Alastair Richardson AMD Episode 48

0:05 Alister, Alister Richardson of AMD, I am so pleased to welcome you to the showbiz show that is David Anderson talking market data. So, I I hope we're going to have a good chat about market data. I I'm just I was thinking how long have we known each other? Probably Austin, Texas and a couple of times a year or so, 2 years. But I'm 3, 4 years now. >> Yeah, 3, 4 years. It's been a whirlwind and nice to to see you develop into this new role as well doing these podcasts, which I think is quite good.

0:37 >> did to be it's a bit of fun. You probably noticed I do like the sound of my own voice. So, this this was a this is a good opportunity to to to to do to do that. So, look Alister, let's sort of dive into the chat. But again, beforehand, what I do with everybody, I'm going to put your LinkedIn profile into the descriptions. So, everybody will know sort of where you came from. But it's still worth giving a little bit of an intro as to sort of who you are and where you came from.

1:06 because I mean, you've got a pretty decent pedigree background in the sort of capital markets before you jumped into this wonderful world of hardware, which we'll talk about in due course. But what I mean, how did you end up here, Alister? Sure, I'll give you a a a a brief history of time. so, I mean, I started out off my career really at the in the banking industry working at places like Credit Suisse, JP Morgan, and Barclays Capital, including getting my European trading licenses and and trading in an agency and and Delta One capacity.

1:39 So, very interesting time of the the world really. this was pre-crash and and through the crash back in 2008 and and that era. but what I saw was really the growth of algorithms from their infancy. A lot of them in my first days were still being done in Excel and then they got moving into a far cry from from where we are today really with high performance platforms and then FPGAs and and all kinds of different technologies. So I really saw an opportunity that more and more was moving towards technologist role and I went and did a number of startups helping that progress towards electronification of our market. And that's just continued more and more and expanded into outside of equities and futures which were kind of early adopters into more and more markets. So you you consciously left the sort of actual participation in the capital markets to go into the technology space because you you saw the opportunity.

2:48 I wouldn't say I left it. I've always been part of it. I've always been back and working with the industry to develop new solutions. But yes, I I made a choice to to go into the technology side and leverage a lot of that experience and knowledge from the actual trading and capital market side and then see how we can improve the technology from a fundamental perspective to give people the the best options and and democratization of technology. Well, look, I I sent you my little my three little ideas of the how the Well, we'll take the conversation wherever it goes.

3:27 But just so the audience can know what I'm going to bring up. First of all, I'm going to kick off by asking you, Junaid, what what are you seeing in the market data industry? You've been with us as Ed. Excuse me. For the last four years or so, again, you've been watching a lot of the So I've got a frog in my throat. You've got you've been watching at the FISD events that we've been involved in, general discussions about market data.

3:49 So I want your sort sort thoughts on that, your observations. Second part, which we just alluded to, you know, what part does technology play in all this? You know, is it just about speed or is there a bit more to it there? And then the third and final part is specifically hardware, which is a sub element of technology overall, and frankly a newer topic out there. And I I'd love you to help educate me, but educate the audience to sort of where that all fits in. So, let let's start with that first one. I mean, you you know, you've been part of the market data industry. You've been watching us all, you know, at various events and what have you. What what What do you see? What what do you think? What's the observation? I I won't call you an outsider, but you're a newer a slightly newer flavor coming to our world, which occasionally we do get stuck in our ways. I mean, again, what what are you seeing?

4:42 Sure. so, I mean, I think, you know, if I if I look over the last sort of 20 years, I mean, we've always needed market data in in every different part of trading for for me forever, right? market data is a kind of cornerstone. so, you know, first thing as I mentioned earlier, we do obviously see electronification of market data and the ability to use that in in different ways. And I think we went through a wave where you had the the large firms doing their general market data platforms from a technology and upgrading with Electron and Be Bop and and others at the higher end, but we also saw a lot of innovation in the industry. And I would say probably 10 to 15 years ago where we had companies you know, like X G, like Anex, like Nova Sparks. All All one now.

5:39 All one now. And I would that's actually one of the points I was going to say. There's there's been a lot of consolidation. but you had a lot of these entrants back in the day and even Vela and Wombat and all of these names that used to be and then there's a you know a ton of consolidation. and I think, you know, consolidations are can be a good thing for the industry and you know, there's some great bits of technology across all of these different firms that that had their innovation in the their days and it's really exciting to see that. But what's also exciting is that there's a whole range of new startups coming out as well. which are relatively younger players in the market who are now taking a slightly different approach and and I think, you know, as I mentioned there was the the larger data providers in the past who have their entrenched systems and which are very well known. You had these FPGA and these ultra low latency participants and now you're finding these new players who I think are taking advantage of hardware and well, we can go to that later on in the conversation, but looking at more commoditized hardware and how to really bridge that gap between the classic players, I would say, and the ultra low latency players and and democratize that a bit more.

