AMD Advancing AI 2026: Lisa Su Full Keynote

Good morning.

That's a pretty good morning. How about let's try one more. Good morning.

And welcome to Advancing AI 2026. It's so great to be back here in San Francisco and to see so many friends, partners, and customers, and especially all the developers that are here with us today. And I want to say a big welcome to everyone who's joining us online from around the world. This is my absolute favorite event of the year. It's where we bring the entire AI ecosystem together to show what we've been building and where we're going next. And this year, this is our biggest show ever because we have so much to tell you. So, let's go ahead and get started. At AMD, our mission is to push the boundaries of high performance and AI computing to help solve the world's most important challenges. I often say AI is the most important technology of the last 50 years. The progress we're seeing is incredible. Every month, every few months, we see something new that was far beyond our imagination. You can already see the impact across every industry. In healthcare, AI is helping researchers identify new drug candidates faster. In science, we're solving problems that we didn't think were possible a few years ago. And across every industry, AI is changing the way we get work done. The most important thing is we're still in the very early innings of what's possible.

Now, just take a look at some of these charts. We look at these every year and you kind of see the rate and pace of AI growth. A couple of years ago, we were just starting. People were experimenting with chatbots and it was really cool. But if you look at today, we're seeing more than 35 quadrillion tokens consumed every single month. That's an increase of nearly 160 times in just two years. That curve is just getting steeper. You're going to hear a lot about that today. So just looking at where all that demand is, obviously training remains incredibly important and foundational. Over the last four or five years, we see the compute used for training advanced models continuing to increase by roughly 5x every year, and those models are getting better. We're seeing better reasoning, new capabilities, and a growing number of specialized models. One of the things I really believe in is there is no one perfect model. We're going to use a slew of different models depending on what task you're trying to solve and which industry you're in. But the bigger shift is actually in inference.

And we expected this. We expected that inference would grow faster than training. We can say that in 2026 for the first time the world is using more AI compute to run models than to train them. We think this year roughly 60% of global AI compute capacity will be used for inference. The reason is simple: as billions of people use AI every day, the workload shifts from building models to putting them to work. We're going to talk a lot today about agentic AI, which is accelerating the shift even faster. We see agents as the next big step for AI. This has really just started accelerating over the last five or six months. When we started with LLMs, they were great for answering questions, but to really get the power of AI, you want agents to be able to answer full questions. You give it a goal, it keeps working until it solves the problem. That means compute demand is growing at an incredible pace. We're seeing a step change because when you ask an agent to do something, it has dozens of steps, it has to reason, call tools, access data, and keep doing it over and over until it solves the problem. So you need lots of GPUs for reasoning, but importantly, you also need a lot of CPUs to orchestrate every step around it.

So when you look at what that means from a market standpoint, it's hard to put this market data up because every few months we're changing the perspective based on what our customers are saying. But what we're seeing is that the shift to agentic AI is growing the AI accelerator market very significantly. Last year at this event, we called the AI accelerator market at about $500 billion by 2028. At that time it felt like a very big number. But honestly, you look at it today and it looks conservative because AI demand is continuing to accelerate. Better models, more usage. More usage requires more compute, and more compute builds better models. So we now expect the AI accelerator market to reach about $1.4 trillion by 2030. What that means is by the end of the decade, the AI accelerator market will approach the size of the entire semiconductor market today. Although there will be many different types of accelerators, I'm a big believer that there's no one-size-fits-all. We do expect GPUs to make up the vast majority of that market because algorithms are still in their infancy and workloads continue to change, favoring programmability in the overall silicon ecosystem.

Now perhaps the most interesting part of the last six months has been that GPUs are only part of this story. Agentic AI is creating an entirely new growth vector for server CPUs. I've updated this number a lot over the last six to eight months, but we call it like we see it. We were early. We saw from our largest hyperscale customers that as AI inference goes up, they need more server CPUs. So we saw growth in the server CPU market. Last year we thought it might grow at 18 to 20% CAGR to an overall market size of $60 billion. But the rate and pace of agentic AI adoption is much faster than any of us thought. We're talking about agents going from millions to billions. That requires a tremendous amount of CPU infrastructure. Every customer conversation tells us that this buildout is just beginning. Based on what we're seeing today, we now expect the server CPU market to grow by over 50% to over $200 billion by 2030. That's starting from today's $25 billion market. So there's a lot of excitement about CPUs as well.

