Today's program is being recorded and webcast live. Please welcome Tushar Jain, Vice President Synopsys Investor Relations.
Thank you. Thank you, everyone. Good afternoon, and welcome to Synopsys' Investor Day. We're so glad to have you here. It's great to see so many familiar faces in person rather than on Zoom. It's a very exciting day for all of us here at Synopsys, and we thank you for joining us in person and for those joining us online. Before I go any further, I need to read a short legal disclosure. Synopsys will discuss forecasts, targets, and other forward-looking statements during today's presentation. While these statements represent our best judgment as of today, they are subject to risks and uncertainties that could cause actual results to differ materially. Important factors that may affect our future results are described in our most recent SEC filings.
We will refer to certain non-GAAP financial measures throughout today's presentation, and reconciliations to their most directly comparable GAAP financial measures can be found in the appendix section. A replay of today's event and the presentation materials will be available on our website at www.synopsys.com. All right. With that out of the way, as I said, we have an exciting agenda lined up today. Sassine is going to cover Synopsys' next phase of growth. You're going to hear from many of our customers, and then Shankar is going to join him on stage and go deeper into our AI platform. Following that session, we'll take a short break, and then Shelagh's going to come on stage and put all of that in the context of our financial model. We'll end the day with a Q&A session. With that, let's get started.
[Presentation]
[Presentation]
Come myself. Hello, and welcome to Synopsys Investor Day. I cannot be more excited to be here in New York City with our shareholders and many who are joining us online. Before I get started, I want to send special shout-out to our employees. I know you've been anticipating this day as much as our shareholders, and I cannot thank you enough for the trust in the strategy, the agility and courage to act and shape the future of our company and delivering with excellence. So thank you. I cannot be more humbled to be part of this company. In my presentation today, I want to talk about the what is changing around us and the why and how Synopsys is positioned to maximize this opportunity. This year is a special year for Synopsys.
It's our 40th year anniversary, and it's very hard for companies for 40 years to start with a disruptive technology synthesis and maintain the leadership position, all anchored with staying on the leading edge of innovation and delivering the technology that our customer must have in order to deliver the best product that they are designing. The other important milestone for us this year is the year one of the new Synopsys, of the fully integrated Ansys and Synopsys. So we're very thrilled to celebrate our 40th year anniversary, as well as our new company. Over the last 40 years, we were able to deliver to our customers with different eras of technology disruption, innovation that was necessary, and we became mission-critical to our customers' success in delivering these differentiated products. In the era of pervasive intelligence
The need to go beyond the silicon innovation to silicon to system is a necessity because the optimization cannot happen at one level of the stack. It has to happen at the entire stack, from silicon to systems. Over the last five years, we've been on a mission to transform the company. Five years ago, Aart entrusted me as the President, then CEO, to lead the next chapter of Synopsys. During this time, with discipline and courage, we've been able to make number of portfolio decisions with divestitures of assets that we did not feel they were needed to deliver from silicon to system, or they were not the right asset for Synopsys to hold. At the same time, we opened up our balance sheet and acquired one of the most essential assets for the physical AI era.
When I say physical AI, that same asset, Ansys, is necessary to advance the chip design process as electronics and physics are merging. When we talk about Synopsys as the leader in engineering solutions from silicon to systems, it is very rare for companies to have three parts of their portfolio, and each part holds the number one leadership position in its market segment. We are the number one in EDA, in silicon IP, and simulation and analysis, or, as we call it, SNA. These assets are becoming critical as we envision the world, as we envision the future, the future products, as digital AI and physical AI are converging. The one common thing to drive the digital AI and physical AI is advanced silicon.
You are going to hear me talk repeatedly and more and more about purpose-built silicon, because the world of just a generic merchant off-the-shelf silicon is not going to deliver the efficiency required and needed in order for the workload and the applications to be optimized all the way along the stack. That optimization is needed and necessary in order to drive the right cost, the right competitiveness of the product. When we think about silicon to system and the increased complexity, pace, and cost of these designs, the need to co-design becomes a must. What does co-design mean in engineering? It means you are optimizing in one domain and having the adjacent domain taken into account, so you are not building too much margin in your product. That is what Synopsys is thriving to do.
That is what we call we are re-engineering engineering in this era of pervasive intelligence.
How do we bring the portfolio together to enable our customers to build the most differentiated product with the lowest cost, on time, high-fidelity product with our portfolio? When we talk about co-design and digital twinning and modeling of the end product, it is all to deliver better, faster, cheaper products for our customers. We serve a global R&D spend of $1.7 trillion. If you look at the various industries on this chart and the dollar they invest in R&D in order to build their products, this is where Synopsys' opportunity comes in. About 90% of that $1.7 trillion leans on traditional way of building a product, physical prototyping. Only 10% of that $1.7 trillion is using technology in order to build their differentiated product.
The trend over the last five years, more money is coming towards the technology, and that 10% portion of the pie is growing, just simply driven by the complexity of these products. You cannot build an EV or a robot or a drone or a chip by having a physical prototype. You need to virtualize, model, design, simulate before you build and test. That is the opportunity we have, and we truly cannot be more excited about having our strategy as AI is transforming engineering from silicon to systems and expand our opportunity. At the silicon level, more chips are required to drive this intelligence and to compute for this intelligence. These chips are increasingly becoming application-optimized silicon, custom silicon, and it is evident by the number of OEM and system companies are trying to invest in building their own silicon.
The reason I am saying trying to invest, it is not easy to build the most advanced silicon to support the AI and the system requirements. At the system level, these systems are becoming more intelligent, AI-driven, a lot of software. The increased demand for simulation and analysis before you spend the money to build the physical prototype is becoming as well a necessity, and more important. The physics-aware co-design crosses silicon and system. On the chip level, the need to have physics simulation with electronics, and physics can be fluid, structure, thermal into electronics is already happening. At the system level, having the representation of physics as your end product operating in the real world is another driver as we move into physical AI. The one thing that we are very excited about is the opportunity that AI is bringing.
On Monday, and Shankar will talk about it more, we announced our AI platform, where we have in multiple areas and domains, an agent engineer that is able to call many of the sub-agents and the tools to perform tasks autonomously. That is only possible if you can trust the results of what the agents are producing. First-time right product is essential. In order to build a first-time right product, you need a trust in physics as these agents are generating output and outcomes. Now let us jump into the actual business, and given the tailwinds, how is Synopsys capturing these opportunities? I will go over the EDA, SNA, and design IP. As you know, we have two segments. Shelagh will talk more about the segments that we have, which is design automation and design IP. I am going to start with IP.
The reason I want to start with IP, there is no better place to describe the market than our IP position. Our IP position, when our customers are even thinking, before even committing to design a chip, they come to Synopsys and they ask about maturity, readiness of a node or a foundry, about the connectivity to connect these chips together, and how is the ecosystem is thinking about them. So our IP position gives us the best view of what is happening in the market. So I am going to start with IP for that reason. What is Synopsys IP portfolio? We have what is called an interface IP, which is an IP that connects a chip to chip or chip to a system, and we have the leading position in interface IP. We have the leadership position in foundation IP. What is a foundation IP?
Think of foundation IP as the bridge between a foundry process technology to design. It is the library. When you are a foundry and you are building the next process technology, the way to represent it in design is through the foundation IP, and we have that leadership position in foundation IP. Talking about foundries, we have more than 10 foundry support in our IP business, about 80 process nodes, and about 3,000+ IP products in our portfolio. As we think of data center, and you are hearing many of the leading silicon companies starting to position themselves as silicon to system companies. Why? Because when you think of data center, you cannot think of the chip in isolation. You have to think of the data center itself, the whole system.
Now you need to take it from the data center to the rack, and how to optimize and design the rack as a whole system. At the rack level, then you go to the blade. Inside the blade, there is the compute, the networking, the memory. Then there you can double-click into the chip itself. I want to spend some time on this picture right here because it is very representative of what do I mean by a general-purpose merchant chip, and how are our customers differentiating? Because the chip itself, from an architecture point of view, they all look the same. You need an AI accelerator. You need a CPU. You need a memory. You need a networking. You need bunch of interfaces to move data. If the architecture, you cannot be too innovative or creative with the architecture itself.
Where the innovation comes in is the workload down to the architecture, down to the implementation of the silicon. So when you look at such a picture, everything you see in purple in here are our interface IP. So big part of the system is coming from Synopsys when you are designing that advanced SoC or advanced system. The interface IP has multiple standards, and I am going to emphasize standards. The reason those standards are important, if you are a CPU supplier and you are building your own accelerator, it is important for you as a customer to have optionality and make sure that these different components can connect together. That is where Synopsys comes in with the interface IP portfolio.
We build based on a standard, and we ensure interoperability from the host to the other part, if it is a chip-to-chip or a chip-to-system type of an integration.
The reason customization is becoming very important, many times, if you buy a merchant chip, as I am showing you in here, sometimes the bottleneck can be the interface. That is where you are unable to move enough data. Sometimes you do not need that expensive accelerator or CPU to be idle for your specific workload. Therefore, as a system company, you are trying to optimize based on your architecture. The challenge of that, the interface IP business, when I say it has been built on a standard, the traditional process for a standard, there are standard bodies. They decide what will a next UCIe, CXL, PCIe will look like. Once the protocol definition is done, the ecosystem gets enabled to start developing, then the IP is available. That is the business we have been in now for 28 years with our IP business.
Standards get defined, you build to the standard, you validate, you provide it to your customer. The AI leaders, they are not waiting for a standard, yet they want the interoperability of a standard. They know their workload requirements, and they get started. The workload requirement define the system architecture. Then they are expecting a Synopsys IP to be available way before the standard is defined. You can look at this as a massive opportunity for Synopsys or a nightmare of how to manage to deliver on a standard and customize way before the standard is defined. The custom silicon opportunity, I am sure you have your own numbers, absolutely increasing. These are the forecasts by 2030, which is 6x. The drivers are supplier optionality, cost, workload efficiency. Again, that is why our customers are building and heading towards purpose-built silicon, is to address these exact challenges.
Now, I'm sure you follow, you see, you read, all hyperscalers are building their own silicon. If you see the words that they're using in here, it's all about strategic flexibility and supply chain leverage. The cost of ownership from Andy, from Satya, is optimizing the architecture. That trend, we've seen it. We've been playing in that trend. What we have, and some of you reminded me earlier, decided to do is about a year ago, we said we need to adapt our business model because that's an amazing opportunity, and Synopsys is truly the enabler of all these customers and more to build their own silicon. As customers are buying merchant, doing ASIC, building their own, that build your own silicon cannot happen without Synopsys customization of that IP.
What we have decided to do and have been communicating in that language, Factory One, Factory Two. Factory One, think of it as our standard-based IP, where you wait for the standard, you build it once, you sell it many times. That's a fantastic business for Synopsys. We will continue on feeding and investing in that business because this is beyond just data center. Automotive, industrial, mobile, consumer, all these chips need a standard. So the focus is not only on the data center opportunity. There's the rest of the market, which is fairly significant, that requires that Factory One build once, sell many times. So I don't want any confusion. We'll continue investing in this factory and leading with our IP portfolio in this factory. Then we start talking about Factory Two, where we build an application-optimized IP, AOIP.
