# Intel Computex Keynote 2026

https://www.youtube.com/watch?v=1h_zY377urU

[00:02] Silicon, the foundation of modern technology.
[00:07] Every transistor placed with purpose.
[00:10] Every watt wrestled from physics.
[00:13] Where every instruction set earns its right to execute.
[00:18] This is how performance gets driven, how efficiency gets built, how intelligence acts, not just answers.
[00:26] Our future is with ecosystems.
[00:30] Shaped by architecture.
[00:32] Capable of connecting technologies, tools, and partners.
[00:37] Powerful enough to execute.
[00:37] Efficient enough to scale.
[00:40] Familiar enough to build with speed.
[00:46] Tomorrow, it's about the progress we scale through open platforms, shared standards, and partnerships amplifying each other's strengths.
[00:55] A unified architecture engineered for systems.
[00:58] The next chapter is being written in silicon with our engineering at its heart.
[01:00] Built
[01:02] different, built together, built on Intel.
[01:06] And now, ladies and gentlemen, please give a warm welcome to the CEO of Intel, Li Panel.
[01:31] computer.
[01:42] uh the steps.
[01:45] Uh this is the elephant mountain and so 1,000 step to 184 meter and so I survived and then came down with one piece.
[01:59] So I'm here.
[02:01] So but if I walk a little bit
[02:03] Slower then you know, uh, I'm exhausted.
[02:08] But anyway, it's a beautiful view.
[02:10] I mean, highly recommend all of you to do that.
[02:14] Yep.
[02:14] I think first of all, I think uh, delighted to be here and uh, this is an important event and then I'd like to uh, get started.
[02:26] And about nearly six decades ago, a group of brilliant and highly motivated engineers and venture capitalists, uh, including Rock Don Valentine and many others, if found companies like Intel, Apple and others, broadly set the motion the largest economic known to mankind, uh, create what became known as Silicon Valley and uh, this is a very exciting time and that is a very same ambition.
[03:06] and mindset with the semiconductors make across the ocean.
[03:12] that sparked the creation of Silicon Island right here in Taiwan.
[03:16] And I have been very fortunate to associate with the creation of semiconductor industry in Taiwan 40 years ago.
[03:27] uh because uh 40 years ago uh Minister KT Lee Lording uh invite me to lay the foundation of venture capital concept in Taiwan.
[03:36] It's a very new concept and uh you know you people put money with you and you play the money and then you share the profit of 20%.
[03:52] But you don't share 20% of the losses.
[03:56] So this is a very unique uh venture capital concept but I managed to do that and then with the help of minister KT Lee and then development fund and I set up my venture fund and
[04:07] about the same time you know uh uh Maurice Chang uh from TI came back to E3 and then uh then set up the TSMC.
[04:17] So it's a very exciting time that I'm being involved in this whole science park and uh synchro science park and the foundation of become the silicon island and so I really see the benefit of you know uh from OEM ODM from design to manufacturing it's all here in Taiwan.
[04:40] Taiwan PC ecosystem has played a critical role uh in Intel growth and success.
[04:47] In fact, last year I was here uh to celebrate Intel 40th anniversary in Taiwan.
[04:58] I want to thank all of the suppliers, partners, and customers for 40 years of partnership with Intel.
[05:05] As partnership continue to grow and uh stronger every year.
[05:08] continue to grow.
[05:13] Thank you.
[05:25] It had been the year since I stepped in uh the role of Intel CEO to be more precise 14 months uh to being CEO of Intel and it may be the first CEO can speak Mandarin
[05:47] and in fact couple of uh customer of Taiwan very important partner for us is that lio also So very unusual if a CEO can drink liquor with us
[06:02] but anyway it's I'm part of this community.
[06:03] Uh execution has always always at the top of my list to do so we
[06:11] had to bring focus back to the core.
[06:15] Uh at our heart Intel is a engineering company and that's what I decided from day one I came to become CEO of Intel.
[06:24] I have all the engineering report to me.
[06:26] So they're understanding to really drive the engineering drives success in engineering uh performant.
[06:34] Our customer and partners are always uh you know already seen a shift uh in the Intel showed up.
[06:41] We are just getting started and so stay tuned.
[06:48] We have a journey in front of us.
[06:50] opportunity ahead is enormous and our job is to stay focused, execute and deliver.
[07:00] Every year Intel ship hundred millions of uh hundreds and millions of uh SOC orchestrating silicon across every
[07:11] industry working tightly with our partners ecosystem across each layer of the stack.
[07:22] from silicon to SOC to system and to software.
[07:27] This generates trillions of dollars in value across four core compute ecosystem.
[07:34] First personal computers,
[07:40] second edge, agentic AI and later physical AI,
[07:44] third foundational data centers.
[07:51] and finally emerging intelligence centers that will power digital agents of the future.
[07:58] Each of these ecosystem represent generational opportunity.
[08:08] and increasingly each of these will need purpose-built CPU,
[08:13] GPU and A6 solutions cers for specific workloads and application.
[08:24] The silicon we are building now will be for human use and the digital agent use.
[08:34] Let us begin with the ecosystem that start it all.
[08:37] To talk more about the PC ecosystem, please join me to welcome to stage our new leaders for client compute and physical AI, Alex.
[09:00] Thank you. Thank you. Thank you.
[09:02] Thank you, Lipu.
[09:04] Little story about me.
[09:06] The first time I came to Taiwan was the year 1990 and I was fresh out of school, first job, first international trip, first
[09:16] East Asia country I have ever been to.
[09:19] And I came right here in Taipei.
[09:22] And I quickly realized even back then
[09:24] that
[09:26] you know this the the desire to grow uh
[09:30] the mindset of win-win and cooperation
[09:33] is all over all of our customers and our partners here in Taiwan.
[09:36] And so it quickly became uh one of my favorite countries to visit and work.
[09:41] And now forward the clock 36 years, I'm here in my first international trip with all the great people here at Intel.
[09:46] And where do I come?
[09:49] Right here to Taipei.
[09:51] Thank you.
[09:57] I can't wait to plan and build the future with you guys.
[10:01] So, let's get started.
[10:04] We have a lot to cover.
[10:06] Intel has continuously increased the pace of progress across all PC segments.
[10:08] workstations, desktops, creators,
[10:19] gamers, premium and mainstream laptops.
[10:23] Every major segment, every major segment
[10:27] is driven by an Intel system solution.