Now probably one of the things differentiating AMD is that this is not just a data center opportunity. As AI becomes a larger part of our daily lives, we want intelligence to run right where the work happens. Cloud is super important, but the devices we use every day are going to be extremely important. We're going to do a lot more at the edge as well with autonomous machines that sense and act in real time. We believe AI needs to be infused everywhere, and that's exactly what we're focused on at AMD. Putting all of that together, we expect the market for our high-performance and adaptive computing products to grow at roughly a 40% CAGR over the next several years, approaching a $2 trillion market by 2030. The only way to service a market like that is for us to work together as an ecosystem. No one company can solve it all. But this is the opportunity to bring the best and brightest together. We have never been in a better position to lead. We have the broadest product portfolio, the strongest roadmaps we've ever had, and the deepest partnerships with the companies building this future.

So a little bit about our strategy. We've been very consistent. Our multi-year strategy is built around three priorities. First, compute leadership: we're building the broadest set of compute engines in the industry so that we have the right compute for the right workload. Second, open platforms: we want everyone to come together in an open ecosystem. We believe an open ecosystem is essential to the future of AI. That's how we get the force multiplier of everyone coming together. That's open on hardware and open on software. That's why we're investing heavily in our ROCm software. AI has been a tremendous multiplier in the rate and pace of progress we're making in ROCm and software. It means that if we put all of this together, our customers can deploy on AMD hardware faster than ever. Third, powering AI everywhere: AI in enterprise, AI at the PC, AI in the physical world, and adding new capabilities across all our products. Today, you're going to see all of that in action. We're going to start with compute for the agentic era, show you a lot of hardware, talk about software, and our overall products. We're honored to have some special guests to help bring the technology to life. So, let's start with the data center.

Today, AMD EPYC runs on the most important workloads in the world. We power the largest cloud providers, the digital platforms that billions of people use every day, and the most important critical systems used by the largest businesses. More than 60% of the Fortune 100 use AMD EPYC, and that momentum is building. Last quarter, we reached a record 46% revenue share in the server market. We're continuing to see every major customer move more workloads to EPYC. On the GPU side, our Instinct adoption has also accelerated. We have a broad set of customers across the largest AI labs, cloud providers, leading AI startups, and national labs. Sovereign AI opportunities are being built on AMD Instinct, and we appreciate that opportunity.

Now, AI compute is becoming a lot more complicated. What we're seeing with frontier AI is that it's raising the bar for what infrastructure needs to do. It takes more than a single chip or a single server. You have to design the entire rack as one system. That requires leading CPUs, leading GPUs, high-speed networking that connects everything inside the rack and across the data center. And just as importantly, this is about making these systems super easy to use. They need to be easy to deploy, easy to service, and very reliable. That is exactly what we built with Helios. So today, I'm super excited to launch Helios, the industry's highest performance AI rack.

Helios is built to train and run the most demanding frontier models in the world at massive scale. I have a lot of show and tell for you. For those who know me, I love holding up chips. I call myself sometimes the Vanna White of chips. But with Helios there are lots of components. These are the chips inside the Helios rack: Instinct, EPYC, and Pensando chips for networking. Every one of these chips is part of the system. MI455 is the engine, the highest performance GPU in the industry. Venice which drives it is the world's fastest CPU cores. Pensando DPUs and Nyx connect them with leadership programmability and scale-out bandwidth. Let's look at how these fit into the system. This is the MI455 accelerator mounted on an enhanced accelerator module (EAM). It's a production module that goes into Helios, much more than the GPU package. The EAM integrates the GPU, memory, power delivery, high-speed interfaces, system management, and cold plates for liquid cooling into one compact serviceable module. Pretty cool, right? Specs: 320 billion transistors, built on TSMC's leading 2nm and 3nm process technology. It brings together nearly a decade of AMD chiplet innovation, combining 12 compute and IO chiplets with 432 GB of memory, all connected by leading industry 3D chip stacking packaging. It's the highest performance AI accelerator in the industry.