An application-optimized IP is we build it for a specific customer requirements. I'll describe in a little bit more details what does that mean to build an IP in Factory One, build an IP in Factory Two from an engineering point of view. The business model, the first one, Factory One, is you license once for a program, and if there is any NRE, we'll charge based on an NRE. In Factory Two, there's the license per program. There is customization fee, and you see here it's different than an NRE. There's a royalty. The reason there's a customization fee and not an NRE, the scarcity of our resources needs to be put and placed on the highest opportunity as we open up Factory Two. Factory Two, the reason we can do it is the scale that we have with our IP business.
We have a massive investment position in the market that is giving us the opportunity to be able to support both a Factory One and a Factory Two. What's the difference from a customer engagement? Again, we'll maintain both. A Factory One will start with the standard being defined, the IP gets developed, the silicon gets validated with test chips, and we provide it to the customer, and we license it broadly. In a Factory Two model, we work with the customer very early in understanding their workloads. We become part of their system definition. Based on the workload, we co-design the IP with the customer. We'll be part of the IP integration with the customer and the system validation and production. This is not only an IP opportunity for Synopsys, this is IP, EDA, and SNA opportunity for Synopsys.
Because as we build the IP and they are doing the system validation, we are taking into account thermal, packaging, stress, how to cool off the system. So it is an excellent opportunity to embed ourself inside our customer workflow and deliver to this opportunity. So that is the Factory Two. I remember as well when we talked about it, many doubters. How will you ever change a business model that has been established for three decades? We do not see it. We do not get it. There is no way the customer will pay for this. Do you have the skills to do it? I am so glad and happy to report today, we have committed agreements with compute leaders, with ASIC leaders, with connectivity leaders. The press release you saw this morning is not a one-time, one customer, end of story.
Typically, back to interoperability, a system company or a leading system company, they want their ecosystem to be on the same IP. So as you are working with the system company, they pull you with their other supplier, be it an ASIC or a connectivity, to ensure they are working with you and your IP is interoperable with what they are using. This is what we released this morning. This has been a work of many, many months, not focused on the dollar and cent, focused on the how you would do it inside my engineering stack. How will Synopsys team deliver with high level of confidence to my chips? I know just in the brief few minutes I mingled with you, many questions. What is the duration? How about this? How about this? How about that?
I cannot share many of the terms of this agreement, and I hope you respect that because terms of an agreement between us and a customer are confidential terms. But I can tell you the following. This is a multi-year agreement. It is a multi-generation agreement. The $1 billion that you saw, it is a license fee. Remember, there are three layers. There is the license fee, then there is the customization fee, and then there is the royalty. The $1 billion is a license fee for multiple generations of Graviton, Trainium, and Nitro. The reason all three of them, because each one of them has a need to connect with another chip in the ecosystem. The other agreements we closed are part of that ecosystem for Amazon.
We will talk more as we wrap up this session around how meaningful that is for Synopsys and leading into the AOIP domain.
Now, the other driver, including system OEMs like Amazon and other, is foundry optionality. It is very important for customers, especially now more than ever before, given the shortage of supply chain, the shortage of silicon, and as you start optimizing at the system level, the system is multi-die advanced package or chiplet or 3D IC. It gives you a great opportunity to have optionality in the ecosystem. But in order to drive that optionality, you need a company like Synopsys to have the IP ready, available, tested at all the advanced foundry leaders. When we say we are the on-ramp to foundry, we are the on-ramp to foundry, back to foundation IP, which is the bridge, and the interface IP that is needed to connect the chip-to-chip or the chip to the system. Obviously, TSMC is the acknowledged leader in advanced process technology.
This is a quote from Kevin emphasizing the importance of Synopsys interface IP along with the foundation IP. Of course, the relationship with TSMC or any foundry is not only about IP, it is about IP and EDA enablement in order to drive that innovation moving forward. The other questions that I have gotten from you in an intense fashion over the last year, you have lost Intel. Are you at Intel? Are you on 14A? Are you doing 18A? Remember my answer. You cannot be in the foundry business without having Synopsys IP. I do not care what others are saying. We have actually next year will be a 20th year anniversary when Synopsys and Intel got married. We became a primary partner. The depth of the engagements, the breadth of the engagements are very well acknowledged by Intel and by Synopsys, the importance of both companies.
Now, I hope the video I am about to share with you will reduce your anxiety, and hopefully the questions will be less about are you working with Intel, to wow, we are excited about the relationship and continuation of what you started 20 years ago.
[Presentation]
Before we go to Samsung, I hope you heard the breadth from Ansys physics to the AI portfolio, to EDA, to package, to IP. That is the breadth and depth of the engagements we have with Intel product and Intel Foundry. The other foundry, Samsung, back to optionality and most advanced silicon, is another outstanding relationship we have in the ecosystem with Samsung. Similar to Intel, similar to what we do with other foundry, is how to engage early through DTCO, Design Technology Co-optimization, to develop, validate the process technology, then the IP, then the IP ramp. With that, let us hear from Jinman.
Hello, I'm Jinman Han, President of Samsung Foundry. Investor Day is usually about numbers, roadmaps, and the future. But before getting into any of that, I would simply like to say congratulations to our good friends at Synopsys on this exciting occasion. As AI continues to expand across industries, from data centers and automotive to physical AI, Samsung Foundry has evolved beyond just being a wafer supplier to become a strategic partner, providing comprehensive system solutions to the customers. At the heart of this transformation, our deep IP collaboration with Synopsys, Samsung Foundry has secured a robust portfolio of Synopsys IP across all processes, from mainstream to leading-edge processes. More recently, to provide optimized solutions for AI applications, we have been deepening our collaboration beyond standard IP to develop customized IP.
By combining Synopsys' proven design IP and EDA tools with Samsung advanced process and packaging technologies, our AI platform collaboration provide end-to-end solutions that support complex custom SoC designs, including HPC and 3D IC. This enables our customers to bring optimized products to market faster and with greater confidence across a wide range of industries. The best partnership, much like the best chips, are built layer by layer. We are proud of what Samsung and Synopsys have built together, and even more excited about the many layers of innovation still ahead of us. Once again, congratulations to the entire Synopsys team. We look forward to continuing this exciting journey together. Thank you.
All right. Now to wrap up our IP section. We will be raising our long-term guide from the mid-teens to high teens for IP, driven by more design starts, AOIP expansion, this factory is firing up and accepting and ramping up on customers, the multi-foundry enablement. The projected growth for AOIP by 2030 will be a billion dollars based on, so this is a line of sight based on the current contracts and commitments that we have with customers. The ambition is to have that business where royalty revenue is greater than the license revenue. I know a number of you asked as well, Will you take a dip in your revenue as you're building up the royalty over time? Of course, royalty will ramp up over time as our customers go into production. But there is no dip.
We're raising from mid-teen to high teens with AOIP factory delivering to a billion dollars by 2030 based on the current customer engagements that we have. EDA. I set up the stage for IP and the need for customization, application-specific chips, et cetera. That chip cannot happen without a fast innovation in EDA. The complexity of these chips, the complexity of these systems, they need to implement whatever technology to be with or ahead of the customers as they're architecting their next system, where a lot of our investment and leadership in EDA is coming in. I'm sure you've seen many versions of this slide.
If you look at the days of AlexNet or AlphaGo to Astra, and the massive requirement need for the compute in order to deliver to that intelligence and reasoning, the chips that deliver to it, say, the TPU of 20 billion transistors, to right now a heterogeneous multi-die system with hundreds of billions of transistors. In order for that to happen, you need EDA to lead and deliver on multiple vectors. As you think of an advanced multi-die system, we talked about AOIP, we talked about the memory customization that is required, the whole advanced packaging requirement. Physics becomes essential. The big challenge, not architecting the system, is manufacturing that system with reliability.
What happens is when that system is operating in the field with the intense workloads, that system overheats, that system will fail, unless you are taking all these design into account, the physics impact into account during the design stage. Our position in EDA is truly unique. Starting from the core, the core EDA platform, which is the leading franchise of what we call a hyperconvergent flow for the best PPA. Multiphysics fusion, this is the expansion with the Ansys portfolio. Hardware-assisted verification is more important than ever in order to validate this complex system. Will the AI workload software work when you bring that silicon back? I do not believe the pace of innovation and shorter design cycle that we talked about in IP, which applies here, is possible without bringing more and more sophistication with AI. Let me click through each.
In the core EDA franchise, the innovation vectors are the agentic AI automation. PPA leadership is where customers make the final decision. You cannot have a good enough performance and expect you are going to invest many hundreds of millions of dollars to manufacture the chip. You need to deliver the best power, the best performance, the best area of the chip. How to optimize the system with a multi-die advanced package. The fusion of physics Our leadership position, number one position with Fusion Compiler, 3DIC Compiler. PrimeTime for timing signoff. VCS Verdi for functional simulation and debug. PrimeSim, which is a transistor level simulator. RedHawk, the industry standard signoff for thermal.
About maybe a month and a half ago, when Jalapeño was announced, there was this simplistic extrapolation that if a model can build software, therefore the model can build a Fusion Compiler or a PrimeSim or VCS, and EDA is doomed because a model can create, because all you do Synopsys is build software. The model is going to build that software. That simplistic extrapolation cannot be more far off than reality of what customers need using the power of the model. But the essentialness of the physics and what we generate and what we sign off before you go to manufacturing is more needed than ever. Same thing, here is a quote from OpenAI, Richard Ho, who is the head of hardware, where he emphasized the essentialness of the Synopsys EDA in building that chip.
Not only that chip did not happen through magic, there was a lot of effort in bringing EDA through the design to generate and to sign off before, in this case, handing it over to Broadcom as the back end ASIC partner for them to implement the chip. Broadcom as well, longstanding relationship across the portfolio, and it was used throughout that particular chip. Now, what got the attention of many is the time to design the chip. I will talk more about the AI contribution to the time to design the chip. What went under the hood as a foundation to all of it is the EDA, the hardware, and the IP to deliver such a system. Multi-physics fusion.
The whole thesis with Ansys was, at some point, the monolithic chip, the scale will not necessarily hit the wall that you cannot innovate further, but you need to look at it from a system level. The scale complexity, while it is continuing to advance node to node to node, the architecture of the system moving from a 2D to 2.5 to 3D is what kept pace in order to deliver these AI HPC chips. So both the scale and systemic complexity. The moment you think systemic complexity, you are talking about physics challenges, stress, warpage, thermal, photonics, electromagnetics, power integrity, signal integrity. This is where Synopsys has seen this trend a decade ago. If we go back to the 2015, 2016 era, we have the complete stack. The way our customer use them is through connecting them through their own CAD and workflow.
Then it moved to a fusion architecture, where you start fusing engines from signoff into design, so you have a convergent flow. You are not getting surprised later, and you iterate. In 2018, this is when our partnership with Ansys started. We took their electromagnetic and voltage drop engine into Fusion Compiler to address that challenge. Then, of course, later, as we are now integrating Ansys, and in the first wave of product releases, that technology, the multi-physics technology of Ansys now is fused inside the digital implementation platform. There is no one in the EDA industry can claim this. This is unique to Synopsys. This is the differentiation of our platform. The key is not only the design, is the multi-physics timing power signoff engines that are embedded and fused inside the platform. I started with thanking you for being here and thanking our employees.
When we said we are going to release the first wave of products nine months after closing the deal, there was a lot of doubt because it is a massive effort. Massive effort. We did release the first wave of products. As you read, what is the value? Better PPA, convergence with signoff, better outcome. So it is better, faster, better, faster, et cetera, to achieve the outcomes that you are trying to get to. Since then, last earnings call, I mentioned that we have five customers in early deployment of the technology. It expanded to that broader list of customers since then. Why? They are seeing benefits. 10x faster photonic simulation, 3x higher fidelity for 3D geometry. You click through it, better PPA, better turnaround time. With that, numbers are going to show that traction.