7:02 Well, and before we and I do want to go into detail on those, but again, a general observation I've had cuz I've watched this over the my 41 years in this industry. I remember the whole low latency debate. God, that must be 20 plus years ago we've been talking about sort of low latency. But one observation I would make is that was only a small rarefied part of the market. We we loved talking about it cuz it was quite cool to talk about it. And I think some of the things you and I going to talk about quite cool.

7:33 But I would sort of put it to you, they're only used by very small part of the market. A lot of the market is rather boring and rather doesn't particularly seek ultra-low latency, high frequency, or whatever. So, you know, how do you how do you see the market segmented? I mean, are you only targeting that rarefied upper echelon of the market? What What about the whole swathe of the rather high latency or medium latency or the people who don't really care about speed or frequency? I mean, what Where How do you How do you see the market segmented? Sure. So, I mean, I think first of all, you're selling yourself short. Market data is exciting, no matter where you are. Oh, we'd go there.

8:15 It's even sexy, we've tried to say, you know. Maybe that's pushing it a bit far. But, yeah, but but no, I I 100% agree. So, I mean, there's a lot of different latency profiles depending on what you're looking to do. and, you know, you're right, there's the ultra-low latency. and if you really want to go to the absolute cutting edge on on that side, you know, there's a very very small percentage of the market that that is looking at that, right? And that's in the 1 to 2 3 nanosecond kind of space.

8:46 And who From your experience, is that sort of high frequency traders, hedge funds, trading shops? Who Who do you notice operates in that space mostly? What type of profile? Yeah, I mean, clearly, it's that's kind of high frequency traders that are the ones who have the capabilities to develop that themselves. In order to get the the absolute lowest, you need to customize it completely for your own specific use case and algorithm, right? Yeah. and there's there's value in doing that for for the firms that are playing in that space. But, there's there's very little benefit to that for somebody who is you know, at a much higher latency or to the extreme, a click trader trading market data on your mobile phone, right? that's a the You need to invest in in that same capabilities or space. But saying that, latency has a number of different characteristics, right? And then coming back to your point in how the market is segregated, latency is very beneficial for for arbitrage or applications like that where you need the raw latency. But latency has a byproduct, which means if you can do something faster, you can do more things in the same amount of time.

10:08 Okay, yeah. So I suppose that's the frequency angle. Well, potentially, but I mean effectively when you're thinking about it, if you're at one extreme and I mentioned mobile phones, if you've got a processor that can perform faster and a more efficient algorithm, you can serve more customers on a single server than before. So while latency might not be the core function, implementing it in your environment can reduce your digital footprint and reduce your server rental costs.

10:40 And understanding that to improve your overall application performance means you can support more end customers on the same hardware and equipment. And and and that was sort of segueing into the the second point I made about technology, is it just about speed? And I think what you're saying is not just I mean speed is one factor, but this quantity that that the frequency quantity, whatever this other dimension, is as important or is important as well. Yeah, 100% and and we see this again, I would say this is a going back to the trends.

11:17 10-15 years ago, I don't think anybody thought about computing power in the same way. I remember when I was you know, working in the banks, we just said go and buy more servers. I think we're reaching a point now where they're starting to be pushed back and and challenged because the cost of doing so and the space the data center space and the power requirements are starting to run out especially in the world of AI where power is starting to become a limiting factor in many of the biggest areas. I mean before I met you the sort of big theme that I've been talking about for 10 maybe longer 15 I've lost track of time when I first started was cloud. I mean before we we're all talking about AI now. AI is all we seem to talk about.

12:07 But for a little while first of all there was the low late I remember low latency was a hot debate at at conferences and whatever panel discussions ultra low latency FPGA and all that stuff. Then at some point and I can't remember the exact date we started talking about cloud what is cloud? It was in those early days it was nobody understood what the cloud was and then over time that's matured and now we're talking about AI.

12:32 But if you guys in the world of hardware I suppose you've been on that journey with us as well. I suppose there was a hardware dimension before your time to that ultra low late cuz we were talking about FPGA was a hardware hardware acceleration was the phrase. Mhm. And then cloud I mean your point about service and space I mean cloud is just somebody else's data center. I mean it's the same model it's just you know your AWS your Google's your Microsoft's are running the data centers rather than JP Morgan or Morgan Stanley.