This is the CPU board inside the Helios compute tray. It's a high-speed 96-core EPYC processor with DDR5 memory and all the IO needed to feed the GPUs, all in a single motherboard. We also have our Selena DPU, a critical part of the networking infrastructure that allows us to deliver both front-end and scale-out and scale-up bandwidth. This is our Volcano AI NIC, which allows up to six Volcano NICs on a board, using the open Ultra Ethernet standard. Each Helios compute tray includes two of these Volcano boards for scale-out and scale-up networking. Inside the rack, we have an additional six dedicated networking trays that handle scale-up networking, connecting 75 GPUs together with UA link over Ethernet with silicon from our partners. That's what we mean by an open ecosystem. When you bring it all together, Helios is simply the best AI rack in the world.

Let's take a look at some numbers. It's a very competitive world. When you compare Helios to the competition, we're delivering 15% more compute, 50% more HBM4 memory capacity and bandwidth, and 50% more scale-out bandwidth. That means every Helios delivers more performance for the largest models, more capacity for longer context, and the bandwidth to scale across thousands of racks. Here is the production hardware that customers are deploying. Each tray weighs more than 160 lbs and stands less than 2 inches tall. The numbers are incredible: more than 18,000 CDNA 5 GPU compute units, over 4600 Zen 6 CPU cores, and 31 terabytes of HBM4 memory in a single rack. That's what it takes to run agentic AI at scale.

Today, I'm excited to announce that Helios is in full production. Shipments are on track to start at the end of the third quarter, ramping into the fourth quarter and the second half of the year. Customer demand for Helios is extremely strong. We're proud of the work done across the leading AI labs to adopt Helios. Let's start with some of our guests today. I'm excited to be joined by our newest Instinct partner and one of the leaders in frontier AI. Please welcome Anthropic co-founder and chief compute officer Tom Brown.

Hello Tom. It is so great to have you here and it is really such an honor to be working with Anthropic. We just had a big week this week.

So, look, Anthropic has just had incredible success over the last few years. Claude has made such a big impact on the overall industry. Compute is such an important factor in progress. Can you talk a little bit about your compute strategy and how things have been evolving?

Yeah. Look, I think you've been really leading in that area. I'm happy to say that I've always wanted to be part of your infrastructure. I remember the early times we talked as you were starting Anthropic. We had a big announcement this week: we're honored that you will be deploying up to 2 gigawatts of Helios. Why now? What was the key reason?

Do we like that? Does that work for us?

Tom, I kid you not, when my team told me that you had an engineer working on MI355, I said, 'Hey, are we making sure we help them?' And they said, 'They say they don't need any help. They got it all covered.' I said, 'That's a wonderful story. Thank you for that.' Now, we're also super excited about the work we're doing together because one of the most valuable parts of this partnership is Claude and using Claude within AMD. We're big believers that AI is a force multiplier, and the more we have leading foundational models familiar with AMD architecture, the better we'll be able to service the overall market. So can you talk a little bit about how you see Claude evolving in these highly engineering-specific workloads?

Yeah. No, I think on the software side, the work we've seen with Claude and kernel development has been incredible. So, Tom, one of the things for our audience here: we like to talk about the present, but most of us are working on the future. This is the beginning of what we believe is a very strong multi-year partnership. Can you talk about what you're most excited about in the industry and what we can do more together to satisfy those key opportunities?

No, I think you're completely right. We are super focused on how we can use AI to improve every aspect of our product, whether it's performance, power, software, or security as you said. So, Tom, huge thank you for the opportunity to work closely with you and your team. Our team loves working with you guys, and we look forward to everything we're going to do together.

Thank you.