In 2027, we will see revenue synergy from that first wave of integration between Ansys and Synopsys. Recall we talked about $400 million synergy by 2029. Shelagh will talk in more details what is it that we will see in 2027, and how does it build our confidence toward the $400 million by 2029. Given the first wave of customer engagement and early deployment in production in 2026, we know we will be able to achieve revenue growth from that solution, which is the synergy between Synopsys and Ansys in 2027. Hardware-assisted verification. With hardware-assisted verification, the complexity of these systems. Remember, the reason our customers they build a customized silicon is to optimize between their software, their AI workload, and the chip itself. The vehicle to do so is emulation and prototyping. You prototype the chip before the chip is there, and you start running workloads.
In order to do so, you need a system that has a very high capacity, remember, those chips are massive, and can run at a high speed, that you can actually bring up the software and validate it. A cool example here, actually. There is this startup that came out of stealth mode a few months ago, and I got many questions from investors like, Hey, there is this company, Etched. Do you guys work with them? We worked with them at day one as they are thinking about their own company and how to build their silicon using our EDA, using our IP, and how to validate it. And here, what is exciting was they were big users of our ZeBu platform. They were able to bring up their workloads in weeks, and weeks in here was about 44 days.
In 44 days after the silicon came, they were able to bring up their software. That is a process it would have took about six, seven months. By the time you bring the chip, you start running a lot of the software and tuning, and that took 44 days. Why? Because they started in parallel. They started writing the software, validating the software before the chip was there. So that is a great example of customers using both technology together. Our portfolio span from the emulation to prototyping to what we call EP, which is a hybrid emulation prototyping system. That to provide our customer flexibility. Capacity is our differentiation. We differentiate on capacity and system-level performance. That is the use case that is the sweet spot for Synopsys. That is where we differentiate. That is why our customers buy our system. The software-defined HAV. At Converge, I explained.
Our customers make massive investment in dollar to buy these systems. We are innovating at the software level, so the customer does not have to refresh their hardware with every cycle. They will buy the software that optimizes and increase the performance while maintaining the same system. That is a high, high value to our customers. Now, we monetize at the software and the actual hardware that we ship for the customer. And the need there, the demand for more and more expansion of the hardware cannot really be more insatiable than it is right now, given the complexity of the software that you are building on the chip. While I am not announcing officially the next hardware system, but coming soon, first half of 2027, the code name is Artemis. It is our next ZeBu system. And what is the expectation of the next system?
Larger capacity, higher performance, and a reliable and best TCO for our customers. I mentioned the complexity of delivering on these chips and multiple chips in a system hasn't really been as complex and as intense as it has been now. How do we deliver to it? Our customers are looking for every opportunity possible to introduce AI into their workflow and lean on Synopsys on how to automate further, because there what you're dealing with, not only complexity. As more system companies designing chips, the scarcity of resources and people that they know how to design these chips are not readily available. With that, let me bring Shankar to go through our AI platform and strategy. Shankar.
Thank you, Sassine, and thank you all for joining us here today. I wanted to start by going back to a theme that Sassine spoke about earlier. We are in this incredible era of intelligent systems, where silicon engineering and system engineering are coming together much closer than ever before. You cannot build a die without thinking about the package in which it will reside. You cannot build a package without thinking about the blade in which it will reside, and you can't build a blade without thinking about the rack in which it will design. This type of co-optimization that needs to happen all the way from the silicon die-level design up all the way through a system-level design, like a rack design, is really opening up tremendous opportunities for Synopsys because of the co-design that is needed.
With the portfolio we have of EDA software, our hardware solutions, our simulation and analysis solutions, and our IP solutions, we believe we are uniquely positioned to deliver the silicon-to-systems continuum. The designers of these intelligent systems are struggling with multiple challenges. The complexity of both silicon design and system design has to be tamed because it's compounding generation over generation. The speed at which these intelligent systems need to be delivered is accelerating because market windows are shrinking. We talked about building these chips and systems in three-year cycles just a few years ago, and now we are talking about building them in 12 months with a strong desire to build them in nine months. Last but not least, the cost of a failure, the cost of a mistake, is incredibly high because most likely you will miss a market window.
The need for getting first-time right silicon and system and software all at the same time, the stakes have never been higher. On top of all these challenges that design teams are facing with complexity, with cost, and schedule, there is another huge challenge, which is the engineering resources needed to build these systems. There are multiple reports that talk about the engineering shortage, and a recent one from Goldman Sachs further highlighted this, that at the top line, the number of companies building intelligent systems, hyperscalers, system companies reaching deep into silicon design, is growing. But the growth of the human capital to meet this design is not growing at the same rate. Therefore, there is a significant gap between what the human capital and capacity we have in terms of engineering and the top-line resource requirements to continue this incredible build-out that we are all experiencing.
This gap is really going to be closed by two things, more automation and more AI. With the recent advances we see in agentic AI and all the frontier models, we are very confident that we now see a path of how the current human capital can apply these technologies and really meet these stringent requirements in terms of engineering resources. I want to take you through a little journey of Synopsys' AI story, which started almost two years ago with respect to generative AI. At that time, models were good, but they did not have a whole lot of reasoning capabilities. They did not have much orchestration capabilities. What we could do with these models essentially is build useful copilots. With heavy context and prompt engineering, we provided useful assistance for engineers.
We were able to even provide some level of automation for specific tasks which were repetitive tasks. While these were appreciated by customers, none of this really changed the engineering workflow in any significant way, and thereby did not really add significantly to the capacity of an engineering team that was trying to get more and more done with less. But as models have evolved dramatically over the past two years, in terms of the ability to launch and orchestrate sub-agents, the ability to do extraordinary reasoning and problem-solving, we are extremely excited about what the next wave of innovation is, and it is taking us towards an era of autonomous engineering. We are now able to now move to much higher-level and coarser-level objectives where models can parse those objectives and then orchestrate a collection of agents to execute it.
With the most recent advances in models in terms of reasoning, we are at the cusp of really enabling autonomous engineering with what we call as long-horizon agents. These are agents that are able to just take a desired outcome with the necessary guardrails and the necessary constraints specified by an engineer, and then able to decompose, plan, and then orchestrate a very sophisticated set of tool invocations, task-level agent invocations, using all the proprietary knowledge assets and knowledge graphs that we have built around our tools, proprietary APIs and data we have built into our tools, and most importantly, anchored by the ground truth engines and solvers, which are really fundamental to everything that we are doing.
While the AI models will reason and explore, the generation of the circuits, the generation and validation of the designs and silicon and systems are basically done by the portfolio that we have across the silicon-to-system spectrum. Really, the future is long-horizon agent workflows orchestrating across multiple tools and delivering outcomes, thereby ushering the era of autonomous engineering and helping us close that gap we talked about earlier. Let me take a moment and walk you through the Synopsys agentic AI portfolio. Because of the strength of our portfolio, as Sassine talked about earlier, all the way from silicon architecture all the way to manufacturing, TCAD, OPC, and so on. Then our systems analysis portfolio, ranging from structures and fluids all the way to electromagnetics and optics. Synopsys essentially is in a unique position to deliver the broadest and deepest agentic portfolio across engineering software.
What we are doing in every domain is delivering, per domain, a set of long-horizon agents that are able to take very coarse tasks and execute them through a complex orchestration of tools and task-level agents that are essentially able to do specific things. For example, let's take the verification domain. A verification agent engineer is able to take an outcome or an objective, like here is a spec 200, 300, 500 pages long, and give me a verified RTL and test benches compared to the spec. Or here is a design that I'm trying to improve the coverage on, and I have about 30%, 40% coverage. Take this and make this into a 90% coverage situation.
An implementation agent engineer is now able to specify an outcome, like here is a collection of blocks that I want you to run an autonomous place and route closure, run sign-off, do the ECOs after sign-off, and essentially close this block for me. That's an example of a long-horizon agent in implementation. Moving to analog design, the nature of the task is no longer let me click these 10 buttons to get this transistor moved from point A to point B, or to connect this device to this other device. The nature of the task is, here is a spec of a circuit that I want to build. Go through the design of the schematic, the layout, and the simulation of that resulting design, and meet these objectives.
That's what we mean by long-horizon agents, and these agents are essentially invoking a strong library of task-level agents provided by Synopsys per domain, and of course, everything is anchored by the ground truth physics, solvers, and engines that we have built across decades, across the entire spectrum of silicon to systems. The story doesn't end just in silicon. We've also now extended the same philosophy over to simulation and analysis. A CFD engineer can essentially provide a geometry, describe the kind of objectives that they're looking for, and the entire setup and meshing of the CFD, the execution of Fluent, and then the results analysis of the Fluent CFD simulation, and then the next steps in order to further improve that, all gets handled by a long-horizon agent.
This is really something that we believe is going to bring a significant level of autonomy into engineering workflows, and thereby increase the capacity. All these capabilities, the long-horizon agents, the task agents that the long-horizon agents can invoke, and then all the invocation of the tools are all driven by the Synopsys Agentic AI Platform, which we call Autopilot. It is the broadest and deepest portfolio, as I mentioned, across the entire engineering spectrum from silicon to systems. A couple of things that are unique about this platform. First and foremost, interoperability and openness. In this platform, our customers can onboard their agents to benefit from all the assets on the platform, like our knowledge graphs, proprietary APIs, proprietary data to write powerful agents on our platform. At the same time, we want to meet our customers where they're at.
If they are down their agentic journey and they want to integrate our agents into their agentic workflow, we also connect to our customers' agentic environments through protocols like MCP and A2A to essentially enable that as well. Last but not least, the compute and LLM optionality. Essentially today, we are supporting the entire spectrum of proprietary foundation models all the way through open-source foundation models. The performance of the agentic capability, regardless of domain, is heavily dependent on the quality of the foundation model and the ability of reasoning that it's able to do, the ability of orchestration it's able to do. As you will hear later, there's a tremendous opportunity here as well in terms of how to really work with foundation models that understand semiconductors or physics much better, and thereby get even better results.
A picture conveys a thousand words, but I believe a demo conveys 100,000 words. Let me give you an example of how our 3D IC agent engineer is able to go from a very core spec of what an AI advanced package looks like, almost a napkin diagram, and take that and work through the entire process of advanced package design and simulation. This is an area which transcends multiple domains. The design, which is now anchored in Synopsys 3DIC Compiler, all the multiphysics analysis steps that are anchored in the Ansys technologies of HFSS and RedHawk, and many others. What you will see here is how the long-horizon agent is able to decompose the objective and execute task agents, execute the ground truth tools at the base of it, and really run an end-to-end flow to design an advanced 3D IC.
If you could, please-
[Presentation]
[Presentation]
I hope you can see why we are so excited about how these long-horizon agent engineers can really change the way in which engineering is being done and really increase capacity in very stretched engineering teams. Let me now give you an example from the simulation and analysis area. Here, we are going to look at electromagnetic coupling and electromagnetic interference, which is very important for any electronics design. What you're going to see here, essentially, is how an engineer essentially is using the Synopsys Agentic Autopilot platform to really execute an EM analysis of a design, find specific signals where they think there might be problems, do some simulation sweeps to determine whether or not the simulation results are meeting some industry standard requirements, which is encoded in something called a CISPR 22.