[10:32] With that vast coverage in mind,
[10:36] we're adding another dimension to scale
[10:38] these products even more effectively.
[10:42] The Intel 18A process is now at full
[10:45] scale.
[10:47] We have a full lineup of products with
[10:50] hundreds of design wins.
[10:53] To prove that, at CES,
[10:55] we launched the Core Ultra Series 3,
[10:58] Intel's first product built on 18A
[11:01] process technology.
[11:03] It's setting a new standard for premium
[11:06] mobile performance and battery life.
[11:11] The Ultra Series 3 enables great user
[11:14] experiences across any tasks
[11:17] with a very fast response CPU,
[11:20] a highly improved GPU, low power processing NPU, and the latest multimedia capabilities.
[11:28] It's a perfect blend of IP performance and power for any AI and agentic experience.
[11:37] It's allowing us to lead the way to transform every PC, every PC to an agentic capable platform.
[11:49] Today, more than 300 designs are shipping across consumer and commercial segments, over 300.
[11:59] And to scale these capabilities even further, we've taken the latest core ultra IPs and specifically tailored them for the mainstream market.
[12:13] The result, the Intel Core Series 3 introduced in April.
[12:17] Let me repeat, we just introduced this in April and it's
[12:22] already scaled up to 70 plus designs.
[12:26] That brings Thank you.
[12:29] Thank you.
[12:31] That brings the total series lineup to nearly 400 designs in just a few short months.
[12:37] Now, that is massive scale.
[12:43] Let's look at some of the capabilities of the Core Series 3.
[12:45] And we can start with battery life.
[12:47] You know, I can go over these numbers here that are printed and I can talk about how we measure them and things like that, but I I have a question for you guys.
[12:56] How long is your day?
[12:58] 10 hours, 12 hours, 14 hours, or more like us at Intel.
[13:09] But the great user experience is if your PC lasts longer than your workday.
[13:14] And that's exactly what we're delivering in all segments.
[13:18] And we support ample number of ports for all of your connectivity needs, unlike
[13:25] some of our competitors who only have one USBC interface.
[13:27] But I'll let you be the judge of that one.
[13:37] The goal of this uh uh Core Series 3 is to bring premium feel and experiences to incredibly thin form factors for mainstream PCs.
[13:47] And you know, you don't have to take my word for it.
[13:49] You can look at this wall of incredibly designed, sleek PCs that are here, and all of it is thanks to you, our partners, and our customers.
[13:57] We couldn't have done it without you.
[14:00] So, please, a round of applause.
[14:07] Isn't it awesome?
[14:10] Super light, super thin, really great.
[14:13] Okay.
[14:13] Now, the next proof is scaling 18A IP into growing markets.
[14:19] And the fastest growing portion of the PC market is the handheld gaming.
[14:21] Let's take a look.
[14:38] Hallelujah.
[15:03] Heat.
[15:37] This is the Arc G3.
[15:42] I think a beautiful chip, more beautiful than what was presented yesterday at a keynote.
[15:51] Okay.
[15:55] The G3 is derived from the Core Ultra Series 3 and the ARC G3 is a tuned higher-formance GPU specifically for handheld gaming and it's providing great battery life.
[16:08] The performance tested across multiple games is consistent and stable versus competition.
[16:15] We are more than 40% faster, 40% faster, and at the same performance, we're half the power.
[16:23] And on top of that, we're running all AAA games at 1080p resolution, many of them above 120 frames per second.
[16:33] Now, that is giving gamers a great user experience.
[16:38] All of these devices will be available late later this month, and is just the
[16:43] Beginning.
[16:45] We're going to have plenty more designs coming throughout the year.
[16:51] Thank you.
[16:51] Thank you.
[16:54] Okay, it is indeed true that Intel has a leading lineup of processors and with the versatility of the ATNA process technology, our newest offerings, we're bringing powerful performance and efficiency to scale across the breadth of premium mainstream and handheld gaming segments.
[17:18] These same fundamentals, the same IP, the same capabilities can deliver far beyond the PC ecosystem.
[17:28] The demand for our processors at the edge has been booming.
[17:35] As you've seen, we've already taken existing product lines and pivoted them into adjacent markets, enabling our customers to grow.
[17:45] Their businesses.
[17:48] Now, the edge is demanding the latest products from Intel.
[17:53] And that's why this year, we're taking our latest series 3 products into the edge business with over 130 designs in multiple verticals.
[18:03] For large scale edge business, our customers need the best technology and chipsets, easy to use reference designs, and appropriate software stacks.
[18:15] And at Intel, we've done all of that.
[18:19] We have over 4,000 edge ecosystem partners deploying into such verticals such as manufacturing, robotics, retail, and many more.
[18:30] For those of you here at Computex, you can see some of that at the pavilion.
[18:39] Given that capability, given the IP, given the chipsets, given the scale that we have, there's a massive opportunity.
[18:46] ahead of us across many segments of physical AI.
[18:50] It's projected to be a 25 trillion market by 2050, and it will leverage all of our scale in the PC ecosystem.
[19:02] Physical AI form factors will take shape across key industries, as you see behind me.
[19:10] We will continue growing these markets with the same strategy of leading IP and chipsets, complete reference platforms of enduser hardware and applicable software stacks, enabling our customers to expand into new physical AI form factors and applications.
[19:29] And indeed, this will be our future.
[19:33] Now, back to you, Liipu. Thank you.
[19:42] Thank you. Thank you.
[19:47] Thank you.
[19:47] Thank you.
[19:47] Thank you.
[19:49] Uh thanks, Alex.
[19:49] AI is profoundly impacting the way we use our devices.
[19:55] A major focus area for us is the use of AI on device.
[20:02] Together with partners, we are at the forefront of advancing intelligence.
[20:08] To tell you more about it, let me welcome on stage my close friend and founder CEO of Perplexity, Aravvin.
[20:27] Well, Aravvin, welcome.
[20:27] You and I have been talking about hybrid compute for a while.
[20:34] The reason why are clear you know the privacy cost performance and let's talk about how to make this work.
[20:45] Yeah.
[20:45] So in February we launched
[20:49] perplexity computer. Computer is an AI
[20:52] operating system. It creates a team of
[20:56] agents,
[20:57] uses up to 20 different AI models, and
[21:01] it orchestrates across models, tools,
[21:05] and files in one single system.