All right. So, Tom talked about how important performance is. We have been laser focused on ensuring that the real-world performance of MI455 and Helios comes true. Let's take a look at some results. Here's an example running DeepSeek V4 Flash, a reasoning model. We're comparing MI455 to MI355. With every generation, we want huge leaps in performance. At lower concurrency, MI455 delivers 4x the throughput of MI355. But at the highest concurrency, Helios delivers up to 34 times more throughput than our prior generation. That means more users, much faster response, and much better efficiency at scale. Performance is just one aspect. The other is cost per token. Customers care about cost per token, and with every generation of Instinct, we're driving it down. With MI455, we're delivering up to 18 times more tokens per dollar, so customers can serve far more users with the same investment. And when you think about how that comes together in a rack, many data centers are limited by power. We tested Helios over a full range of workloads at fixed rack power compared to the competition, and we're seeing an average of 10 to 15% more performance across leading inference workloads. That translates to up to 30% more tokens per dollar than the competition. That's a pretty good value proposition.

Now I'm happy to say that customer demand for Helios is extremely strong. From the largest AI labs to hyperscale and neocloud providers, we are working across the entire ecosystem to enable Helios together with our OEM and ODM partners. One of our deepest and earliest partners deploying Helios is OpenAI. To talk about where AI is headed and the work we're doing together, please welcome OpenAI's head of infrastructure, Sachin Kati.

Thank you, Sachin. It's so great to have you here. Most importantly, we are so thankful for the partnership with OpenAI. We've been through so much on this journey together. You're really operating on the frontier of AI. Can you tell us a little bit about what you're seeing and what compute looks like for you?

Sachin, I don't think I've ever spoken to you where you haven't asked me for more compute. You're very consistent. Look, our teams have been working really closely together. It's been a true journey thinking about all the technology we've been doing. Last year we announced a landmark partnership where you're going to deploy up to 6 gigawatts of AMD infrastructure. You were one of the first to have MI455 racks a few months ago. Can you talk about the journey?

Like I said, I know more faster.

So, Sachin, one of the most exciting parts of our collaboration is some of the research work. Thankful for the opportunity to bring our best engineers together with your researchers to talk about how AI should be implemented and used for future silicon and software. Can you talk a little bit about that work?

I think super excited about that. It's really the tide that lifts all boats. As we work together, it helps our work together but also helps the overall ecosystem. So a little bit of your crystal ball looking forward: we all know that compute is super critical. What do you need from the next generation AI infrastructure, and what can we do as a key partner to help you accomplish that mission? If you haven't heard already, I need more compute, more, more, more. But at least he's consistent.

That's fantastic, Sachin. We really recognize that input is so valuable. Thank you again for the extraordinary partnership. The amount of work our joint teams are doing together, we couldn't be more excited about the opportunities in front of us. Thank you, Sachin.

All right. So, now let's turn to the world of CPUs. There's been a lot of talk about CPUs recently, and I can say this has been the foundation of our AMD data center strategy for the longest time. We launched EPYC in 2017 and have been on a very clear mission: every generation, more performance, more efficiency, more capability for a wider range of workloads. Naples was our first generation, establishing the foundation. Rome and Milan changed the economics in the data center, adding core count and throughput. Then Turin and Genoa expanded performance, and Turin set a new bar for density, throughput, and total cost of ownership. That's the EPYC formula: a clear roadmap and consistent execution, with leadership that grows every generation. Fifth-gen EPYC Turin is the best server CPU in the world, with up to 192 cores and 384 threads, delivering leadership performance across cloud, enterprise, and HPC workloads. But compute never stands still. With agentic AI, it's creating a whole new class of infrastructure, and the CPU matters now more than ever. That's exactly what we built Venice for. There's been a lot of talk about agentic AI, but it's a very new field. Server computing is splitting across different workloads. First, GPU servers: the CPU's job is to drive the GPUs, all about speed, highest frequency cores, fastest IO, keeping GPUs fully fed. The biggest middle part is the largest growth: agent servers or agent sandboxes. Here you execute code, call tools, query data outside the model. The priority is density: highest performing cores per watt to run thousands of agents at once. Then we have traditional general-purpose servers for enterprise, running applications, databases, data services. With these different workloads, you need the right CPU for the right workload. EPYC is the only CPU portfolio that leads across all three.