Let's take a look at Ansys HFSS and Ansys Electronics Desktop embedded within the Synopsys Autopilot platform and enabling an agent engineer flow. Here, what you can see is, look at the objective that has been provided here. Again, it's all natural language interface. There's no clicking of buttons and no pulling down menus. Essentially, the directive was given to run the EMI scanner on this PCB, and then the necessary tools are getting set up and invoked. The results from those tools are being summarized, analyzed, and now the engineer says, Zoom into the clock net, because I think that's where the problem is. Again, the right tool level invocations are happening underneath.
Then the next query is, Hey, do me a sweep across a range of frequencies and show me what the fields look like. Again, this is what I mean by specifying outcomes and objectives rather than having to know the intricacies of the tool and how to drive the tool to get specific results. In this case, you see the radiated emissions. Then the next request here is plot the psi wave results and then compare it to CISPR 22, which is a standard for EM interference. Again, here, it's basically opening up the CISPR 22 PDF, understanding what the requirements are, firing a plot, taking these results, and essentially generating the comparison to the CISPR 22 standards, which is basically illustrated by. In fact, it doesn't find the file.
It goes and locates it correctly. Then it basically plots the final graph, comparing the field values from the EM simulation, against what the guidelines are with respect to the standards. So this is really illustrating how the nature of design is changing with agentic AI and how, in this case, the red line is the CISPR 22 requirements, and you can see the plot is basically satisfying those requirements. So natural language interfaces, all the tool calling is being handled by the agents. The proprietary knowledge and the skills that Synopsys has built up over decades is all encoded into the platform. This is how we are really revolutionizing the way in which agentic AI interconnects with engineering. One key point to touch upon is really what is happening underneath in terms of the tool calling.
Let's take an example of a verification team in a hardware design group. They've been handed a spec, and they have to write the test plan and the tests corresponding to that spec in parallel to the design team implementing the design. In the typical setup, there's a lead who takes that spec, breaks it up across their team of verification engineers, assigns each of them a portion of the spec. Then they then go off essentially running our verification tools like VCS or our debug tools like Verdi, and essentially building those tests. But all that work that is being done is essentially gated by things like, Hey, I've got meetings all day today or I don't work necessarily all weekends.
I don't work late into the evening. So to a certain extent, the total work that can be accomplished is gated by the amount of human cycles that are available. Contrast that now with an agent engineer, a long-horizon agent engineer that is now handed that same spec, and essentially with the right knowledge graphs from Synopsys embedded within the Synopsys Autopilot platform and all the task agents available, there's a rapid decomposition of that spec. Essentially, agents are fired off to work on different parts of the spec. Beyond just writing tests, these agents can also explore a much larger solution space to write high-quality tests.
As a result, when you look at the tool profiles and the tool calling profiles, essentially, we have an expectation of a much higher tool calling profile than in the case where we don't have agent engineers, because you're only gated by the compute on the right-hand side. The more compute you have, the more exploration, the more agents that can run in parallel, and as a result, finish this task with high quality and much, much faster and essentially expand the capacity of this team. That's really why we believe that the move to agentic AI and agentic execution is going to significantly increase the tool usage. Let me finally conclude with the momentum that we have with our portfolio across the industry.
Over 50 engagements with all the top customers, and many of them are now seeing the value of the agent engineers and the agentic platform that we have delivered. For example, Intel is seeing a lot of value in the work we are doing with them on reducing their verification bottlenecks and improving the engineering efficiency of their design and verification teams. MediaTek, we are working with them very closely on something which is a very, very laborious design step, analog and mixed-signal design. Here, MediaTek is engaging with Synopsys to essentially take our agentic analog design capabilities and essentially use agent engineers to achieve significant productivity benefit in the design verification as well as optimization of analog circuits. Samsung Memory is also working with us very closely to use our agentic portfolio to accelerate their engineering processes and design steps for high-bandwidth memory design and DRAM design.
Again, very close collaboration there. NVIDIA is both a partner as well as a customer. Of course, as a partner, we work together very closely on the agentic platform and many of the components in our platform, we co-develop it with NVIDIA. But then they're also a consumer of all the agent engineers and the agents that we are delivering through the Autopilot platform, and they're also seeing tremendous benefit and value from all the innovations that we are driving. With that, let me hand it back to Sassine to now talk about the AI monetization. Thank you.
Thank you, Shankar. All right. What you heard from Shankar and the snippet from Richard at OpenAI, I hope I don't have to convince you that AI is absolutely a TAM expansion for EDA. You need the essentialness of EDA to generate, to validate, and therefore, we expect our tool usage will only expand with the combination of a human and agent engineers. I have been, over the last five, six months, describing that our customers will not apply AI in the same way across multiple customers. Customers try to differentiate in the way they're going to implement and evolve their workflow. Therefore, Synopsys' strategy is to provide a solution that adapts to the various customer flavors of adopting AI. Let me walk through the various choices that we have and can offer customers.
The Synopsys full stack, that is what Shankar just described, where the customer can come to Synopsys, they get the platform, the agents, the tools, and they provide objectives to AI and to achieve a certain outcome. This is a significant investment we have been making. We will continue on leading and put our resources, energy, effort in the Synopsys full stack. The second choice customers are making, and there are a lot of customers, by the way, in that second category, where they are saying, I have my own special sauce. I want to build my own agent. I want to build my own platform. What I need from you, Synopsys, is access to your tools and access to some of your agents. I do not want to reinvent a debug agent.
I will get the debug agent from you, Synopsys, but I want to own the rest of my platform for various very good reasons. The third model or choice is a frontier model. As you have seen, and we have discussed it many times, when you have a model that comes out one day and say, I was able to perform this specific task in semiconductor chip design, in many cases, those models were using open source because that is what they have available, and they saw good either productivity or proof of concept. With the advancement in frontier intelligence and reasoning, I have no doubt that there will be a convergence with frontier models, with EDA in order to achieve the best outcome as another customer choice.
Across all choices, EDA and SNA engines are essential as the physics-based ground truth. You will not tape out a chip if you are not going through the sign-off gates in order to ensure that whatever has been explored, proposed by AI is being validated. Now let me walk you through quickly, how are we thinking in terms of monetization. We talked about subscription and consumption in the last couple earnings calls. In the Synopsys Full Stack, the customer can subscribe through a subscription license to the Synopsys platform and to the Synopsys agents, as well as the Synopsys tools. Now, Shankar showed 5- 10x more consumption sometimes the agent can trigger. In number of cases, the customers, they may choose to have a subscription for the tool and a consumption-based. The consumption can be on-prem, cloud. We have it available, whatever customer choice they choose.
So that is the monetization stack using the Synopsys stack. The second revenue stream is when the customers are saying, I may want to have a hybrid. I want to license some of your agents, and I need your tools to build my agents or your tools to run your agents, because it is going to consume more licenses. We will have the agent subscription that sits inside the customer platform and the tool subscription and consumption. The third, when you are training a model, you need tools to train it. So that is a tool subscription. When you are influencing the model, you need tools to run it. So that is a subscription and consumption, and then there is a revenue share. A lot of the questions will come up, how and based on what will you have revenue share? It is going to be based on outcome.
If you're able to achieve a certain PPA with the best optimized RTL that you created, but the model, given it can explore much broader, can provide a better PPA. Will that better PPA be worth X dollar, and therefore, what's the revenue share model in this use case? Here, we've been working for, actually right now, many months, close to two quarters, assessing who, how, what's the model to protect Synopsys IP if the model is trained on Synopsys. Because when a model is trained on open source is one way, when it's trained on Synopsys, what is the role of Synopsys in training that model, in influencing the model? Who owns go-to market? Who owns the monetization process?
As we've gone through many of these iterations and talking and collaborating with many of the AI labs, I'm very pleased to announce today a partnership, multi-year, with Synopsys and OpenAI. The cool thing about it, there will be a specialized GPT- Synopsys model where it's post-trained using Synopsys agents, tools, skills, and workflows. So we bring in the knowledge with our tools, with our workflow and skills. OpenAI brings in their frontier intelligence and reasoning, and the combination of both should provide the best outcome, period, in terms of as measured by an objective. As Shankar showed in the demo, you can provide it a PPA objective, certain targets, et cetera.
This multi-year arrangement is available now to set of customers that are in early engagements to see what is the outcome of a model converged with the best-in-class EDA versus an open source or other alternatives EDA. Before I go into more details, actually, let's hear from Greg at OpenAI.
Hi, everyone. I'm Greg Brockman from OpenAI. We're really excited to be partnering with Synopsys to accelerate chip design for everyone. At OpenAI, our progress in AI depends fundamentally on the chips that we run on, and now we have an opportunity to use AI to help design those chips. Synopsys brings deep engineering expertise and the tools that chip designers already rely on. Together, we're building a specialized AI model that is built specifically to master these tools. The goal is to bring together frontier intelligence with the ability to use Synopsys EDA tools, and this will help engineers autonomously explore a much greater range of design choices to efficiently trade off design targets and to get to a working chip faster, to be able to shave off weeks, months from the design process, and to bring more chips to the world.
And what I find really exciting is that this is something that can build on itself. Better AI helps engineers design better chips. Better chips makes AI more capable, more efficient, more accessible to everyone, to empower people around the world. That is something that our work together will help accelerate. So thank you, Sassine. Thank you to the whole Synopsys team. We're excited to build this with you and to see what your customers will achieve.
All right. This is actually very exciting options for customers as they're assessing how much do they invest in their own intelligence models, workflows, the Synopsys full stack, and the OpenAI Synopsys option. I know most likely you have a lot of questions on your mind. I did not give you much runway on this one like we did this morning with IP. So we'll take the questions later. But just to give you some color, the key things to emphasize, this is a specialized model. It's a GPT- Synopsys, meaning whatever learning that goes into the model, it stays Synopsys proprietary inside that model. So our IP does not leak or disappear into the bigger base model. The objective is to optimize, verify, deliver best outcome. So you give it objectives, you deliver outcome. And there are engagements underway. Let me add a little bit more color.
How will customers access this? Think of it as a service. A customer will go to Synopsys, go to market, or Synopsys go to market, goes to the customer, and they can order an OpenAI directly with a customer. The customer will go to OpenAI. They get access to the model, GPT- Synopsys. They get access to the compute. They get access to the EDA tools, agents, and through that engagement, it will be an outcome-based. And that's where we monetize, back on the prior slide I showed you, the subscription of training the model, the subscription/consumption, and the revenue share component. I cannot be more thrilled for EDA that we'll be able to capture value based on the high impact we deliver to our customers as yet another option in this era where AI is making significant progress in intelligence and reasoning.
If I were to summarize AI for EDA, is it a tailwind or a headwind? The key concepts that we all need to be very clear on what AI can do, AI can reason and explore. You need EDA when that intelligent model is providing a task or guiding. You need EDA to generate. You need EDA to generate a GDS, a layout, a clock, et cetera. And you need EDA to validate. So there is AI can explore, can recommend, but you need Synopsys to generate and validate. Agentic AI will reason, orchestrate, execute. Synopsys ground truth engines are a must because they are foundry-certified sign-off back to we're the bridge for DTCO, we're the bridge to foundry. They're deterministic, they're physics-grounded, and we've gone through the rigorous validation in manufacturing and yield assurances. That's what our industry and Synopsys has led and done over decades.