[21:08] The agent harness inside computer is
[21:12] model agnostic,
[21:14] perfectly balancing intelligence,
[21:17] accuracy, privacy, and cost is the
[21:21] orchestration problem it solves. And so
[21:24] this allows you to run smaller models
[21:28] locally on the Intel Core Ultra Series 3
[21:32] GPU. And so for the first time ever, we
[21:36] work together to create hybrid aenic
[21:40] inference. And so what we are showing
[21:43] today is just the start. Hybrid agenic
[21:47] inference is how we maximize token value
[21:52] per watt per user.
[21:56] >> So should we show them how it works?
[21:59] >> Yep.
[22:04] >> All right. So here it is.
[22:07] Let's say I'm an associate at a private
[22:10] equity firm and um I'm working on
[22:13] something that has a confidential
[22:15] project code name project falcon. Here's
[22:18] the query.
[22:22] So, think of it as me trying to
[22:25] understand if a certain private company
[22:29] is worth $1.1 billion and I'm feeding it
[22:33] confidential deal materials.
[22:36] The work begins on the laptop. It sees
[22:40] that project falcon has private dealroom
[22:44] files and an NDA, a local leverage
[22:48] buyout financial model, a whiteboard
[22:51] diagram and bilingual transcripts that
[22:54] are very confidential. You don't want
[22:56] these materials to be shipped to the
[22:58] server. So what the local model does on
[23:03] the core ultra series 3 is it first
[23:05] decides this is all very important work
[23:08] and shouldn't be sent to the server. It
[23:10] reads the files classifies what is
[23:12] sensitive and what is not and then
[23:15] computer decides what should leave the
[23:17] device and what shouldn't and each of
[23:20] these things is done with local AI.
[23:25] The orchestrator can spin up additional
[23:27] agents as necessary.
[23:30] And so if you need a research agent to
[23:33] bring in outside file materials against
[23:37] local model without exposing any private
[23:40] files, that's what you want in the
[23:42] hybrid system. And so computer arc acts
[23:45] as one single system, brings all inputs
[23:48] and outputs together. And so let's
[23:50] actually skip and see what the actual
[23:53] result would be.
[23:59] All right. So, the result is a document,
[24:02] a research report, and sporting data.
[24:05] And it's being created by agents on
[24:09] large cloud-based models, keeping your
[24:11] sensitive information only on your
[24:14] device. And so, all your local device
[24:17] models will take care of the private
[24:19] files and the server side models will
[24:21] take care of other things. through
[24:23] hybrid inference orchestration.
[24:25] >> This is the architecture we both believe
[24:27] in and the future is more compute in the
[24:31] data center and more compute on the
[24:34] local machine.
[24:37] >> And so I think of this as a big
[24:40] milestone for engineering on both the
[24:43] agent harness AI side as well as the
[24:46] chip side. And so um it's been really
[24:49] fun to partner with you and Intel on
[24:50] this. So thank you so much Libu.
[24:53] >> Definitely. Thank you Aravvin and
[24:55] looking forward to continue partnership.
[24:57] Thank you.
[25:03] >> We talk a lot about the exciting new
[25:05] developments in the PC edge and physical
[25:09] AI space. I want to take a few minutes
[25:12] to talk about the foundational IP that
[25:16] power all these advancements.
[25:19] Let us talk about x86.
[25:24] When most people think of generalpurpose
[25:26] computing, they think x86
[25:31] and that is a good reason for that. X86
[25:35] is architecture that has power data
[25:39] center for nearly five decades
[25:44] and the leadership continues.
[25:48] uh according to the IDC expect eight out
[25:51] of the 10 servers installed through 2030
[25:57] to be x86based
[26:00] powering modern computing from
[26:02] foundational to emerging intelligent use
[26:06] cases.
[26:09] Intel pioneer most of the breakthrough
[26:12] architectural innovation
[26:15] that have enhanced 886 over the last
[26:19] four decades starting with the 8086
[26:24] that become the foundation of modern
[26:27] computing.
[26:28] Uh if you can see the chart today we
[26:31] have two flagship CPU cores
[26:34] PC and ECores.
[26:37] One optimized for performance,
[26:40] one the others is for efficiency.
[26:46] These are Intel most advanced CPU cores
[26:50] with the accelerator building
[26:53] built in spec uh specifically for
[26:56] foundational workloads like security.
[27:00] Our x86 cores power our PC client edge
[27:05] portfolio and also power our data center
[27:09] and AI portfolio.
[27:11] Under my leadership, we are committed to
[27:15] building the best CPU cores in the world
[27:19] and we will enhance ensure that the most
[27:23] compute intensive workload run best on
[27:27] 886 x86.
[27:30] Next, let us talk. Thank you.
[27:36] Now let us talk about how x86 is
[27:40] enabling foundational data centers.
[27:44] To tell you more about it, let me invite
[27:47] on to the stage Kavoke.
[27:58] Thank you. Thank you. Thank you, Libu.
[28:01] Wow. It's so great to be here. uh
[28:04] specifically in this point in our
[28:07] history, global history, collective
[28:09] history and be at Computex with a blue
[28:13] badge. So I'm very happy, I'm very
[28:14] humbled to be here to share with you
[28:16] some of the innovations that we have. So
[28:19] uh let's talk a bit about uh see what
[28:22] this AI thing is about. So when we say
[28:25] foundational,
[28:26] we mean the workloads that keep the
[28:29] world running. So currently we have data
[28:32] centers and there's a number of items
[28:35] and workloads and entities that run on
[28:37] these data centers. So uh for example we
[28:41] have 5G networks that uh keep us
[28:44] connected. We have databases that keep
[28:47] our data safe. We have cloud services
[28:50] that power our daily lives. So we expect
[28:53] and demand for these workloads to to
[28:56] grow in size and capacity between uh now
[28:59] and 2030 from 80 gawatt to about 100
[29:03] gawatt and uh yeah most of you involved
[29:06] in this uh domain understand the the the
[29:10] extent of this type of an expansion.
[29:13] These workloads are broad. They are
[29:15] mission critical. So attention special
[29:18] attention has to be taken uh when
[29:20] running them but also they require
[29:22] performance, efficiency, security and
[29:26] resiliency. And we can't emphasize
[29:28] enough uh all these four factors.
[29:32] That is why we are excited to have Intel
[29:34] Xeon 6 Plus introduced at Computex this
[29:38] week.