Venice is our newest and most advanced server CPU family, designed for the agentic era. We're extending Turin's leadership across every metric: performance, efficiency, TCO for cloud, enterprise efficiency, HPC workloads. It starts with our all-new Zen 6 core, with higher IPC, higher frequency, delivering up to 1.8 times more performance than Turin. That's one of the largest generational gains in the history of EPYC. It's 203 billion transistors built on TSMC's newest 2nm process, and uses a next-generation chiplet that supports up to 512 threads per socket. This is a really important point. It gives us the highest compute density, but also allows us to double both memory and IO bandwidth. I have a few more chips to show you. Venice is not one chip, it's an entire family of chips. Our chiplet architecture is a huge advantage, letting us take the Zen 6 architecture and build a whole portfolio. Let's start with this guy: Venice HF. This is the highest performing CPU that drives AI host nodes. It has eight compute chiplets, each with 12 cores running at up to 5 GHz, with the IO and memory bandwidth to keep GPUs completely fed. This is the CPU I showed you before, shipping inside Helios. Now, this is its bigger brother: Venice 256 core. This is built for agent sandboxes. It also has eight compute chiplets, but each has 32 cores, scaling to 512 threads. It is the highest compute density in the industry. Customers get more agent capacity per watt, per dollar, per rack, bar none. And one more member of the family: Venice 128 core, optimized for enterprise and general-purpose servers. It has the same leadership performance per core, but in a lower cost, lower power design for enterprise applications, tuned to give customers the best performance per dollar. You can see the power of the family as they come together. Next year, we're adding a few other chips: Verona, the next generation for AI host nodes, with more power-efficient low-power memory and faster interconnect between CPU and GPU. Also for high-performance computing, Venice X, using our 3D V-Cache stacking technology, stacking memory chiplets right below compute chiplets to accelerate performance. One architecture, a CPU is not a CPU; there are many different types, and for us, one architecture spans all.

Dozens of different chips and that breadth is something that no other CPU vendor can match. And that's what makes sixth Gen Epic the best CPU in the data center.

Now, you're going to bear with me because I'm going to take you through some performance because the numbers are just incredible. So looking at AI workloads, again, you have different use cases and different workloads. Starting first with GPU servers. Many of the largest frontier models require you to move data between the CPU and the GPU. And in that scenario, Venice moves the data faster than the best competitive x86 processor, delivering up to 1.8 times more tokens per second. And when you look at the Agentic CPU servers and sandboxes, because of our density, Venice delivers more than twice the agents per watt for agent sandboxes. And for general purpose applications, in this case, we're standardizing on a 100 kilowatt rack, Venice delivers more than twice the performance per watt. So pretty incredible results.

And the gap is even wider if you look at ARM processors. So if you look at the leading ARM processors, when we're comparing at the chip level, Venice supports up to 2.8 times more agents per watt. And when you look at the rack level, Venice is delivering up to 3.3 times more performance per watt. And what that means is at the data center scale, you can just get a lot more agents in the same power envelope.

Now, there's been a lot of talk about what matters most for agentic AI with some saying that per core performance under load is the only thing that counts. But that's really only part of the story because agentic AI actually runs as a distributed platform and that is across databases and vector stores and orchestration and all of that has to happen at once and that requires not just raw speed but you absolutely need density efficiency and per core performance all together. And what I'm happy to say is that Venice leads across every one of them. So when you compare Venice against the highest performing ARM CPU from our competition, Epic delivers 20% higher per core performance.

I like that number. I like that number. And when we pair that with leadership core density, we're delivering 2.2 times more performance at the socket level. And so the main point is whether a customer needs per core or per socket performance, Venice is the leader. And there's one more point that a benchmark doesn't really capture because when you think about what agentic AI is, it's really like you're adding thousands of employees to your enterprise, to your traditional workflow, and that software enterprises have been running on for years.

The truth is that software runs on x86. So Venice runs all of it. And this is where x86 has an advantage. You must have the best performance. You must have the best density. But having that software compatibility is a big big plus. So customers scale up their agents on the stack that they already have.

And now let me just show you what that means at the rack level. So at the rack level, what we find is, density is super important. Everyone's trying to optimize their data center. Everyone's trying to optimize their power envelope. And with Venice, every major server OEM is offering a broad spectrum of racks for agentic AI. So what that means is we have the full spectrum from 25,000 cores to 50,000 cores and that's more agents per rack and that gives customers the choice. The key is every data center is different and with that flexibility allows you to pick the right cooling, the right space requirements for what you're trying to do in your data center. So there you have it, Venice the best server CPU in the industry.