When you think of an intelligent model, frontier model, that's a huge value. It's fantastic value to do the exploration, et cetera, with EDA to bring the high impact. Of course, the outcome will be a verified outcome. I'm not sure if you've noticed the sentence in quote. If I were to summarize what customers tell us all the time, and you see it consistently, can be summarized here. In AI we believe, but in physics we trust. Customers are not saying, I'm not sure if AI, I'm going to adopt it or not. They are adopting AI. At the same time, they need physics to validate. They need sign-off to validate. From an EDA growth, we talked about the synergy of electronics and physics, more design start, and AI-driven chip design. We are raising our long-term guide from double digit to mid-teens growth.
We talked about it will be 2027 will be the year where we will see the revenue synergy from the Ansys acquisition. We will see revenue from AI, and we will see continued acceleration in EDA as well as HAV, the hardware-assisted verification. Let me next go to SNA, but let me introduce the SNA section actually with a short video. It's always cooler to have SNA videos versus chip videos. Chip videos, they seem so boring. SNA, you can look at the car, the airplane, the data center between electronics, et cetera. That's a great video, actually, of co-design. Co-design of electronics with the whole rack, the structure of the data center, the cooling, et cetera. What are the complexity of intelligent system design? The failures that are caught late are so costly, and it takes so much time.
The trade-off companies need to make, if I do a lot of modeling and simulation upfront, I need certain skills. I need to change my workflow. It's hard to do it. Companies, customers, they don't innovate unless the constraints are high. When the constraints are high, can be driven that those systems are becoming more intelligent. Those systems need to serve various different applications that is necessary to bring in a different workflow for these systems. Today with our SNA portfolio, we pretty much serve every one of these markets. The Ansys acquisition expanded the Synopsys customer base by 10x. 10x expansion of our customer base. The beauty, many of those customers are looking for the next method of designing these intelligent systems. If they're building their chips and expanding to build their chips, that's fantastic. That's even a bigger opportunity for Synopsys.
If they're sourcing the chip, but they're looking for a new way to improve from a traditional development model, which historically, until now, back to that 90% of the $1.7 trillion, is done through this traditional model, where you have requirement, you design, you may or may not simulate, you build, you test. Then if you find failure, you go back through the loop of testing, then redesigning and building. The simulation usage is fairly limited by few inside these companies that called analysts. That after the design is done, it goes through a different group to do some simulation as the product is being built and tested. The opportunity is how do we make simulation more accessible, where the simulation is done during the design and more and more simulation is, again, accessible before the product goes into build and testing.
Now, that concept is not new to many companies that they are building the intelligence systems, and they need and have to deliver these products on time. This is a set of customers, out of many, that have adopted early the simulation in the new way of building a product. You can see what they see is a closer correlation to an actual physical testing while reducing the need for as much physical testing. Now, what Synopsys has with our SNA portfolio is the multi-physics trust and sign-off. That is very important because you can simulate all day long. If you do not have fidelity in that simulation, you will not use it. So Ansys and our SNA portfolio across five physics, fluid dynamics, structure, electromagnetics, optical, and thermal, we have the leadership position. It is the ground truth of sign-off for the physics domains.
AI is not only impacting EDA. It is the same thing for SNA. The democratization of simulation has been a long, multi-year effort to bring simulation from few analysts to a broader set of users of simulation. The next set or opportunity is AI-driven surrogate models. We have technology called SimAI and optiSLang. What SimAI does is you can upload a prior simulation data from the prior design and can predict and gives you, through optiSLang, options of design. If you were to change the curvature of a blade in a jet engine or in a car or in a robot, how does it change the performance of the product you are building? So you can get it done through surrogate models in fraction of the time. Then you do the final simulation for sign-off.
It is an opportunity to do a different type of simulation faster through models, which is the surrogate model opportunity. Shankar talked about the agent engineer for physics. This is an investment. Immediately when we closed the acquisition, we brought the Ansys R&D to operate into the same platform, rhythm, strategy around AI and the agent engineering. As Shankar mentioned, we have the first wave of customer engagements and announcements there with agents. With frontier models, the same as I just announced with OpenAI for semiconductor, I see it as a potential for systems. Not only for silicon, for systems as well. The opportunity is very similar. Can you leverage frontier models with the ground truth physics to explore, simulate, test physical AI products? Shankar already went through this.
The platform, the Synopsys Autopilot platform with domain-specific agents expand into physics, and that provides Synopsys the opportunity to have an AI-powered simulation across multiple domains and multiple industries. That is the simulation aspect of it. You recall about a year ago, we announced the partnership with NVIDIA, where we will leverage the Omniverse from NVIDIA and Synopsys simulation and analysis across various industries, where Omniverse provide the customer the ability to envision, visualize their product, design their product, and Synopsys' engines are used to simulate, because, again, they have the ground truth sign-off of physics. We have many customer engagements there. Actually it was very cool earlier this week to see AMD buying World Labs. It is the same approach. Why? Those intelligent systems, you need a digital twin of electronics, you need digital twins of the physics.
You need a digital twin of the environment in which that system is operating. If it's a data center, it's a data center, it's a car, it's the world environment, et cetera. The direction we took with NVIDIA Omniverse is a validation earlier this week that that's a market that is expanding, and there will be similar type of collaboration, that we expand as, of course, the acquisition will close, et cetera. To summarize, for SNA, the long-term double-digit growth, that's organic, and that's higher than the traditional growth that Ansys as a standalone organic has been able to achieve, and is driven by number of tailwinds. More simulation are needed, more value capture through accelerated simulation. This is a GPU acceleration and other. The AI-driven inflection point we just talked about, and the new opportunities through digital twin and other.
Now to bring things to close and summarize, the long-term growth outlook is mid-teens CAGR through 2030. If you recall, that used to be double digits, and the reason it's mid-teens is the growth we've talked about for EDA, mid-teens IP, high teens, and SNA, the double digits. FY 2027 outlook at 15% year-over-year growth. Some of you asked me earlier, "How much did you prep for this meeting?" I'm like, "Actually, the meeting itself, it was less the prep." The prep was the FY 2027. We pulled by two and a half months to provide you the FY 2027 and not miss the opportunity to be here and share with you our enthusiasm, our excitement, what is driving it, and of course, some of the major collaboration and redefining some of the businesses that we have. Just to summarize, this is our opportunity.
At the silicon level, the expansion of application-optimized silicon gave us the opportunity to create a new category of IP, which is AOIP anchored with a license fee and a royalty, which is a significant opportunity for Synopsys. AI for EDA, along with everything that we do in terms of multi-physics fusion, the entire platform, HAV, et cetera, the OpenAI option for customers, as well as the customer choice of building and mixing their agents or a full stack from Synopsys, that will all contribute to our growth in 2027 and will only expand beyond 2027. Systems, more SNAs needed for these intelligent systems. Truly what differentiates our assets and company is the ground truth, highly trusted physics that we do in both silicon and systems. With that, big, big thank you. Now we'll take a break and I look forward for the Q&As. Thank you.
We will now take a short break. Please return to your seats in 15 minutes.
[Break]
Please welcome Shelagh Glaser, Chief Financial Officer.
Thank you. It is great to see everybody. Thank you so much for coming. I know there is a lot going on, so thanks for prioritizing time with us. I am going to bring together everything you have heard today and show how it comes together in the financial model for the company. First, Sassine talked about this being a major inflection point in the industry, and why is this moment different? The complexity and the pace that is happening in system and silicon design has never been faster. There is a talent gap and a compute gap. So AI is going to help bridge the gap on that. Customers are increasingly designing integrated physical and digital systems, which require co-design. Shankar went into a lot of detail about that. First time right economics have never been more important to customers.
Having to do another spin or another tape-out is hundreds of millions of dollars of a new tape-out, but even more importantly, it is missing a market window. So that is missing revenue. When customers have to over-design, that leads to bloated die sizes, which leads to yield problems, which leads to less units to be able to sell. So customers need to come to us because we are uniquely positioned with a leadership portfolio in EDA, in SNA, and IP, so that they can be confident that they have the ground truth physics in sign-off. So the models they build of the products are going to be the same as what they are going to see in high-volume manufacture. We are the link that allows them to have that confidence. We have traditionally been seen as we are solving R&D problems.
As Shankar and Sassine laid out, we are driving even more value for customers. The problems we are solving are making sure that they are able to hit their revenue targets, their chips are on time, they are hitting the right PPA, margin targets, they are achieving the yield and the cost that they want. We are making sure they have time to market and high-confidence product schedules and high-confidence product success as they move from doing designs and models into actually taking those products into high-volume manufacture. So this is the early innings of increased value capture that Sassine laid out and the change in our business models that we are driving, and we are in the early innings, and this will build over time. Let me take a step back and talk about where we have been.
Over the last five years, we have doubled the size of revenue of the company.
Over that same period of time, we have driven significant margin expansion, seven-plus points in margin expansion, and we did that while integrating one of the largest acquisitions in the software industry and the largest acquisition our company has ever done. So we have proven that we can grow both scale and we can grow profitability. In 2026, this has been a year of execution for us. It is our first full year, as Sassine said, bringing Ansys in. Over the course of the year, from our initial guide, we have raised outlook on all key metrics. We have raised outlook on revenue, non-GAAP operating margin, non-GAAP EPS, and free cash flow. Importantly, this momentum has been broad-based across all of our businesses, and we have been translating more profitability into more cash versus disciplined execution.
This helps set the stage for the next era of growth for us as a company.
Before I go into the long-term model, I want to make it clear how we will present the company starting in fiscal 2027. We will continue to have the same segments, Design Automation and Design IP. What we will change is the revenue disaggregation. Specifically, what we committed to this year was to provide full transparency on Ansys, our first year of this consequential acquisition, and we have delivered on that. As we move forward and we build out these multi-fusion physics product lines, it will be harder to separate Ansys and EDA products. So what we will be presenting in fiscal year 2027, we will move the semiconductor business unit from Ansys into EDA. That is about 10% through Q3 2026 of the revenue of Ansys. We will also use that remainder of Ansys and show that to you in Simulation & Analysis. Design IP will remain unchanged.
These changes align to how the industry views these, and it will allow us to give you full transparency of the performance on our EDA business and our Simulation & Analysis business. Throughout the course of the year, we will provide apples-to-apples comparison because obviously we will have the comparison with 2026, how we report it. Now, let me get into the growth algorithm. Sassine laid this out in his section. For this year, for 2026, we anticipate revenue of $9.7 billion. As we are driving to fiscal year 2030, we are driving to a model of mid-teens overall growth. Underpinning that is mid-teens growth for EDA with a floor of 13%, double-digit S&A growth with a floor of 10%, and Design IP with a floor of 17%.
This is all underpinned by what Sassine talked about in terms of increasing design starts, increasing complexity, increasing need to do co-optimization, and in the Design IP, Sassine laid out the new business model that we are driving with application-optimized IP. Underpinning this is revenue synergies, which I will talk about in a minute, and increased value capture as we evolve our monetization model and change the way that we work with customers on that. Let me go into synergies. When we announced the Ansys acquisition in January of 2024, we committed to both revenue and cost synergies, and I want to provide an update on both of those. Let us start with revenue synergies. On revenue synergies, the commitment is $400 million run rate in synergies by fiscal year 2029.
What we have already done, which we talked about today, Sassine laid out again today, is we have already built a new joint roadmap, the multi-physics fusion products. We are seeing great customer enthusiasm on that, and those will start to revenue in 2027. We have also brought the sales teams together so the sales teams can have cross-selling, really across the entire product line. As we exit fiscal year 2027, we will have greater than $100 million run rate in revenue synergies. So the synergies begin in earnest in 2027, and we have confidence in our ability to achieve the $400 million run rate synergies by 2029. Now let us talk about cost synergies. On cost synergies, we had committed to $400 million run rate by fiscal year 2028.