13:04 I'm I'm I'm tying myself in knots here. What what we've been on a journey though haven't we? We we have and I think you know the same with with everything right? Not everybody needs an FPGA and and not every use case can be satisfied by the cloud. What the the right approach is to do is to take the the best of all options and use them in the right, field and and places. so, yes, cloud has has been, very important. and I think, you know, what we've also seen though is cloud was heralded as the panacea.

13:40 anything that you want, you can just put in the cloud, don't have to worry about it, it'll all just work. yes, that that can be true, but what we've seen in even in the market data space is a number of different partners come up who help optimize your usage within the cloud. Yeah. >> and to lower your cost because if you don't understand what you're using and how you're using it, it's very difficult to maintain that, cost advantage, if there is one by doing that. I mean, you know, cloud's great for for, bursty behavior.

14:12 it's great for managing variable amounts, because you can scale very very quickly. >> Yeah. but, you know, from a a latency perspective, it might not be as good, as you can optimize with with on-premise, based solutions. And if you have a consistent workload, it also can be challenging to to compete on on costs if you know, your your workload and it's a a constant flow. And it's all horses for courses, isn't it? And and I guess most firms have probably got a hybrid approach. They've got on-prem stuff, processes, and what have you, hardware.

14:51 And then it makes sense to do certain other processes in the in the cloud, but, you know, do you think we we've been on this educational journey, and I suspect some mistakes were made where stuff was tried to be put in the cloud, and then it didn't really make sense. I think people are learning about the cloud. I mean, egress and ingress is an interesting debate, isn't it? Hey, yeah, put all your data, don't we'll we'll we'll charge you nothing to put it on here. Oh, you want to take it out? Oh, that we'll We'll you the invoice for that. I mean, I mean, there's been a few little gotchas on the way I've noticed. Yeah, there certainly have and and I think, you know, not just from that perspective, but also even selecting the the instances within the cloud. A lot of people don't even realize that you know, there's often two or three different processes that you can choose from.

15:43 Yeah. Yeah. You know, AMD is one of them. lots of people don't necessarily do the benchmarking when they move to the cloud to see which one is actually more cost-efficient based on the the running costs. is it better to go for a cloud instance which is a virtualized and you get a virtual CPU. and are virtual CPUs between CPUs even the same thing? you know, one thing that we notice quite a lot is on some of them some of the the providers, you will find that you know, one vCPU from an Intel Intel is different to one vCPU from an AMD.

16:24 because we have things like multi-threading and and others. So, trying to do a direct comparison just off naming conventions actually can affect your your overall performance look at at how an application runs. So, I think it's becoming more and more important to understand what it is how the infrastructure is built that you're running on in order to get maximum efficiency in terms of cost and best performance as well. Do you think in the early days of the cloud that people just said, "Oh, we're going to we're going to pick between AWS, Google, Microsoft, whatever, Oracle, newer player?"

17:05 And and do you think people didn't ask those questions about, you know, they they picked AWS and and put their eggs in that basket, but then didn't ask much more about the basket. Didn't say, "Well, what type of straw is used to make that basket?" I'm I'm mixing the metaphor terribly. But, did they just say, "It's a basket, and I don't really care what the basket is made of." It was that the early days? Yeah, no. I think I think that that was.

17:31 And you look at and I think most of the clouds in the very early days, I mean you really had one type of instance. But, I think what's evolved is that the clouds have gotten smarter. The clouds want to be agnostic in in one sense because they want to give all of the different providers an an opportunity. But, they also have their own custom designs now as well to attract specific customers in for specific use cases. And you know, you see this with people developing their own arm Well, yeah. Because I I read about that. So, Google have got their own chips. Because I think again, we're going off in a bit of a tangent, but I think it's an interesting one.

18:15 Again, one of you you guys, but I'm you won't mind me mentioning Nvidia because Nvidia is you know, got a lot of profile not at least because of its share price. Although your share price you I think lunch is on you next time I we see each other because I think the AMD share price has done rather rather nicely. But, were some of the cloud providers feeling uncomfortable that they were almost too dependent on the chip providers, and then they got in on the game. They said, "Well, if you know, we can make our own chips." And describe that story. I mean, what's going on there? What are you know, obviously you've got an AMD perspective, so it's you're going to give it your spin. But, I mean generally educate me.

18:56 >> I obviously I can't speak directly on behalf of any of the providers or why they made the decisions. But, you know, what we can certainly see is there's within the semiconductor industry there's a lot of competition to to innovate and differentiate, right? And it's the same as being in the market data space. People are looking for ways to create new advantages with their products. And you know, but there are some limitations, right? By going for a single vendor's dedicated piece of silicon. If that's not widely available both on prem and in the cloud or multi-cloud, then it raises the question whether or not you need to maintain two or three different code bases to be able to run your software across multiple different platforms. So, there are challenges by by doing that and you might be able to gain an advantage on a specific use case, but the trade-off is going to be maintenance and longevity of that product. And also being able to move between the different cloud providers. I think that's where the x86 technology that both Intel and AMD use, which underpins our technology, is widely available on prem, in the cloud, and across multi-clouds.