And I'm very happy to say also today that Venice is in full production. Customer demand is incredible. It's the strongest we've ever seen for a new Epic generation.

We're seeing every major server OEM, every major cloud provider on track to begin rolling out in the fourth quarter as we start with making the broadest Epic launch we've ever had. So with that, I want to turn to my next guest who runs some of the largest and most advanced compute infrastructure in the world. And it's really been our privilege to partner with them as they've deployed multiple generations of both Epic and Instinct. To share more about our work together, please welcome to the stage Meta's head of infrastructure, Santo Dennard.

How are you? Wonderful. It is so great to have you here. Thank you for being here.

Santosh is a true friend. I have to say, it's been a tremendous opportunity over the last few years. My story about Santosh is the first time I sat in his office, he said to me, Lisa, we're going to deploy a lot of CPUs. Please make sure they work. And I said, I will. I will.

I try.

Thank you Santo. And look, we completely agree with your philosophy. Now you are actually one of our broadest partners because if you think about our work on CPUs, GPUs, we developed the rack scale OCP systems together, you've deployed lots of CPUs, millions of CPUs, and you're a lead partner on Venice. Can you just talk a little bit about your CPU and infrastructure and some of our work together?

It's like every month.

Yes. No, absolutely right. And we've gotten a lot of feedback from your technical team. I think that's what I really appreciate about the partnership with Meta. Now clearly you're deploying lots of GPUs too, a lot of accelerators, and you're also one of our deepest partners on the accelerator side starting with MI300 and then we had a very large strategic partnership announced around up to 6 gigawatts starting with MI450 and what we're doing is quite unique with Meta. So can you talk a little bit about the evolution and the MI450 plans?

They're not shy for sure.

Yeah. Absolutely. Really appreciate the effort together on MI450. Now I'm going to ask you also about your crystal ball. So you look out into the future and you see what the next few years really means as you push forward in the AI industry. What are the biggest challenges and really opportunities for us? This is the industry ecosystem here. So what should we be focused on as an industry?

Fantastic. Santo, I completely agree with you. I want to say again, thank you. We are so happy with the amazing partnership that we have across the board, and most importantly, with your engineering teams, and we truly are excited about what we're going to do in the future. Thank you so much.

So, you heard Santos talk about just the diversity of workloads and the massive scale that you require with AI. Now I want to turn to another part of the inference market where the requirements are actually very different. So lots of workloads and as inference is moving into more products and services, we're actually seeing it segment into different workloads depending on what you're trying to do. Some of these are less interactive, so they're really tuned for maximum throughput or the lowest cost. Other applications require more balanced throughput, so you have to balance throughput and responsiveness. And then there's this new class of applications that actually want very fast results. You have extremely smart engineers and they don't like to wait. And that is where ultra low latency or every millisecond matters. Each one of these types of workloads requires a different type of compute. And one of the best ways to reach ultra low latency today is disaggregated inference. So if you think about the different pieces, both prefill and decode are two different jobs. Prefill tends to need more compute, decode tends to need more memory bandwidth. And with Instinct we have a very balanced machine so that we deliver both great compute and great memory bandwidth. But you can actually take this a step further if you know what workloads you're trying to run. You can actually let customers tune each of these pieces independently. And to address this market opportunity, we've been working with Cerebras. So to talk more about what we're building together, please welcome to the stage Cerebras co-founder and CEO Andrew Feldman.

Andrew, it's great to have you here. It's been an exciting few months for you. I know that for those who don't know exactly what you've been working on, can you talk a little bit about Cerebras and what problem have you been trying to solve?

There's no question, Andrew, that you have some tremendous innovation with what you've been working on and our teams have been collaborating really closely over the last few years to really have Helios together with the wafer scale engine for ultra low latency inference. Can you just kind of educate the audience a little bit? What are we trying to do and what does that mean for customers?

Yeah. Look, the technology is pretty cool. So just talk a little bit about how it comes together.

So very exciting. I know that customers are excited about what we're going to do together as well. So, when are we going to have it in market?