As we've talked in each of the earnings calls throughout this year, in fiscal year 2026, we've been accelerating those synergies. I'm pleased to announce today that we will be complete with our $400 million run rate synergies in fiscal year 2027, which will be one year early. We are executing against our synergies and feeling very confident in our ability to achieve these. We'll be done with cost in 2027, and we've got strong line of sight to 2029, given the strength of the first year of revenue synergies we'll have in 2027. Let's talk about operating margin. Sassine laid out the change we're driving in the business model. That change flows into operating margin. I'm pleased to say that our objective in operating margin is approximately 50% by fiscal year 2030.
That's up from our prior expectation of mid-40s, and it's driven by the changes in the business model that Sassine outlined. How we're doing that is on multiple levels. We're scaling and bringing efficiency into everything in the business. In the scaling, we're working on higher value capture, as Sassine outlined today, as we evolve the business model and IP, and we infuse AI into our products. Portfolio optimization, that's something you've seen us do year after year, making sure that we've got our investments in the highest return areas, and that's a constant evaluation that we do. Operating leverage in everything we do. Literally, how do we simplify every process and every approach in the company so we remove friction, so we focus on high value add?
While we're doing that, it isn't about cost-cutting, it's about efficiency and leverage and investing in critical innovation, which fuels the strategy that Sassine laid out. Specifically, large areas of investment we're making is advanced node and multi-die design to be able to support our customers as they endeavor on these more and more complex designs. Making sure that we're building innovation and simulation and digital engineering to be able to make sure that we're supporting those customers, and infusing AI for engineering in all that we do. Just as we're working with our customers to infuse AI, we're infusing it so that our team gets the benefit of that. Funding the application-optimized IP, formerly called Factory Two, in IP. Building that out. We're keeping both factories.
Sassine talked about the Factory One, which is our traditional IP, so we're keeping that, and we're adding on top of that investment to build out this new model. Let's talk about the capital allocation framework. Our priorities are clear. Our priority first and foremost is to invest in growth in the business and invest in R&D that fuels the innovation that drives the business. On a regular basis, to make sure that we've got the investment in the right areas to drive the growth of the company. We will drive selective and disciplined M&A. The second is to maintain a strong balance sheet and maintain investment-grade rating. You've seen us pay off the term loans early. That was an important part of our operating plan as we drove through this past year to get the term loans paid down.
The third is to regularly return capital to our shareholders. Here, we will return up to 50% of free cash flow to shareholders. This framework allows us to invest for growth, to build the innovation that fuels the strategy while creating shareholder value. Another thing that we've talked about, many of us have talked about, is how we think about stock-based compensation. We are targeting 8% SBC as a percent of revenue by fiscal year 2030. We will do this through revenue growth and scale and disciplined application of our equity program. You've already seen us make significant progress on this. We're down 3% from our peak in 2025 of approximately 13%. Like our investment dollars, we think in a very disciplined way about where we put our equity.
This allows us to ensure that we're investing in attracting and retaining top talent, which is absolutely necessary to ensure that we're building the innovation out and building the capabilities that will allow us to achieve our strategy. All of this will compound in both EPS and free cash flow. With revenue growing in the mid-teens, operating margin scaling to approximately 15%, we will grow EPS and free cash in the mid-20s. This is a durable compounding growth plan that translates into profits and shows up in cash. Putting it all together, these are our long-term financial objectives. It's underpinned by three scaled market-leading businesses and the change that we're driving in the business model with AI and the application-optimized IP and underpinned by our synergies realization. We're not dependent on any one single variable.
We have multiple growth vectors that we're driving, and what we're driving is both growth and leverage, which shows up in EPS and free cash flow growing even faster. We're doing all of this while we're investing to ensure that we can continue to fuel the growth and returning capital to shareholders. Let's zoom into 2027, and as Sassine mentioned, we're giving our full 2027 guide today, which we would normally give in December. Since we're together, we felt it was very important that we talk about what does 2027 look like. Let's shift into that. 2027 is a very important validation year of this long-term model that we're laying out today. We expect to grow revenue 15% to $11.15 billion, to expand operating margin 250 basis points to 44%, and to grow EPS even faster at 27% to $19.08.
All of these numbers are obviously at the midpoint. This represents the strength of our underlying business, and it's a great start towards the long-term model that we've laid out today. I want to talk a little bit on operating margin, because this has been a question for quite some time, so I want to make sure we're really clear on this. What we've talked about in 2026 is our expectation for full year 2026 is 41.5%. That's 420 basis points improvement from fiscal year 2025. On top of that, what I just guided for 2027 is another 250 basis points improvement.
And again, we'll have the full realization of synergies next year, and we're driving the greater scale in the business while we're continuing to invest in the business and continuing to invest in important strategic objectives to achieve the product lines that will help our customers scale. Putting it all together, our guidance for 2027 is extremely strong. We're very confident in this. This is why we feel comfortable giving this to you in September instead of waiting for December. You've seen us accelerate into the second half of 2026. You've heard the new deals announced today, so we have strong progress towards these long-term goals, and 2027 is an important year to be able to demonstrate that. On top of the metrics I already talked about, free cash flow will be approximately $3.1 billion, and that's up about half a billion dollars year-on-year.
Just one more thing before we get into Q&A. Given the strength of the balance sheet and our business model, we are announcing our intent to repurchase $1 billion in Synopsys shares over the coming months. This is based on our confidence in the model and our cash generation, and it aligns exactly with the capital priorities I just outlined. Number one, to invest in our business. Number two, to ensure we have a strong balance sheet. And number three, to make sure we're returning capital to shareholders. This allows us to offset dilution, and over time, the repurchase program will allow us to reduce share count. So let me sum up. We're at an industry inflection point. Complexity is outpacing engineering capacity. First time right is absolutely high stakes for our customers. Customers cannot afford to miss critical market windows.
We have the portfolio that allows them to have high confidence in being able to ensure that they're building products that will achieve their goals and will be right the first time. Our financial goals are mid-teens revenue growth with margin expansion and free cash flow expansion. We're confident in our execution, and we're confident in our value creation, and the first step towards this long-term model is our fiscal year 2027 guide. With that, thanks, and we're going to move into Q&A.
Please welcome back Sassine Ghazi, President and CEO, and Shelagh Glaser, CFO.
Ooh, questions. We are going to have two mic runners, Chris and Christine, and they will come to you guys, and then you can ask your question. Please state your name and the firm. I ask you to limit yourself to one question, and then I can come back to you guys if you have more questions. We will begin with Siti.
Siti Panigrahi from Mizuho. First of all, congratulations. It is an amazing Investors' Day and thanks to Sassine and his team, who has put together a really good Investors' Day, and thanks for inviting us. Sassine, as you said, guide, not only 2027, even your long-term guidance, it is amazing. Much better than we were expecting. The questions come, when you laid out a lot of growth opportunity, how do you rank order, and what gives you that confidence to hit that number? Specifically on the AI opportunity, I would like to ask the different scenarios you talked about. Who owns that value capture part? Which scenario you have higher value capture versus other models?
Yeah, thank you for the question. We will not talk about 15% unless we have confidence we are going to meet the 15%, and it is really the layers that we described. Starting with IP. Even though the royalty will not show up in 2027, but starting in 2028 and then it starts ramping into 2029 and beyond, that becomes a fairly large percentage of our IP business that comes through royalty. While Factory One will continue on executing and delivering to the double digits expectation as well. All in all, in IP, as Shelagh mentioned, the floor is the 17% with an objective to grow in the high teens. On EDA, it is the same. I want to remind us that the floor is the 13% with the objective for mid-teens. The EDA between synergy of the multiphysics AI definitely will start contributing in FY 2027.
I want to call it the classic EDA growth of delivering to just the best-in-class software and hardware in order to achieve these opportunities. In terms of AI for EDA, we talked about three revenue streams. Revenue stream one, the investment is essential because those agents, if the customer is subscribing to them, either in option one or two, that investment and delivering to these agents is essential to have that differentiated capability. It is clear for our customers how to pay Synopsys in stream one and stream two because they know they need to subscribe for the platform, the agent, and the license, or some consumption of the license. We already have number of customer engagements. That is what gives us the confidence in 2027 we will see revenue from stream one and two.
Stream three, the revenue share, we are currently in a number of customer engagements testing the GPT-Synopsys based on outcome. You give it an objective. If the outcome is better than what they are able to achieve, the monetization will happen based on what I call the service. That model runs on OpenAI cloud, so they will provide the model to compute. The customer gets a model, compute, EDA licenses, EDA agents, and that harnesses the context, the skills that are required. Then there is a revenue share split between us and OpenAI based on these outcomes. That is a whole new stream of revenue that we were not able to capture before. That is a new revenue stream that we are looking at. That is why the mid-teens for EDA with the floor of 13% starting in 2027 is something that we are confident about delivering.
We will go with Jim at Goldman.
Jim Schneider at Goldman Sachs. Thanks for doing the presentation, and congratulations on the targets. I was wondering if you could maybe talk a little bit about, given the accelerating growth rates you are expecting across the businesses, to what extent is pricing and pricing to value sort of driving or underpinning those targets? Can you maybe talk about any like-for-like pricing conversations you are having with customers to sort of drive that accelerating growth? Or is that purely on the basis of either mix or top-line benefits from elsewhere? Maybe just as a secondary point, can you maybe talk about, to the extent the agentic solutions gain traction in the market, what is the impact on the company's gross margins?
I will take maybe the first one, and Shelagh, you can address the second one. Pricing conversation with customers, they do not go far unless you are able to deliver more value to the customer. In EDA, typically, there is a renewal cycle. When the renewal cycle comes in, customers assess their needs, and that is the opportunity to inject new technology. With almost every customer right now, the conversation is, I need more licenses because I am building my own agents, or, I need to buy an agent from you, or, I need the new 3DIC Compiler fused with multi-physics because I need to achieve my next program that I need to plan for. That is really where the opportunity comes in to lift the value that we are getting from the customer based on the value that we deliver. IP, I want to say, is a different story.
With IP, if we are talking about a Factory Two, really the reason we were able to successfully bring in number of customers in Factory Two is the scale, the trust, the quality that we have in our IP. It does not mean it is more essential than EDA for chip design. They are equally essential. But the dependency for them to build a customized chip depends on Synopsys. When I go to the extreme and say there are no other options to deliver for a customized IP at the scale that we can deliver, it opened up the conversation with the customer. We need to capture more value given the impact we are delivering to you, and this is where the royalty is coming in.
The point I made as well, royalty will be higher than the license that we capture, and that is the other opportunity to deliver value.
Maybe, Shelagh, if you want to take.
Yeah. Let me take margin. I think about AI in sort of two buckets. Obviously, we are working with our customers, everything that Shankar and Sassine talked about, and our cost doesn't really change. A lot of that is accretive to margin. Also think about our own internal consumption of AI, and the way that we are looking at it is also outcome-based. How do we basically create more capacity for ourselves? We have areas where we are short on engineers. How do we create engineering capacity that allows us to get products out that otherwise we wouldn't? So that is also beneficial in margin because those are products I wouldn't have even had.
I don't want to be accused of prioritizing the front row, so we go to Vivek in the second row.