20:26 And it doesn't require any code change to move between any of those different options. So, it really gives you that that full flexibility to pick up your application and just move it. Obviously, if you each of those architectures are can be slightly different. So, if you want to take the get the best out of it, you're going to need to make some small minor modifications. But, in general, the technology the the basic instruction set is the same, and so you can run your applications without changing the code, which is significant.

20:57 And and that I've seen that happen a lot. You know, you can get the absolute best in a sort of customized way. And that and that can seem like you know, for the precise thing you want to do that particular solution is the best, but then it's sort of you're stuck with that that provider, that particular setup, and you've got no flexibility to change. And and the supportability, you're very dependent on that provider and what have you. So, yeah, I've seen that happen in my long career that sometimes a slightly more you know, standard if standardized approach can be more safer.

21:34 Yes. And and I think you know, especially now it's it's interesting to see the approach that for example now in Europe, the European market is taking where they have active programs now to have a technology independence and come away and use their their own European homegrown technology as well. and including clouds for data sovereignty and and other use cases.

22:07 I think the EU announced that only in the last week or so. and for example even you know, the the French government is now moving entirely to a French-owned and and made video conferencing software for the the local government. I mean And I think those kind of things are are interesting because and again relate to this freedom of the underlying architecture being standardized because you know, if there are new regulations that are proposed in the future by Europe or or another location which requires people to hold their data in a European sovereign cloud, that could be a a major change.

22:51 I don't think we're there yet. I can't predict what the European regulators are going to do. But I think we all know that things can things can change fairly rapidly. I'm going to risk something here cuz I'm going to stray in a little bit of political area. I not asking us to take sides or anything but just observing the political machinations that are out there. So, you you can ignore or duck these questions if you don't like them but I'll I'll I'll throw them around a bit anyway because again to that last point is this sort of EU I mean it is because of some of the politics that has gone on more more recently. Am I right? Is is AMD apologies for not knowing this? Is that an American I I I'm sorry.

23:30 >> yeah. Yeah, we're based out of Santa Clara. Yeah. And again, I think this is one of the observations that you know the vast majority of these big tech companies are are sort of US based. But again, help me out here a couple of other issues. Isn't Taiwan the sort of this massive manufacturer of of silicon? And then you've got a sort of China discussions of what China is doing. And then I I'm digging even date deeper the rare earth metals that are needed to build all these chips. I mean there's a sort of layers of politics here. Again, bit of an unfair thing to throw at you but at a very generic level how does that all play out?

24:10 if I had a magic ball David then I would love to use it and and go to the right place. But no, Taiwan is been a great partner for us TSMC which is the factory. >> Why why is Taiwan why is Taiwan this this big important center? I haven't fully understood that. So, there there's a company called TSMC who manufacture chips. and they have done exceptionally well at manufacturing on the the lowest nodes. So, basically very very small transistors which are used to make any silicon that that you can think of.

24:48 now not every application needs the latest node and the smallest chips. as you get smaller it's better power efficiency, better performance, and you can do more in the same amount of space or physical space because it's smaller. and Taiwan, TSMC happen to be leading the way, with the most advanced nodes. >> And and they make some of your chips. They they make They make >> They they make a lot of our chips, and they make a lot of, our, people like Nvidia and and others, also, use that factory.

25:26 And it's one of the key, differentiators. there's a lot of different foundries, around the world, specializing in different, sizes, and people are are also catching up in in various different markets as well. So, but right now TSMC is, is the leader in, in that space. So, AMD, Intel, Nvidia, you your trick your your skill is designing the chip. Is that is that your role rather than being a manufacturer? Or do you also have your own manufacturing plants? No, so so we, you're you're right. We, we're we're what's called a fabless, company at AMD.

26:05 so we don't, manufacture, in our fab. Intel on the other hand does have its own manufacturing capabilities. and, and they can can, design and and manufacture their own circuits. whereas we, and, and Nvidia primarily use TSMC. Got it. Okay. So, that that's a sort of the the layer. And and again, I'll throw in that rare earth. I mean, do do does the rare earth stuff is it lithium and thing? I mean, what Is there a problem there brewing or is is that, you know, the shortage of those rare rare earth? Does that matter in the grand scheme of things?