[29:40] It has 288 eores, a massive 576 megabyte
[29:46] of L3 cache
[29:48] built with our Intel 18A technology. And
[29:51] we can't emphasize enough the the value
[29:54] of uh Intel technology that brings to
[29:57] data center products. But most
[29:59] importantly, it delivers
[30:02] efficiency and density which enables our
[30:05] partners to save uh very precious real
[30:08] estate, have more compact servers and
[30:12] the racks.
[30:15] So this is a leadership compute for the
[30:17] next era of cloud and network
[30:20] infrastructure.
[30:24] So Xeon 6 Plus launches with the
[30:26] strength of our ecosystem that's been
[30:28] built over decades and decades of data
[30:30] center development both from a hardware
[30:34] but also from a software and
[30:35] infrastructure perspective.
[30:38] Moreover, our ODM partners are bringing
[30:40] Zeon 6 plus solutions to the market
[30:43] today.
[30:45] So these range from full rack scale
[30:48] deployments to server level designs.
[30:53] Xeon 6 Plus joins
[30:56] our lineup of data center processors
[30:59] next to our already launch Xeon 6 based
[31:02] on peores.
[31:04] Both of these uh category and class of
[31:07] uh solutions delivers new performance
[31:10] and choice for all the enterprises whose
[31:12] infrastructure backbone is built on x86
[31:16] and zeon.
[31:18] This is critical for enterprises that
[31:20] need to increasingly balance preparing
[31:24] for AI workloads but at the same time
[31:27] running their day-to-day mission
[31:28] critical applications.
[31:32] So let's switch gears and talk about how
[31:35] Intel was certifying the deployment of
[31:38] intelligence at scale.
[31:41] It's undeniable that enterprise
[31:42] infrastructure today will have to evolve
[31:45] to keep up with the AI demand.
[31:48] Recent research forecasts that AI
[31:50] inference workloads are expected to
[31:53] become 40% of all data center power
[31:56] demand and much more than uh they are
[31:58] today.
[32:02] So we have these two paradigms where we
[32:04] have the foundational data centers keep
[32:06] on running their traditional workloads
[32:08] but at the same time they have to figure
[32:10] out ways of building their
[32:11] infrastructures to serve intelligence at
[32:14] scale and this is where Intel and Xeon 6
[32:17] plus come in.
[32:25] Now up to now training split the data
[32:28] center into two. So on one hand we have
[32:31] CPUled enterprise infrastructure
[32:35] and the other hand we have GPU heavy AI
[32:37] factories and that was very clear divide
[32:40] for a while right and we've all been
[32:42] accustomed to that uh that reality
[32:45] but as a moves into real workflows data
[32:48] tools governance the needs change. The
[32:52] next wave is not just about training
[32:54] models. It is about putting AI to work.
[32:57] So let's look at why Agentic AI changes
[33:00] the infrastructure equation.
[33:04] The way AI inference works is
[33:06] straightforward. We take a prompt. It
[33:09] gets fed into an LLM where it spends
[33:11] most time reasoning about the prompt. I
[33:13] we've all seen this. We've done this
[33:14] thousands of times. And out comes an
[33:17] answer.
[33:18] In this case, a lot of time is spent
[33:20] computing the large language model which
[33:22] is mostly GPU and compute intensive.
[33:29] Now the way agentic AI works is
[33:30] radically different. It's given goals
[33:34] rather than prompts. So we all seen the
[33:37] uh the different types of loop that
[33:39] people are running on this agentic AI.
[33:41] It's also very iterative in nature but
[33:43] also prompted by automation
[33:46] and thinking, planning, acting and
[33:48] reflecting are a natural way of these
[33:51] agents interacting with us.
[33:54] As it works, it uses tools, reads and
[33:57] writes files, checks rules and other
[34:00] aspect that were, you know, in the
[34:02] traditional realm of CPUs and x86.
[34:06] And then for each step, the type of
[34:08] underlying
[34:10] compute needs is very different. And
[34:12] we'll show that in a bit.
[34:15] This is particularly important as agents
[34:17] scale up their work. Spawning new agents
[34:20] that work concurrently. And the category
[34:23] and the complexity of agents are going
[34:25] to be very different depending on the
[34:27] complexity of the work.
[34:30] That's the main reason that there's such
[34:32] a rapid increase in CPU demand for
[34:35] Aentic AI. The CPU orchestrates the
[34:38] show.
[34:40] Now, what we're seeing is we're also
[34:42] seeing the balance and the ratio of uh
[34:45] one CPU to 8GPU and more is uh is coming
[34:48] much closer to par. So, let's take a
[34:51] look at the real example. John,
[34:54] >> thanks. Gaborg, you talked about how
[34:56] Agentic AI is changing the compute
[34:58] requirements. Let's take a look at a
[34:59] real example. I have a traditional AI
[35:02] inference set up on the lefth hand side
[35:03] of the screen. Let's send a request.
[35:06] Write a Python function that calls an
[35:08] OpenAI compatible chat completions API.
[35:11] The model gets the response, generates
[35:14] code, and sends the request back. Take a
[35:17] look at the slider on the top of the
[35:18] screen. GPU dominates nearly 7:1 GPU
[35:22] heavy.
[35:24] In contrast, let's take a look at an
[35:26] Aentic AI system. Across the top, look
[35:29] at the pipeline stages. Green is GPU
[35:32] work. Blue is CPU work. Linting is
[35:36] happening on our Xeon 6 Plus processor
[35:38] with effic efficiency cores. Web fetch
[35:41] and compile is happening on our Xeon 6
[35:44] performance cores
[35:46] and unit testing is coming back and
[35:48] running on our Xeon 6 plus effic
[35:50] efficiency cores.
[35:52] The right class CPU for each stage of
[35:55] the pipeline. Take a look at the slider
[35:57] across the top again. We're near par,
[35:59] but CPU heavy this time.
[36:03] What's this look like when we multiply
[36:05] that by millions of queries a day?
[36:10] As you mentioned, each Xeon 6 Plus
[36:12] processor has up to 288 cores. That's
[36:16] 576 cores per two socket server.
[36:20] When we look at that from a rack scale
[36:22] perspective,
[36:24] that gives us over 36,000 cores per 32
[36:27] years of compute space.
[36:30] Thank you, John. Wow, this is pretty
[36:32] amazing and some data to ponder on.
