to work across thousands of agents. And I'm hoping, and today we're scaling it up to tens and hundreds, but we'd like to scale it up to thousands and hundreds of thousands of agents all working around the clock trying all kinds of different permutations so that you could solve this many variable, impossible to find a perfect answer, you know, optimization problem. And so this is an ideal place for agents. Analog design is very similar to that. The design space is extremely large and it's hard to know when you're pushing the limits of analog design. We're basically at the limits of physics and there's a lot of different permutations, a lot of different ways you could design the analog circuitry. Having agents explore the space for us, you know, simultaneously design for signal integrity, for power, for size, for manufacturability, exploring all of that simultaneously across all the agents. These are all exploration problems. And so I think there's a misunderstanding about fundamental design that you have a problem statement, you come up with a solution like you're solving a puzzle and that's the answer. Well, it turns out, you know, there are a thousand plausible answers. A million plausible answers. The question is which one are you going to go and optimize towards? Well, having more agents having you explore the design space, the better. And that has been history of EDA and chip design.
That's right. Because like you have said before, there's no one right answer. There are thousands of right answers. You have to find the one with the best PPA. That's right. So, anything that helps find the best right answer is invaluable in the design process. It's a searcher algorithm and one of the most complex search algorithms in the world. We're going to need a lot of agents to help us do that.
Yeah. Yeah. Super, super exciting.
So, Jensen, also you talked about at GTC, by the way that was a great keynote. This is just a warning to all the companies: your tool use is going to skyrocket. You better secure your licenses soon.
Yeah. Call your Cadence salesperson right away.
Yeah. Like the tool use, which is I think we all know that the number of computers are going to skyrocket. The number of licenses of software tools are going to skyrocket. Agents, we've historically been limited, both of us have been limited by, you know, butts in seats. In the future we're only limited by compute. Yeah. And so more virtual teams.