26:46 I mean, look, there there's always, new technologies and people always find new new ways around problems. there there's lots of demand for a lot of these rare earth. You can see that with the commodity prices recently, but I don't think it's going to affect the market data industry too much. >> I I'm just whimsically I'm just curious about some of these things, but I I I digressed a bit there. So so back back to the market data industry. So we're we're in the market data industry, you know, we followed that low latency debate, then we went into the cloud debate. We're in the AI debate as a sequence. And then in parallel to all this, you you came along, you know, and started talking to us. And it was fascinating to to sort of throw in the the hardware dimension. I mean, can you is it unfair to ask you to give us an idiot's guide to the sort of the hardware? I mean, what are the sort of elements that somebody looking at different hardware I mean, frankly, you know, I I've done, you know, I've got my laptop. It's an AMD laptop rather than Intel cuz that's the CPU.

27:51 But then there was the graphic cards where Nvidia and you were I don't I don't think Intel does GPUs, do they? I mean, is it's mainly They've started to do a few things and they've had various realms into it, but mainly on the consumer side, but yes, I mean, AMD is is definitely the number two behind Nvidia and we're catching up pretty quickly on the GPU space. And I mean, I I remember cuz I in my youth I used to play games. I mean, graphic processing units were more about games in, you know, 20 years ago.

28:28 When when did that when did the capital markets wake up to GPUs? It was a it was like FPGAs were earlier and then GPUs started to be appreciated as a way of hardware acceleration. Have I have I said that in the right way? I think what you'll actually find is that GPUs have been used for quite some time in the industry. Okay. as far back, I think is, you know, early 2010s, GPUs have been used in in the finance space. But, GPUs and and FPGAs solve completely different problems. And same with CPUs.

29:05 So, you know, I think where where people if So, a CPU is a general-purpose piece of silicon, right? It's easiest to program. You don't need to have any custom languages and and you can run all of your different applications on it. So, CPUs are great for general-purpose use cases. and, you know, if you want to take advantage of some of the specific microcode, you can get even better accelerations with CPUs. So, we've seen a significant increase in adoption even of AMD CPUs in this space because for low power benefits and performance.

29:42 And that power and the cost of electricity is Yes. Yes. It's it's become more and more significant. So, we've seen customers deploy AMD architectures and replace their entire like grid infrastructures as an example. and save on up to a third a year on their annual electricity costs with increased performance over what they had before. So, that's a significant saving and and something that COOs and and others are very interested in, right? How technology can play a role in improving your ESG scores, lowering your impact, but also saving you bottom dollars. So, you can go and actually start investing that elsewhere.

30:24 Now, an FPGA is which AMD also makes after our acquisition of Xilinx. that's really designed for low latency customizable programming. it was actually designed for prototyping chips and and other things in the past, and being reprogrammable hardware, but the financial industry has used this for for trading applications. Now And you have to program it with a very specific type of language. I can't remember the name of it. >> Yeah, very low level VHDL is is one of the ways. So, you need a specialist engineers to do this and it can be very complicated.

31:02 And the industry partners, some of which I mentioned earlier, who leverage this and and and can sell you full baked solutions. It was the Exegy world and the the old Active Financial. Now, Active Financial is now I've Is it part of Options? I I've lost track of who's bought who. Yeah, yeah, exactly. So, there there's been a number of third parties who've you know, used this technology and are offering it to customers or building it in-house, but it is very specialized and it can be quite costly to to maintain.

31:35 Now, GPUs are actually very highly programmable devices as well, but they are really not low latency devices. So, the what they can do is do a lot of processing simultaneously and in parallel, much more than a CPU can, but also at a much higher power draw and All right, right. and a higher latency. But if you've got a big enough set of data, your overall time of job completion is going to come down, which is where GPUs have really made a big inroad and a big difference. So, it's not about latency of a single transaction there, it's about the latency of your entire model that you're you're running on the GPU.

32:25 And this is where AI has come in as well. There's some pretty large models which are relying on highly parallel programming, which is where GPUs are really useful to to do that. But inferencing is still often done on either an FPGA if it's an extremely small model or on a CPU. So not all inferencing is necessarily done on a GPU today even in the AI space. So let's let's segment our audience, the people who are probably watching and listening now. We I hope we're going to have some quite technical people listening and I and I might ask you to dig into a bit of your inner geek on on that. But again, there's probably a lot of people and I include myself who who don't have a deep technical background. So what what are the sort of discussions and questions when you're going talking to Well, I I suppose you've got different audiences. You are selling to you know big bulge bracket banks and high frequency trade In other words, the actual financial institutions that are doing the trading. So they're one audience, but then are you also talking to and selling to the sort of the old Refinitivs now LSEG and Bloomberg and S&P and FactSet. Are you selling to the exchanges, the the people who are providing this data?