[36:39] By far the density of CPU we showed is
[36:44] is has the highest density per rack
[36:46] ever. But also looking at the number of
[36:49] agents and these are the new metrics
[36:52] that are emerging. We can safely say
[36:54] that that particular rack can run up to
[36:57] 150,000 agents. So good news to all the
[37:01] CIOS in the audience. Now your very
[37:03] expensive GPUs can be can see more
[37:06] utilization because of uh our solutions.
[37:12] Now both Zeon 6 with pores and ecores
[37:16] are built on for intelligence at scale.
[37:19] There are different cores of course but
[37:21] we've seen the workloads that require
[37:23] very high performance cores pushing the
[37:26] frequencies but also there's a need for
[37:28] very high density power efficient cores.
[37:31] So we've seen all the workloads we've
[37:33] run all the analysis and we are
[37:35] delivering these solutions uh to to all
[37:37] of you
[37:40] now.
[37:42] Having said that, we are working with
[37:44] our customers and partners to make sure
[37:46] that each solution is uh tailored to to
[37:50] your needs. So,
[37:52] I'd like to welcome Libu back on stage
[37:55] to talk about the server and rack sale
[37:57] solutions that our partners are working
[37:59] on. Thank you.
[38:05] >> Thank you, my friend.
[38:06] >> Thank you. Thank you.
[38:11] Uh thank you Kavoke. It is great to see
[38:15] the momentum uh in the data center.
[38:19] As we look forward to see that is for
[38:23] intelligence at scale. Discrete compute
[38:27] alone is not enough. Our customer are
[38:30] asking us to think of system level to
[38:34] help them serve real agentic workloads
[38:38] at scale. It push us to rethink how we
[38:42] deliver our compute beyond the socket
[38:46] and to the rack.
[38:49] That is why we start the initiative
[38:52] called Rex scale blueprints
[38:56] working with ecosystem partners to
[38:58] develop Rex scale blueprints built on
[39:03] open standards.
[39:05] So customer can rapidly scale their
[39:08] intelligent infrastructure with
[39:10] confident without proprietary
[39:13] workarounds.
[39:16] Behind me as you can see two examples of
[39:19] these blueprints. One is for agentic
[39:23] performance based on Intel Xeon 6 with
[39:27] pores.
[39:29] The other is agent density with the
[39:33] Intel Xeon 6 Plus with ECores.
[39:37] We are working closely with our partners
[39:40] ecosystem including Foxcomanova
[39:45] to expand our Rex scale offering. Let me
[39:49] call on stage one of our partners, chief
[39:52] product officer of Foxcom, Jerry Xiao to
[39:57] talk about how we partners on Rick scale
[40:00] solution.
[40:04] >> Thank you Libu.
[40:06] Thank you, Jerry.
[40:09] >> Um, I'm so excited to be here today.
[40:12] Wonderful product and amazing event.
[40:18] Jerry, Intel and Foxcom have been
[40:20] working together to many decades and
[40:24] Foxcom has been instrumental in driving
[40:28] technology innovation in Taiwan and
[40:30] around the world.
[40:33] >> Yeah, that's right, Abu. Um I'm proud
[40:37] the work we have done together from AI
[40:40] servers to data centers
[40:43] and to age computing all together and
[40:47] today we're excited to announce the next
[40:51] step in our partnership.
[40:53] Intel andcom are working together to
[40:56] develop Rex scale products built upon
[40:59] Intel Xeon processors. Together we will
[41:03] focus on exploring the development
[41:07] integrations and commercialization
[41:10] of differentiated
[41:12] uh rack scale AI infrastructure solution
[41:17] leveraging comp complementaryary
[41:19] architecture to address diverse AI
[41:22] workload requirements.
[41:25] Yeah, together we will continue to
[41:27] deepen and expand our partnership
[41:30] unlocking new opportunities ahead.
[41:32] Through this collaboration, we will
[41:35] deliver system level AI solution to our
[41:38] joint customers in airballing more
[41:41] integrated and scalable computing
[41:44] environments. This makes it marks an
[41:48] important step ahead and we look forward
[41:52] unveiling more in the near future. Thank
[41:54] you Liu.
[41:56] >> Today is an exciting uh milestone for
[42:00] our continual partnership uh with
[42:03] Foxcom. Jerry, thank you for joining us.
[42:06] >> Fantastic. Thank you for having me.
[42:16] Thank you to the many partners in the
[42:19] audience today that is helping to bring
[42:23] this rack scale vision to life providing
[42:26] choice through our uh through the
[42:29] ecosystem power. Uh we do not believe
[42:32] one size fit all approach for
[42:35] intelligent centers. Each enterprise
[42:39] will run unique workloads. So their
[42:43] infrastructure needs will also need to
[42:46] be unique and purpose-built. As you can
[42:49] see from the screen here,
[42:52] just look at the server in front of me.
[42:55] Uh this is whole series of partnership
[42:57] we have.
[43:14] Intel is working with a lot of partners
[43:17] to provide service rack scale solution
[43:21] designed to fit your existing
[43:24] infrastructure ready for AI at scale as
[43:28] you can tell in front in front of you.
[43:32] We see token usage exploding.
[43:36] Agent now consume 1,000 time more tokens
[43:41] than single event reasoning.
[43:44] In addition to building the best CPUs,
[43:48] it is critical that we deliver compute
[43:51] solutions optimized for token
[43:54] consumption and token generation.
[43:59] The bottom line, AI at scale will
[44:02] require hetogeneous computing. To this
[44:05] end, Intel recently announced a
[44:08] partnership with Sonernova.
[44:10] To talk more about this, let me call to
[44:13] the stage founder CEO of Samberonova,
[44:16] Rodrigo Leong.
[44:25] Rodrigo, welcome to joining me today.
[44:28] >> Thank you, Lupu.
[44:30] >> And over the next few months, we have
[44:32] announced few updates on our joint
[44:36] development partnership. Can we talk a
[44:38] little bit more about the work that
[44:40] Intel and Sonova are doing together?
[44:43] >> Absolutely. We've been busy. Earlier
[44:46] this year, we announced a multi-year
[44:49] collaboration to deliver high
[44:51] performance, costefficient AI inference
[44:54] solutions based on Xeon infrastructure.
[44:57] We've been building something really
[44:59] special. Excited to show you today.
[45:04] This is the This is the SM50
[45:08] sambar we announced earlier this year.