Thank you. Vivek Arya from Bank of America Securities. Thank you so much for an informative Analyst Day. I had two questions. One, Sassine, for you on the revenue side, and then Shelagh, for you on the operating margin side. On the revenue side, Sassine, if you go back to the Analyst Day you had in fiscal 2024, at that time, I think you had set expectations of somewhere in the low teens growth. You are definitely raising that bar towards mid-teens. But the growth rate in the last few years, as an industry, was lower than that. I am just saying as an industry, what changes in the next few years to help you accelerate and have more confidence in that growth rate? And then also clarification there, how much is royalties for next year and as part of your 2030 model?
If you could help quantify that would be helpful. And then on the operating margin side, you want to grow faster and you want to expand operating margins much faster. Is it that the thing you answered to Jim, which is really just getting leverage? Is there something else? What if you were to limit your operating margin, Shelagh, to mid-40s, which is still pretty decent, would you be able to grow even faster? Then what is that trade-off between sales growth and operating margins? Thank you.
So for EDA, as Shelagh clarified, the 13% is the floor. In order for the company to grow at mid-teens, you need EDA to grow close to where the company needs to grow. Otherwise, just the numbers do not add up. With that, the confidence we have, given the solution, and I want to anchor on the multi-physics fusion monetization starting in 2027, will ramp up to the $400 million in 2029. AI starting in 2027. And the expectation that the rest of the portfolio will grow with the market growth for hardware EDA and as the classic core EDA, as you exclude AI and multi-physics. In terms of royalty, the $1 billion agreement with Amazon, the $1 billion is for license fee, and it is for multiple generations of the three products that I mentioned earlier. Royalty is not part of it.
Royalty will get captured as they go into production. So the moment they go into production, there is volume, and we start capturing royalty. Some of these designs will start in the next few months. So then anticipate 14-ish months of design to tape-out and production, and that is when you start seeing the royalty ramping up. Similarly to other agreements we closed with ASIC, with the connectivity as part of that ecosystem, is roughly in the similar timeline.
I will make sure I answer. There is specifically no royalty in 2027 for what Sassine just outlined because that ramps over time.
In the 2030 model, it is a billion?
Shelagh, just to—
The total AOIP is $1 billion. That includes licenses and royalties, but obviously royalties has built up over those generations of chips. We did not give a split.
—Nothing in 2027.
But Sassine gave the intentionality that we're driving to, that royalties will be a lot higher than license. Back to your question on where are we putting the balance between revenue growth and operating margin, our focus is on both. What gives us confidence in doing that is everything Sassine just talked about on the business model, and you can think of the royalty as being 100% pure margin over time. So that also creates yet another lever in margin expansion that we didn't have before.
We will not trade off growth opportunities to just raise the operating margin from 44%- 45% or 46%. We are absolutely investing in the business. Absolutely investing while making priority, leveraging technology to do exactly what Shelagh described.
Thank you.
Okay. What we will do is let us get Jay, and then I will do Josh, and then we will come back to Jason.
Thank you, Jay Vleeschhouwer , Griffin. One of your slides earlier showed core EDA as foundational to growth, and that is undoubtedly true given the size of, let us call it Synopsys classic core EDA, which is still your single largest piece of business. However, and we talked about this, you and I, just a few months ago, that business has shown low to maybe mid-single digit growth. There has even been some sequential decline in a couple of quarters. So to get from that percentage growth to mid-teens, you would have to add anywhere from $350- $400 million a year for that classic business and compound upon that, all else being equal. So what drives that Synopsys classic core EDA business, that improvement over what you have seen in the last year?
Secondly, with regard to the investments you have been highlighting, can you speak a little bit more in detail about what you are doing, particularly on AE expansion, and go to market?
Sure. I am assuming when you talk about the EDA classic, you are talking about the core EDA of software and hardware.
Just software.
Just software. Okay. On the software side, there are two tailwinds that it is becoming very obvious that we are seeing them, and we are in active conversations with customers as they are looking at the next renewal. We have a number of renewals that we are in discussion with customers that they are looking for more capacity for AI. They are looking for the advanced technology that we have, the multi-physics fusion. As we modeled FY 2027, and we communicated this a couple of months early, it was modeled based on a bottom-up roll-up of our contribution that comes from multi-physics, AI, as well as the growth in the business due to more consumption and more need for that software. As we look for the long term, the double digits is based on further acceleration on all these vectors. Our confidence in delivering to it is fairly high.
Otherwise, we will not put it as our long-term guide or specifically for FY 2027. In terms of investments, with AI, the workflow, the engagement with customer is changing, is absolutely changing. As our customers are deploying our full stack or their hybrid approach, further, when you go into the OpenAI Synopsys model, the agent is becoming the expert of how to use the tool. That is very different than the past. Our AE investment needs to evolve and our field investment on how to sell in the world where an agent is becoming the expert user of a task, of a domain, where you have a model that is able to orchestrate reason across. This is not only an opportunity for Synopsys. As you know, the entire software industry is trying to evolve, too.
How does it open up new use cases when you have a model that is able to explore far more than a human can? How do you support the model in that use case? The support is going to come through fidelity checkpoints, because whatever the model propose, recommend, the tool is generating, you need to have a sign-off to check it. That is the uniqueness and differentiation we have in our portfolio. This is where the Synopsys Ansys portfolio brings in the richness of that sign-off as we integrate more and these agents are able to generate and validate.
Yeah. I know you know this, Jay, because you are a deep study of it. But when we gave the EDA growth, that does include hardware. Sassine gave us a preview for an announce, we will be introducing a new hardware platform. The need for hardware across our customers is critical because they are building bigger and bigger and bigger, more complex designs, and they absolutely have to have that insight so they have confidence when they go to tape out a product.
Joshua. Can we make sure that mic is on, please?
Do I need the mic? Oh, there we go.
Yeah.
Joshua Tilton, Wolfe Research. Thanks, guys, for doing this. I thought it was an awesome use of time. I apologize, maybe I am going to sneak two in here. The first one is just, can you help us understand what the pipeline for these Factory Two deals look like? Amazon is awesome, but we are always looking for what is next. So help us understand what some of these deals look like coming down the pipe. Then maybe my second one is there anything in the OpenAI partnership that you announced with this GPT-Synopsys that will keep this type of relationship unique to Synopsys, or do you expect some of your competitors to come out with something similar down the road?
Okay. I chose my word carefully when I say the billion dollar by 2030 is based on the current agreements that we have signed up. Factory Two is not limited to few customers. Factory Two will expand because that application-optimized IP is needed for any COT. Today, you cannot invest and deliver a competitive COT, and as you know, every hyperscaler is building COT without having an application-optimized IP. The strategy we took, and it's been almost a year in the making, is how to use our scale and the current investment in Factory One, continue on delivering without missing a beat, open up a new factory, and have a completely different engagement model with the customer as I outlined. We're going to be embedded with the customer from a system requirement to a system validation.
The best way to learn how to do it is to run with the leader and the company who has been most successful in building their own silicon. The conversations are happening with others, and today is a very important day that we can right now be more open in the conversations with others to say, Here's the model, here's what we've done, here's how we're doing it. That does not limit our opportunity to IP. While we talk about IP Factory Two, when you're embedded at a system level to a system validation, that brings in the multi-physics, that brings in the packaging, that brings in the whole portfolio. I cannot be more excited to have a validation point with the lead COT and their ecosystem and start opening it up as we engage further with customers. So that's on IP. On OpenAI, we're not biased.
We don't pick, we want to work with this versus this versus that. The strategy we took, I want to say close to now seven, eight months ago, AI labs were approaching Synopsys and saying, I'm experimenting with my model using open source, and I'm seeing something very cool, some good outcome, good results. But I know for a fact I'll get far better outcome if I collaborate with you. It's a great conversation. But then, how do we engage while making sure that we protect the Synopsys IP and the skills and knowledge that we're bringing? The requirement we have is our knowledge and IP cannot get sucked into a model that becomes the base model and without our control is available to the world. That's a non-starter for Synopsys. The GPT- Synopsys was a big investment from OpenAI.
OpenAI will have to invest hundreds of millions of dollars to post-train GPT to make a GPT- Synopsys. That does not come for free. That's a big investment they have to make. The investment we are making is bringing skills, assets, R&D, to work with them to fine-tune that model and make it achieve the best outcome between the intelligence and reasoning with our tools. As others are willing, and there's a market traction behind that willingness, we'll assess. As far as what do they do with the rest of the market, et cetera, that's not for me to answer. It's really for the others to answer. But we're very pleased actually with the leadership position we took to architect it, define it for the market.
Hi, Jason Celino from KeyBanc Capital Markets. Maybe to build off of Josh's question with the OpenAI relationship, presumably, outcome-based pricing has been difficult to prove in software, right? Can you maybe just tease out what that would look like with this OpenAI partnership? Because customers, they may be using other models, right? They may be using other competitor tools. What if there's an outcome where you do improve the PPA, but an alternative method improved it more, would that customer pay in that situation?
Yeah. That's why we have three choices for the customer. If the customer in lane two, that they're using our agent, their agent, their model, we're very neutral to that. That's fantastic. We will sell our agents if they're using any of our agents. If they're deciding to just build everything themselves, they need our tools. And the tools, as Shankar showed in one of his slides, it's anywhere between 5- 10x for the verification, the VCS Verdi use case that Shankar showed, that they need more capacity. We have customers coming to us, they're needing more capacity for our software because they're using that hybrid lane number two that I described. Of course, the same thing applies for the Synopsys stack. By the way, if you're using a full Synopsys stack, the one thing we've been able to demonstrate to customers, you can be more token efficient.
Why? We have access to the guts of the tool. We have deep API that they're not available in lane two or three. Again, it's customer choice, one, two, three. The third one, the outcome base. If you heard Greg, the comment he made regarding RSI, where a model can recursively determine the architecture of the silicon in order for the silicon to determine the next architecture of the model. This is not an OpenAI-only thesis. AI has demonstrated over the last year and a half, that each generation of models, the capability is truly exponential.
With that frontier reasoning and the frontier intelligence, with our tools being post-trained, our skills being attached to it, and the knowledge of the chip design, I have no doubt there will be many use cases that the outcome will be able to achieve much better PPA in a much faster time. In few of the customer engagements that it's happening, to be clear, the few customer engagements that they're happening today, they're OpenAI-driven customer engagements, meaning OpenAI buy chips from many. They decide the architecture, they decide what type of spec they need to provide their chip suppliers. In the various handoffs the model is able to prove that use this RTL because it will provide me better power or performance once you take it into implementation.
In these early use cases, that's a wonderful opportunity for Synopsys because that's a revenue stream we would not have captured before because we were not there. We did not play. Right now, we're part of that service offering that OpenAI has. Now, again, for FY 2027, we will see revenue from AI, and as the technology and the partnership evolve, it will only accelerate into the future.
There's a question right there.
Yeah. Here, we'll go with Lee.
Don't want to get into the middle of who gets the question. I'll be quiet. You decide.
Yeah.
Thanks. Lee Simpson, Morgan Stanley. Thanks for today. It was very informative, actually. Just maybe going back to AOIP, I'm just trying to understand where are you hoping to impact here? Because obviously interface IP has been something of a go-to for you guys, particularly SerDes and PCIe, et cetera. Is most of this work going to be done around the IO block? Is that how it's going to work? And where does it stop becoming a chiplet? And then alongside that, it looks as though a lot of this is going for the hyperscaler market. Will this include China as well? Will we see regrowth in China? Thanks.