33:46 I mean, what's your audience? What's your marketplace out there? What what what sort of things are they thinking about and asking you about? So I mean, I think you know, everybody needs technology in this space and and AMD is one of the I think or if not the only firm silicon firm in the world which produces FPGAs, CPUs, networking cards like AMD's Solarflare and GPUs all under one roof.

34:18 So really we have a quite a wide variety of different topics to go and speak to people. So if you look at the the most basic or fundamental improvement that to performance that that people can make exchanges and and others is is just dropping in something simple like a Solarflare networking card. and that is why nine out of the 10 top exchanges in the world run on Solarflare. So, you know, you can take your application, put in a a Solarflare card, use our onload, and immediately get significant performance improvements in terms of messages per second and latency to help your software application get a leg up and be more competitive without making a single line of code change, which is pretty significant. And that's one of the reasons >> and that's interesting. The whole latency discussion is a a sequence of different things. And and and you've got to look at your latency from start to finish, and there might be 10 different steps. That's right. And and you know, it depends which one you want to to do, and that's why I sort of said that what we found was that starting with the Solarflare networking card was the biggest individual uplift with the shortest amount of involvement necessarily from the software development, reprogramming, and and so on. And we've got a great team in Cambridge who who support customers in this space and are very knowledgeable about how to help I I need to have an affiliate link, don't I? I need to have an affiliate link.

35:56 Yeah. Another day. and then you know, what we see from there is people look at the CPU as the next step, right? And because if you're running a a software-based application, the first thing is get some free boost, which is your networking card. Second thing is how do I then get the best out of my existing software? And that might be to consolidate your data center, reduce the number of servers, do more with less power in in the in the same footprint, which is really important, especially if you've got sort of co-location or or other use cases.

36:34 But also, as we've seen market data volumes grow pretty substantially, every year we're setting new records in terms of total flow. you just have to look at Opera or any of the feeds, it's just getting larger, it's not getting smaller, right? And especially as these exchanges are upgrading their technologies, that in turn means that people can now put in more messages and and the volumes continue to grow as well. so having the capabilities to to process that, in the same footprint, is is really important for for customers. And and being able to analyze that market data and make use of that market data, capture it, do the timestamping, all the rest requires, a good hardware. You can do timestamping on the StellarFlare, you can focus on your CPUs to, AMD CPUs are giving a a massive leg up, allowing you to do more, in that same power profile as well.

37:34 >> is is by far the most focused on server-based CPUs. I mean, does does does the desktop feature much? Do do people care Is it Is it servers where their game is really? it's it's it's both, right? data center is is definitely where the larger volumes necessarily are are coming from these days and in terms of the higher end and most market data platforms are doing that. But AMD also provide a large range of, desktops, so, with our Ryzen, which is in our laptops, in our gaming, servers, they are really leading the way, in terms of, performance and and and value and value. But now you can see, we've got another product called Threadripper, which really sits in a class of its own, but that is for trader workstations and and others.

38:28 >> Got it. Be able to process all of your market data locally with incredible speeds and and that's really the fastest options available. So, it's a sort of think about FPGA it and that's probably only a very small portion of the of the industry. Then, most people are going to be very focused on the CPU. And then, it's the GPU. So, there's a sort of see and and and the network card, which I yeah, so those are all I mean, just going back to one point and I and I don't know, I've got to be careful about using my words here.

39:01 Power has two connotations. There's the power of the pro How How powerful the processor is is one definition of power, but then the consumption of power, the consumption of electricity really. Yes. I mean, if if if if you go along and and you're talking to a customer and you say I can I can give you 5% more computing power and somebody else comes along and can offer 10% more computing power, but then you've you've reduced the electricity consumption by 20% versus their 10% or something. In other words, which is more How are you finding out there? How How do people trade those two things off? Will people give up a little bit of compute power to save on electricity costs? Or obviously, they want both really.

39:49 Yeah, and I think that's really it's a good point and and you know, we have a an AMD a large number of SKUs. so, we've got some high frequency components, which are a little bit more power hungry, but give you that extra clock speed and and that extra Sorry, a SKU. What's a SKU? You used the word SKU. SKU it's a a version of our CPUs. Not every CPU is the same. Some have more cores, some have less cores, and they're all made with conscious design choices. Just to your point there, we go up to, for example, 192 cores on one processor.

40:27 Which And and so what? I mean, what does that Is that more powerful, more computing power? I mean, what is that That's more cores in a smaller footprint. So, what it means is within inside the same rack size or server footprint, you can get more cores to do your your application. And that's really good for things like Monte Carlo simulations or any large-scale data because you want the large number of cores. But, those cores aren't going to run as fast as a lower core count CPU. For example, our 64-core CPU is a high-frequency part, which is very popular in trading applications because you get much higher clock speed in order to run your trading algorithms when you need to respond to market events. But, not everything needs to respond to a market event in that same It comes back to what I was saying earlier. It's If you understand the the hardware, you can make the right technology choice, and this is where people like myself from AMD and our our team around the world helping customers make the right choice because just testing one of these devices might not be the right device for your specific use case. So, having that bridge between hardware and software and application uses is going to be really important to get the best choice for your your application.