[45:10] Rack scale AI infrastructure built for
[45:13] agentic workloads.
[45:19] It uses Intel Xeon 6 processors with
[45:22] Sonova SM50 RDUs and shipping to
[45:26] customers later this year. Today, we're
[45:29] also excited to demonstrate the world's
[45:32] first heterogeneous disagregated
[45:35] inference using Simonova's RDU with
[45:38] Intel's CPU and Nvidia GPUs.
[45:42] What you're about to see is the same
[45:43] prompt, the same model running side by
[45:47] side, two different stacks.
[45:54] So the one on the left is GPUs, RDUs,
[45:59] and CPUs. This aggregated difference and
[46:02] this one on my right is GPUs on their
[46:06] own. They both get the fed the same
[46:09] prompt in the same model just different
[46:11] stacks.
[46:13] The disagregated inference stack is
[46:15] taking off. And what's happening here is
[46:18] you have the Xeon 6 processors doing all
[46:22] the tooling execution. You have
[46:24] Salmonova RDUs doing the decode and
[46:27] generating all of the tokens. And then
[46:29] you got the GPUs performing the PROM
[46:31] caching and the faster prefill reducing
[46:35] overall time.
[46:37] When all three chips are working
[46:39] together, you dramatically reduce the
[46:42] end toend latency and the agents for the
[46:45] fastest for enentic AI and the other
[46:49] side the GPU stack is still working
[46:52] away.
[46:54] So the initial result of our work is
[46:57] disagregated inference is the the GPUs,
[47:00] the RDUs, the CPUs that's the fastest
[47:04] and artificial analysis and our test
[47:06] found it to be two to three times faster
[47:09] than just the GPUs alone. And this gives
[47:12] us an early look at how fast this can
[47:14] be. Ro, the most exciting part about all
[47:18] this is that we have tremendous customer
[47:21] interest uh in these solutions.
[47:24] >> Absolutely. So, let's see who comes
[47:25] next. So, turn it back over to you.
[47:28] >> Thank you so much.
[47:28] >> Thank you. Thank you. Thank you.
[47:34] To continue this conversation, I'm
[47:37] delighted to invite my good friends uh
[47:40] you know, Robert Smith uh is a Vista
[47:44] Equity Partners, chairman, CEO and and
[47:49] uh you know of the partners and Roger
[47:52] Smith to tell you more about how they
[47:54] plan to use this racks from Intel and
[47:57] Samnova. Robert.
[48:01] >> Hey, thank you.
[48:05] >> Lipu, good to see you.
[48:06] >> Same here. Thank you so much for joining
[48:08] me.
[48:08] >> Pleasure. Thank you, my friend. Thank
[48:09] you, my friend.
[48:10] >> Yes. So, Robert, AI is driving huge
[48:15] demand uh for computing and it is
[48:18] reshaping the silicon system software
[48:21] and at all at once. What are you seeing
[48:24] and hearing from the enterprises that
[48:27] you work with? Yeah, first of all, I'm
[48:29] excited to be here at Computex to join
[48:30] you at this wonderful event. Um, for us,
[48:34] it's been quite quite incredible. Uh,
[48:37] there's been a huge focus right now to
[48:38] bring AI to enterprises around the
[48:41] globe. Uh, we want to make it usable. We
[48:44] want to make it impactful for the
[48:45] organizations that we work with. You
[48:47] know, we have over 90 portfolio
[48:48] companies and well over half of them
[48:50] have now uh have converted to Agentic
[48:53] Solutions. And with over 750 million
[48:56] users of our software, that really
[48:58] translates to over 10 billion agents.
[49:02] That's why we've launched Vector Core
[49:04] Computer VC2 with our partners at
[49:06] Cambium Capital to offer the world's
[49:09] first commercially available
[49:11] architecture for disagregated inference.
[49:15] This novel agentic neocloud is built to
[49:17] deliver the fastest enterprise inference
[49:19] throughput of any architecture to date.
[49:22] The demo you just witnessed with Rodrigo
[49:24] was conducted live in our Los Angeles
[49:28] data center and we have over 50
[49:30] deployments planned in the US which are
[49:32] targeted to convert existing data
[49:34] centers to inference data centers. This
[49:37] is very exciting and as we saw from Ro a
[49:41] few minutes ago uh we are already
[49:44] starting to see strong momentum for
[49:46] these offerings. Can you talk a little
[49:48] bit more about how Intel Victor core
[49:52] compute and our partners like SANOVA are
[49:55] bringing this aggregated uh inference to
[49:58] life?
[49:58] >> Of course uh I'm excited to share that
[50:00] first together AI is the first
[50:02] commercial customer uh and is excited to
[50:05] use this architecture as a service to
[50:07] accelerate inference workloads. We
[50:10] expect many of our enterprise software
[50:12] companies and their customers to quickly
[50:14] follow as the demands for inference keep
[50:17] growing and this is has to be and it is
[50:19] more efficient than anything they pre
[50:21] previously have had access to. Most
[50:24] critically VC2 is built and utilizes the
[50:29] senova stack which is an aircooled data
[50:31] center. We believe it will deliver what
[50:34] enterprise customers and communities are
[50:37] asking for which is reliable, low
[50:40] latency, lowcost inference at scale.
[50:44] Partnering to advance AI is one of the
[50:46] best ways to develop this
[50:47] transformational technology making it
[50:50] usable and economically viable for
[50:53] enterprises worldwide.
[50:56] >> Uh we are excited about that. Thank you
[50:58] for joining me today and delighted to
[51:01] have you here.
[51:02] >> Always a pleasure, Liu. Thank you.
[51:03] Congratulations. 14 more months. We're
[51:06] excited to see what you're doing.
[51:07] >> Thank you.
[51:08] >> Thank you.
[51:11] >> As you just saw from Ro and Robert,
[51:14] picking the right silicon architecture
[51:16] for your needs is critical for
[51:20] enterprises today. There's a broad range
[51:22] of architectures to choose from. As
[51:26] large workloads increasingly become
[51:29] strategic asset for companies, they are
[51:33] increasingly looking for silicon built
[51:35] around their exact needs.
[51:39] Next, I would like to invite Sini, a
[51:42] semiconductor design veteran and a
[51:45] leader of our purpose-built silicon team
[51:48] to talk more about the work we are doing
[51:51] in this area. Swini
[52:01] Hi Leu, thank you. Thank you so much. A
[52:04] very good afternoon to you guys.