Yep. Yeah. Excellent question. There are really about five IP titles interfaces that customers are needing desperately an optimization of that IP because it consumes quite a bit of area. The optimization of that IP will provide a higher bandwidth, lower latency, and these are the one that you could imagine what they are, the PCIe, the 224G type of an Ethernet SerDes. HBM is all customized, so custom HBM, it is to optimize the logic to memory interface. UCIe, even though it is called a standard, there is nothing close to a standard the way UCIe is implemented by many customers. The way we have engaged with AOIP with the customers, we are not picking and choosing which IP needs more customization, and I will pay you only for this.
They will come to Synopsys for their IP needs for a system, and the reason is the customization fee, not an NRE. Think of it as a priority fee to put our scarce resources to work on that engagement, and based on that, we will capture the royalty. Will other customers ask for it beside the hyperscalers? Yes. I want to say over time, meaning if you are looking at an IP optimization for robotics, for automotive, it is really a trade-off. How much do you get by customizing the chip? Do you, the customer, wants to make that investment to customize that chip? The decision Synopsys needs to make, is there much upside if we do that effort? Because again, it is a royalty base and scarcity of resources to deliver to it.
As far as China, the China opportunity as it relates to IP has slowed down, in particular for the most advanced IP, because China is unable to design in China the most advanced IP. They do not have access to Gate-All-Around, they do not have access to 3D IC. It is not us, our desire to do it. Our Chinese customers are looking for ways to continue on investing and designing given the restrictions and constraints. We will absolutely engage with them in a similar model as we have for the rest of AOIP, but it all depends on the when, the how, to drive it.
Okay. We will go to Andrew, then we will get Charles, and then Ashish.
Thanks. Andrew DeGasperi from BNP Paribas. Just in terms of the three lanes you discussed earlier—
Andrew
—I was just curious to know. I am right here, sorry.
I am looking for Andrew. I just hear the voice.
Just curious to know, which one do you think delivers the best economics to Synopsys, if you were to fast-forward in four years, and which one do you think would be the most popular with your customers?
I missed the first part. For AI?
Yes. AI.
The three revenue streams?
Yeah. Full stack, customer-owned platform, and I think frontier models.
Today, I want to say most of the explorations is in lane number two, naturally, because customers, they are all racing to figure out, "I am seeing some good outcome with AI. How do I integrate it into my workflow? How do I change the way I am doing design, leveraging AI?" Almost every customer in lane number two have came to Synopsys and said, "You know what? For the debug agent, for the linting agent, for the whatever other agent, you already have it. I benchmarked it. I do not need to do it myself. I will buy it from you." That is why it is not where does Synopsys make the investment. We will absolutely continue on investing on having the full stack because that investment in the full stack are needed for two and three.
Because the OpenAI, I go back to what is the relationship with OpenAI?
Synopsys is bringing not only the tools to train the model, we are bringing our agents, we are bringing our workflow. Today, most of our customer engagements are in two, then one, then three. Do I envision there will be more post-trained models, frontier models, similarly to what we have done with OpenAI, with other companies that they see the same opportunity that OpenAI is seeing, and they are willing to invest in it? I believe yes. It will absolutely expand there because investing in a frontier model and frontier intelligence, you are not going to make money by just having the intelligence. You are going to make money by having the whole stack, including the compute, the service that comes with it for different markets. Engineering, it is such a sweet market because it is complex. Can you bring in that added value?
I do believe lane three will expand beyond one frontier relationship that we have right now with OpenAI. I believe lane number two will mature and use more of Synopsys agents because the customer does not need to reinvent and put their resources if the agent is doing a better job that they can get from Synopsys, and that is how the EDA matured over the years. Customers or IP, they build their own, then they say, "It is not worth the investment. I can get good support and a competitive offering from Synopsys." That is how it is today, and that is how I see it going into the future.
Charles.
Sassine, Shelagh, thanks for taking my question. This is Charles Shi from Needham. I have a question on lane three, lane number three. I think one of the things people were worried about throughout this year has been can AI design chips completely bypassing the EDA tools? There has been a thought that, okay, to post-train the AI model, you do need the design data, and frontier model companies, maybe except one, do not have chip design data. But when you do this collaboration, there seems to be a possibility for them to use your EDA tools, generate synthetic data to train a model. The question is this: Does this eventually evolve to a future where there is actually going to be fewer tool costs or maybe no tool costs at all in the future? You completely just use the models. That is question number one.
And the other one, kind of related to this, I understand this is a specialized model. The Synopsys IP is kind of contained within that model. It does not go to the general purpose LLM. But will your customers be concerned about connecting their data into this model? Because there is probably going to be multiple customers using the same specialized model, and how are they thinking about protecting their own data? So that is the second question. I think this is going to be related to the adoption, any inhibitor to the adoption. Thank you.
Yes. So, on the first question, definitely not. There will not be a point where you can bypass EDA completely. How will the model create a placement of the gates that the RTL. Once you create an RTL, you synthesize it, you have to place it, you have to route it, you have to create the clock tree for it. All of that has to obey the rules of manufacturing that comes from TSMC or Intel or Samsung. Do you know how often these rules are updated? Sometimes customers get new PDKs two, three times in one tape out. Who validates them? The model cannot do that. The model can reason, recommend, explore. Let us assume in some cases it can generate. When you generate, you need to validate. That you need to validate that step C can go to step D without wasting energy.
Then you get to GDS, and TSMC will say, No, thank you. This is violating all kind of my physics rule base. So I do not see how a model can bypass EDA generating of data and signing off and checking the data. As far as question number two, that has been big part of the conversation with OpenAI. As I mentioned, there are already number of customers in that model where OpenAI, that is not a Synopsys, OpenAI will own the security, the containerization of the customer data, protecting the data, securing for the customer that the data is protected. Same when Synopsys engage with the customer in the classical model. If we get a customer data or use case or what have you, we lock it up for each customer, so there is no contamination.
In that new model that is owned by OpenAI, what we do in that relationship, that is why I go back to it is a service relationship that covers many aspects. What we bring in are the tools, the agents, the skills, where the model, et cetera, will be run and through the OpenAI, either on compute or where they host the model can be through AWS, through Azure, through whomever.
Go on.
That discussion has happened with few of the early customers, and those are leading customers that they are comfortable with the working model and how to engage that way.
Back of the room.
Hey, Sassine. Hey, Shelagh. Ashish Bhandari from Throughline Capital. Thanks for taking the time and doing this event.
Over here. Yeah.
Can you see me? Great. I just had one follow-up on GPT- Synopsys I think this opens the aperture to partner with OpenAI on different fronts, including on the traditional Ansys simulation portfolio. I guess I would be curious what those kinds of partnerships could look like in the future and just how those conversations are going. Thanks.
We will absolutely expand beyond semiconductor because it is the same problem. As I mentioned in one of the SNA slides I had, there is the surrogate model evolving into what Shankar presented, the Autopilot to, Frontier model for physics. So it will absolutely evolve there. In semiconductor, the workflow is fairly well-defined. So I want to say the scope of the engagement, the workflow with the customer is fairly well-defined. In simulation and analysis, different industry has a different workflow, if you are working with an automotive, with aerospace, with robotics, with drones, but the opportunity is absolutely there. There will be an expansion. It is not limited for semiconductor, and these discussions are happening. As they evolve, of course, we will be very excited to share with you how do they evolve and how do we monetize them.
Gary.
Thank you so much for taking my question. Gary Mobley at StoneX. Thanks for hosting this event. Very informative. So under the idea that some of these long-horizon AI agents are market expansion opportunities, I think that is another way of saying you can drive higher average deal sizes at customer renewal. So assuming that somebody takes Agent Engineer in its most fully loaded form, how incremental can it be for a license renewal with a longtime customer? Then maybe if you can help explain how using Agent Engineer or any other customer's agentic engine, how that drives more usage of traditional EDA copies per chip design specifically.
Yeah. Shankar mentioned we have 50-plus current engagements. Those engagements are mostly around our agent engineers sitting at our customer platform. There are few engagements that we are offering the complete platform, and the platform is much more than just the orchestration of the agents. There's a lot of telemetry and intelligence that goes into connecting from the compute layer, model layer, all the way up to the domain-specific agents. The example we showed where it's 5- 10x more VCS and Verdi licenses for that particular use case. Our customers are seeing it. We have number of renewals discussions happening today with customers that they're constrained by licenses. They're saying, "I don't know how much I need. I don't want to overcommit." That's why we're offering both a subscription and a consumption for the license, regardless what you're using for the subscription or the platform or the agent.
With few large customers, and they happen to have renewal timeline in the window that we're talking about, the conversation right now is how do we provide flexibility as our customer is learning? Because they're not going to commit for three years at a certain level of capacity if they still don't know yet what capacity they need. The one thing they know is, "I need more capacity." As you know, each customer engagement is different. It depends on the baseline of the tools they have, how do they grow it, et cetera. That's why I go back to the confidence we have that it is more consumption for the tool because we're having these conversations with the customer. How much of it will be consumption versus the classical subscription? In this early stage, it's primarily the customer comfort zone is subscription because that's what they're used to.
They're comfortable with it. We're completely okay with it. That gives us a much better visibility and a revenue streamline that is very predictable. It's meeting the customer where their needs are at. That's the approach we're taking right now, and in each one of those cases, it's starting at the most fundamental layer, which I need more licenses from you.
All right. Is that Kelsey? I can't see.
Hi. Kelsey from Citigroup. I have a question on operating margins. As you build out your Factory Two more customers, would you need to allocate more R&D dollars there?
The ex—
Would that impact your long-term operating margin target?
—I have answered that before in the following way. We were already doing customization for customers, but we were not getting paid for it. That is why our confidence, if you remember two, three quarters ago, we are like, "We know how to do it. We have been doing it. We charge for it as NRE." Our engineering team, we know how to do it. To the earlier question, why do you stop at IP, not a subsystem? We are delivering subsystem to customers. We are delivering, in many cases, customized IP. The Amazon engagement, what is unique to it, and I am truly looking forward for us to learn how to be embedded early at the system definition all the way to the system validation.
That does require some new skills, and Charlie and team are expanding and prioritizing these skills as we had many conversations with Amazon to set the right expectation of what does Synopsys own, what do they own, what is the handoff, how do we become part of their team? All of this took the last four, five months of conversation to say, "Here is the engineering expectation and the type of engineering I need to move forward." From an operating margin point of view, I will have Shelagh comment more on it, the reason, again, we have the confidence that the operating margin will only improve is, again, we were doing a lot of the work. Right now, we need to get paid for the work at a much higher rate than what we have been getting paid for.
The opportunity to use a new method of customer engagements and technology, we are absolutely absorbing it at a fast pace inside our IP R&D team.
Yeah. Kelsey, the operating margin does include investing in AOIP and building out that factory to support those further customers and design. So that is fully incorporated. Over the horizon, we add in royalty, which is also 100% margin, and we have not had that before in our business.
Thank you.
All right. Thanks, everyone. I know there's other earnings today, and you guys want to get to those. Sassine, if you just want to close out and say thank you.
No, really. Thank you so much. I hope our enthusiasm and excitement around how do we take advantage of the market opportunity and redefine our business model in IP, in EDA, and in physical AI. I truly cannot be more excited about our opportunity, and I look forward to every quarter committing and delivering, and we do what we say and say what we do, and we need to keep up with that promise. Thank you for taking the time. I look forward for more conversations with you. Thank you.
This concludes Synopsys Investor Day. A replay of today's program and the accompanying materials will be available on the Synopsys Investor Relations website.