41:54 >> how sophisticated is the customer base out there, or does it vary massively that some are are asking very intelligent, very perceptive questions, and others a lot I mean, you know, how much education do you have to do out there? Or or you It varies widely, David. You know, there are there are some people who technology decisions is their bread and butter and and that's what makes them makes them money. And they're they're very in tune.

42:22 They understand a lot and they can customize a lot for for our applications. But there are other customers out there who really need a a bit more education from that perspective. But I think this is not a new problem, David. You know, we've always had these partners. I spoke about the FPGAs. Not everybody needs to write an FPGA cuz you can go and buy a solution from our market participants. And this is what I said earlier. I think what we've seen now is in the market data space a lot of new market entrants who are starting to do that optimization, get that benefit, and be able to showcase this to the industry so the industry can just leverage some more expertise from from this this industry. Throw throw some examples out.

43:11 I mean, who are these new these new upstarts, these new I'm curious just for my education. No, no. I mean, but you know, I mean, I'm when I say upstarts, some of them can be very new and some of them can be a little bit longer, but people like Cairo is brand new. Okay. Who is you know, they they've got some really interesting technology and and launched last year. Markets IO Oh, yeah. very new in price as well in the grand scheme of things and are also leveraging new technologies to to help bridge that difference for for customers. So, and there are others that that that are coming up and and and people are innovating in that space. So, I think it's really interesting and a lot of these guys are doing it on commoditized hardware and this is for 90% of the market, right? Rather than the cutting edge.

44:05 As I said, you know, you've seen some consolidation in the industry with some of the FPGA providers and and the ultra low latency space. But, you know, our industry goes through these these events all the time, David. You know, there's consolidation and and then new technology startups are coming. So, it's just nice to see that that we're not at the end. We're just at the beginning and we always are. Actually, that prompts a question that I've I've talked about a lot of my podcast episodes.

44:34 When I started in the industry in the '80s and maybe even in the early '90s, I mean, banks and stuff were throwing money at at different solutions all trying to beat each other. It it was the wild west in a good sort of way, an exciting way. Sort of money was no object. Just throw money at it and and we'll see what happens. But then, various crashes happened, various things went on. And what for the last good 20 years, really, we've been in this world of cost-cutting, of of being so conservative with a small c, so careful.

45:07 I mean, how how cost-conscious is the market? I mean, how you know, in terms of What was it? Some somebody said to me This I've told the story many times on podcast. You know, it's like I've gone out, I'm trying to sell this new product called a 20-pound note and I'm selling them for 10 pounds. And the customer said, "I've just got no budget. My procurement process process won't You know, it's like you've got this slam-dunk solution, but people are just so cautious and and cost-conscious. I mean, how is What do you see out there? What what is the the nature of the market at the moment in terms of its willingness to spend money?

45:47 So, I mean, from my perspective, David, certainly what I see is that, you know, it it comes back to the the customer that you're you're looking for and understanding how they're going to make benefit. If if you can show a material business benefit by investing in something, I think the money is always there. I think but there's also a driver to reduce costs, which is a a constant feature of any industry, right? so, you know, and I think with a lot of these new providers as well, if they're able to show, "Well, we can do the same or give you a performance improvement, but only using half the hardware." then that's a material cost saving as you're upgrading. So, it makes sense to go and do those even from a cost perspective. If you can also show that you're going to increase your performance while reducing costs, it's a win-win for everybody. and so it's a a meaningful process to to go down. and I think, you know, there are customers who are willing to invest.

46:54 you know, at AMD, we've got some specifically designed FPGAs, which are at the very, very high end, and people are willing to pay for for that latest technology at at a a significantly higher rate than a standard commoditized piece of hardware. So, it it really depends where your you're investing your money. And I think what we see now with AI is budgets are going into AI, and people are spending a lot of money. I think the estimated value of the market in a couple of years will be nearly a a trillion dollars a year. so it's a it it's fairly significant investments, and I think people are investing. They're just looking at where to invest their dollars and and how they're going to get their best bang for the buck. I mean, we're we're in the final 10 minutes. AI, I mean, where do you start?

47:47 I mean, there's there's there's all this talk about a a stock market bubble, but which we won't go into. But I mean, in terms of of actually where we're going with AI, what are you seeing out there? Where where is AI being used most? I mean, are you seeing an increased demand for your products directly because of projects that are running in some sort of AI context? I mean, there's there's lots of different flavors of AI.