[52:06] Purpose-built silicon it is. This has
[52:09] been a journey that the industry
[52:11] industry has been using almost for the
[52:13] last decade or so and especially
[52:15] hyperscalers have tapped into this to
[52:17] its full potential and shown us the
[52:20] benefits in every way possible. Libu,
[52:22] last year you challenged us to see in
[52:25] this space given the fantastic assets
[52:28] and the breadth of assets that we have
[52:29] at Intel, how could we be relevant to
[52:32] this, not just be focused on the stuff
[52:35] that we do internally. How do we bring
[52:36] this out to the external world and do
[52:38] something more? With that said, we we
[52:42] had a proposition. We've been working on
[52:44] it. And today I'm very happy to share a
[52:47] couple of good outcomes that we have.
[52:49] The first on the hyperscaler side we
[52:52] have Google and Intel or Google and
[52:55] Intel have gone into a partnership
[52:57] wherein Intel is delivering what is
[52:59] called as the infrastructure processing
[53:01] unit. I would call it Intel processing
[53:03] unit actually but infrastructure
[53:05] processing unit which is a a piece of
[53:07] silicon very vital for hyperscalers
[53:10] performance and that journey continues
[53:12] and by the way this is a deployment
[53:14] today so it is not just something that
[53:16] we are doing but it's already designed
[53:18] and being deployed while this is this is
[53:21] working on we've been Intel as a company
[53:24] has been pretty active in the telco
[53:26] market and in this telco market another
[53:29] marquee customer Ericson has been uh
[53:32] partnering with us and Ericson chooses
[53:34] us wherein we deliver or Intel delivers
[53:37] the next generation infrastructure
[53:39] silicon for at a global scale for them
[53:42] across the board. This just gives you a
[53:44] very sneak preview at the highest level
[53:48] to see the kind of work that's happening
[53:49] in the purpose-built silicon space which
[53:51] is a very exciting space and more
[53:54] importantly a high growth space and I
[53:57] was just thinking what better place than
[54:01] Computex and Taipei where custom silicon
[54:04] really is the name of the game here to
[54:07] announce that Intel has officially
[54:08] entered this market. So looking forward
[54:10] to working with many of you guys and see
[54:13] how we can be relevant to you some of
[54:15] your aspirational goals on silicon.
[54:17] Okay. Thank you Libu.
[54:19] >> I'm super excited about all this
[54:20] partnership that you announced and more
[54:22] to come.
[54:23] >> Yes, absolutely. More to come. Yes.
[54:24] Thank you so much. Thank you so much.
[54:26] Thank you.
[54:26] >> Thank you.
[54:30] >> The work Streiny and the team are doing
[54:32] with purpose build silicon is really
[54:35] important. I'm super excited to be
[54:38] partnering to build custom silicon with
[54:41] many leading edge companies as well as
[54:44] some of the most dynamic startups across
[54:47] the industry vertical. I would like to
[54:51] highlight some of this partnership
[54:53] today.
[54:55] One of the most exciting areas where we
[54:58] can deploy advanced silicon is biomemed
[55:02] engineering. For years, emulating the
[55:06] function functionality of the human
[55:09] brain has been the holy grail of
[55:11] computing.
[55:13] One company that is in the forefront of
[55:16] brain inspire computing is echo neuro
[55:22] technologies.
[55:24] Let us hear more from Eddie Chan,
[55:27] founder CEO of Echo Neuro Technology and
[55:32] also one of the world best
[55:35] neurosurgeons.
[55:37] Hi, I'm Eddie Chang. I'm a neurosurgeon
[55:39] at UCSF and co-founder of Echo
[55:41] Neurochnologies.
[55:43] For decades, AI has been brain inspired,
[55:47] meaning borrowing ideas from
[55:49] neuroscience at a distance. Neuromorphic
[55:52] computing has carried that vision the
[55:54] furthest. It built silicon around the
[55:56] brain's core principles like spikes,
[55:58] sparse communication, memory, and
[56:00] compute all in the same place. That
[56:02] architecture is right. But what's been
[56:05] missing is direct evidence of how the
[56:07] brain actually performs the computation.
[56:11] That's now within our reach. For the
[56:13] first time, we can study how the human
[56:15] cortex computes language in real time at
[56:19] the resolution where computation
[56:21] actually happens. This opens a whole new
[56:24] possibility.
[56:25] Algorithms that are not just brain
[56:28] inspired, but new ones that are trained
[56:31] on the brain activity itself, measured
[56:33] against the brain itself. That's the
[56:36] shift in our collaboration with Intel.
[56:39] Together, we're developing brain trained
[56:41] algorithms for streaming speech that
[56:43] approach the efficiency of biological
[56:46] computation.
[56:47] The payoff runs both ways. AI that's
[56:50] faster, lighter, and closer to how we
[56:53] actually think, and new tools to restore
[56:56] speech to people who have lost it.
[56:59] Together with Intel, we're building AI
[57:01] that learns from the most powerful
[57:03] computer ever discovered, the human
[57:05] brain. We're excited about what's ahead.
[57:08] Thank you.
[57:10] >> Thank you, Eddie. I'm amazed by the work
[57:13] you are doing. I'm confident that our
[57:16] work together will help lay the
[57:18] foundation of highly efficient AI
[57:22] computers uh in the future.
[57:26] Another company doing work at the
[57:29] cutting edge of biology is Greenstone
[57:33] Technologies.
[57:35] We are partnering with Greenstone to
[57:37] establish scalable reference
[57:40] architectures
[57:41] applicable across the broader life
[57:45] imaging ecosystem.
[57:48] Dr. Joseph Wu is the head of cardiology
[57:50] at Stanford and the founder CEO of
[57:54] Greenstone. Let's hear from him.
[57:58] Hello, my name is Joseph Wu and I'm a
[58:00] professor of medicine and director of
[58:02] the Stanford Cardiovascular Institute as
[58:04] well as the co-founder of Greenstone
[58:06] Biosciences. Thank you so much for
[58:08] including me in Computex. Intel and
[58:11] Greenstone are working together to speed
[58:14] up the development of new medicines. Our
[58:16] partnership combines state-of-the-art
[58:18] human genetics and biology from
[58:20] Greenstone with advanced AI computing
[58:23] for Intel so that we can scale data
[58:26] processing, storage, and analysis.
[58:29] Greenstone has built the world's largest
[58:31] bio bank of human induced flu potent
[58:33] stem cells. From just 10 cc's of your
[58:36] blood, we can make your brain, heart,
[58:39] liver, kidney, gut, and any type of
[58:42] organoids in your body that are
[58:44] genetically identical to the patient.
[58:46] This will then allow us to test existing
[58:49] and new medications more quickly and at
[58:52] a lower cost. I believe the combination
[58:55] of human biology and AI computing will
[58:59] help shape the future of biio medicine
[59:01] in the next decade. And this is why
[59:03] we're so excited about the partnership
[59:06] between Intel and Greenstone
[59:07] Biosciences. Thank you very much and
[59:10] enjoy the event. Shaja,
[59:17] thank you Joe. I'm amazed by the work
[59:20] you're doing and my excited about the
[59:24] potential of our partnership.
[59:28] Another key partners is Hitachi.
[59:32] They have a wide range of capabilities
[59:35] that help accelerating our our work our
[59:40] plan around foundry tools and quantum
[59:43] computing system. Let us hear from
[59:46] Hitachi CEO Toko Nagasan.
[59:51] >> Hello Computex. I'm Toshiakunaga, CEO of
[59:55] Hitachi. For decades, Hitachi and Intel
[59:59] have worked together to solve key
[01:00:02] challenges for society. And today, we
[01:00:06] are bringing our strength even closer.
[01:00:09] By combining Intel's advanced computing
[01:00:12] with industrial strength in the physical
[01:00:16] world, we will create intelligent
[01:00:20] solutions that will benefit both
[01:00:22] businesses and society.
[01:00:25] Thank you, Libu. I look forward to our
[01:00:28] future together.
[01:00:31] >> Thank you, Titi.
[01:00:36] We are really looking forward to working
[01:00:38] with you.
[01:00:41] Finally
[01:00:43] uh if you look at you know we have the
[01:00:44] brains inspire computing biomed medicine
[01:00:49] and then energy. The last one is
[01:00:52] industrial automation. Finally one
[01:00:55] partner I would like to hear like you to
[01:00:58] hear from is known for their pioneering
[01:01:01] work in industrial automation. Let us
[01:01:05] hear from my very good friend uh Rhoden
[01:01:07] Bush at Seammens.
[01:01:10] >> Hi Libu. As a customer of Intel, we all
[01:01:14] know that global semiconductor demand
[01:01:17] has hit a high record. In 2023, Semens
[01:01:21] and Intel already joined forces to meet
[01:01:24] it. And now we are taking our
[01:01:27] collaboration to the next level. We are
[01:01:30] expanding our partnership across the
[01:01:32] entire value chain from design to
[01:01:35] manufacturing to chip applications in
[01:01:38] seammen's products.
[01:01:41] We improve design quality through EDA
[01:01:44] automation and software solutions built
[01:01:47] with Atlantic AI. We partner on all
[01:01:51] areas of the manufacturing process
[01:01:53] including product life cycle management,
[01:01:56] automation, electrification, quality and
[01:01:59] sustainability.
[01:02:01] And what makes this even more relevant
[01:02:04] for us, the chips created in this value
[01:02:07] chain will be used in our own seammen's
[01:02:10] products.
[01:02:12] Looking forward to what's coming up.
[01:02:16] >> Thank you, Roland.
[01:02:22] We are delighted to expand our long
[01:02:24] partnership with the seaman groups.
[01:02:28] I'm looking forward to disclose more
[01:02:32] about this partnership in the coming
[01:02:35] months and we are working with several
[01:02:37] other partners to keep pushing the
[01:02:40] boundary of what is possible.
[01:02:46] I would like to close by returning to
[01:02:48] where we start our conversation.
[01:02:52] The opportunity for Intel and for our
[01:02:55] partners is immersed PC edge agentic
[01:03:01] physical AI data center and emerging
[01:03:05] intelligence center from silicon to SOC
[01:03:10] to system and applications.
[01:03:15] This opportunity is only made possible
[01:03:19] by all of you. Look at the list and the
[01:03:22] largest ecosystem of partners, suppliers
[01:03:27] and customer.
[01:03:31] Intel is an iconic company. We lay the
[01:03:35] foundation of modern-day computing and
[01:03:39] we are proud of our heritage.
[01:03:42] But we do not want to rest on our honors
[01:03:46] and gora. A year ago, I joined as a CEO.
[01:03:52] I challenged my team to work with me to
[01:03:55] build a new Intel.
[01:03:58] That is exactly what we are doing. We
[01:04:01] are not encumbent by the past. We are
[01:04:04] building something wonderful.
[01:04:08] It is have been year of transformation
[01:04:12] for Intel. We ram our 18A to high volume
[01:04:17] with multiple products. We are executing
[01:04:20] well on our advanced packaging
[01:04:23] milestones. We make tremendous progress
[01:04:26] on engaging customers and building our
[01:04:29] foundry business. We introduced new
[01:04:33] SOC's for all major compute platforms
[01:04:37] from premium mobile to high density
[01:04:41] cloud and 5G.
[01:04:45] We are be rebuilding and strengthening
[01:04:48] partnership across the ecosystem.
[01:04:52] where I double down on creating new
[01:04:55] business opportunity
[01:04:57] across existing and emerging domains.
[01:05:03] We are working at the forefront to
[01:05:05] imagine re-imagine computing and make it
[01:05:09] highly efficient for the AI era.
[01:05:13] And this is just the beginning. I super
[01:05:17] excited to continue executing at hypers
[01:05:22] speed.
[01:05:24] >> Before the lights go out,
[01:05:27] the race begins.
[01:05:32] From simulation to strategy, performance
[01:05:35] begins with compute. With electrons that
[01:05:38] power pace and data that backs
[01:05:40] decisions,
[01:05:42] the race never ends. Engineering never
[01:05:45] stops.
[01:05:46] Intel, official compute partner of
[01:05:49] McLaren Racing.
[01:05:57] Ladies and gentlemen, this concludes the
[01:06:00] Intel Computex 2026 keynote. Thank you
[01:06:03] for joining us this afternoon to witness
[01:06:06] the future of technology. We look
[01:06:08] forward to seeing you again at Future
[01:06:11] Intel events. And please don't forget
[01:06:13] your personal belongings.