48:14 Again, set the scene. Where where is where are we going with AI in 2026, but also over the next 2 or 3 years? Are you just seeing a sort of an exponential growth or a linear growth? I mean, what what are you seeing out there? What are you hearing? I mean, there are there are plenty of reports about the industry and and the growth of AI and and most of them look towards more sort of exponential growth in in terms of volume and and and trade there. I mean, what you know, we're still in a relatively early stage of AI. and if you look even at the announcement that our CEO made just under a month ago now at CES. if you look at our MI320 5 and and some of the older GPUs, and when I say older, I mean like 2 years ago. Yeah. It's all relative, isn't it? Yeah.

49:09 The the GPUs that we need 2 years time will be a thousand times more powerful. so the the growth in processing capabilities is immense. >> Has has has Moore's law gone out the window? It's a completely different law. cuz you know, Moore's law really related to I guess compute power and and CPUs really at the time. and GPUs have just And on a different level, yeah. on a different level. But saying that, GPUs can't do everything, right? And this is where it's important. You still need your different technologies for different use cases. But but yes, I think one thing that we're going to see is a lot more competition in the GPU space.

49:50 more movement towards open standards. You're already seeing this with technologies like UA link and UEC, which are all competitive to proprietary products from the market leader today. because I think people again, they don't want a vendor lock-in. They want that open source. They want that open standards to be able to >> whoever's got the lead quite likes a lock-in. The I mean it Of course. you know, I mean it wouldn't you?

50:22 Look. I mean well, AMD is is always been open source and pushing open standards, right? That's our bread and butter from from an organization standpoint and something we strongly believe in. We want to win because we've got the best technology, not because we force people to be stuck with you. Yeah. Be be stuck, you know, and and we believe that we've got the the best technology and the great road map across all our product lines to to enable people to choose AMD and and make the choice that's right for them. so I think you know, we'll certainly see that that further adoption. We've got some really exciting AI clusters coming out now. I think to your point earlier, you know, the data center versus the the desktop and and the workstations, we AMD historically you know, was gaming in the the desktops. and we're starting to make a lot more inroads into those workstation use cases with our ROCm software, which is giving you the ability even on your laptops to run these AI models and transfer them directly into large-scale cloud deployments. and Helios is coming out later this year, which is our rack level AI clusters. it's done in a very different way.

51:45 again, AMD is not selling the rack itself. We have you created a reference design and working with our partners in the industry, all of the OEMs to build these scales and and provide that service and and really building that community ecosystem. So, we're we're we're fast approaching the end. You know, you you I think you were skeptical whether we get to an hour at the beginning, but I knew we would. I mean, I I always and I I warned you about this or I advised you, not warned you. this chance, you know, what haven't we talked about? What couple of messages, what couple of educational points would you put out there? Apart from buy AMD, which I think you've eloquently said quite But what I mean, couple of parting thoughts? I think even more than than buy AMD and I think back to my point is that hardware has made significant changes over the last 5 to 10 years in general. more significant and it's changing in a much faster rate than it has in the decade before. And I think that's been driven by AI, it's been driven by a number of different changes in the industry, which require higher data rates, higher lower power usage and and other use cases. So, what I think is is important for for the industry and why it's relevant to market data is to understand the platforms you're running on. take a look at what's out there. and see how you can lower your cost and increase your performance by making smarter and better choices and understanding how to use the the commoditized hardware that's available for you today. And I think that's that that's the real key is that there's a lot of choice out there.

53:34 Go and look at it, evaluate it, and make your own decisions on on what makes sense and and how you know, you can bring new value back into the organizations as well to showcase a cost saving and then a performance improvement. But it requires a bit of self-education but you know to understand this because people haven't been as engaged with hardware as as as you're saying they should be now. So it requires that audience to up its game a bit its skillset and whatever and and and look I I applaud you. I mean I've met you because you've engaged with the FISD world. This is David Anderson. This is not me as an FISD person but obviously I know you primarily through my FISD conversations and I think I think it's really refreshing that you come out to talk to us and to engage with the the overall community. Intriguingly so now that providers have sort of way back in the back they they haven't chosen to quite come forward as much as you have.

54:33 So I think that's that's exciting and that's positive. So thank you for doing Thank you very much for the invite David. It's always a pleasure to to talk to you and and speak to the market data community at large. And and hopefully we will see more of each other although as I am pulling a little bit away from the FISD world you might see me a little bit less in the FISD world but I'm still I'm still here.

54:56 So you know anyway look thank you so much and I'm going to stop the recording. Don't go away but cheerio. Thanks very much everyone.