Raghib Hussain, Altera | theCUBE + NYSE Wired: AI Factories

Palo Alto studio connection Silicon Valley and Wall Street. I'm John F co here with Dave Volante my co-host. Hello, I'm John Furry with the cube here in the PaloAlto studios. Of course, we have our NYC studio connecting Silicon Valley to Wall Street. This is the NYSC wire cube AI factory series. Rahib Hussein is here, CEO of Alterara, recently taking the helm as the chief executive. Thanks for coming on the cube. Great to see you. Thanks for coming in. Appreciate it. >> Thank you. Thank you for for the time today and I really look forward to our discussion today. >> So you had many entrepreneurial journeys around data uh in semiconductors at Marll and now CEO of Alterara. There's a lot of history with Alterara. If you look at what's happening in the semiconductor business, it is mainstream and there's a headline every day more capex, more AI infrastructure. you've been at the center of it and now as the world grows the demand for mathematics, the demand for data to be processed in an AI way is not your yesterday's general compute that we all know in our PCs. It's kind of moved to a whole another system. We've been covering it with AI factories. Talk about the current state of Alterara. You're at the helm. Talk about the company, where it's at, what it where it came from, and where is it today. >> Yes, Altera is a 40 plus years old company. um it it has uh you know history of delivering the programmable solutions um throughout throughout the history. We have shipped over 5 billion units over over time. So a lot of experience, a lot of expertise um and then um it was uh acquired by Intel 10 years ago and for [clears throat] the right reason of Intel they focus it to be a more of a x86 accelerator um which which made it focus on a certain segments of the market. But uh when I saw that Intel is um deciding to divest this company, I I looked at it and I I realized that um Altera is still has the uh industryleading fabric because in FPGA fabric is everything. If you have a great fabric, if if your end customer can get the highest frequency in your design, if they can get use it for the highest utilization, that's is the most important part. So then I thought that hey it's a it's a perfect opportunity because we are living in a world which is going through a you know huge transformation and everything pretty much is changing you know AI is going to force redesign of every equipment out there and models are changing on a weekly or monthly basis. So I I think it's a prime time this is where FPJs are designed for. So I I I looked at this a huge opportunity. All we have to do is to go and and you know kind of make it independent uh put a full focus you know 100% pure play FPJ company which is independent and which can serve the the entire market u invest properly in the in the portfolio and uh really help our customer solve their problem and that's why um I was personally very excited about taking ultra private um team is excited There a lot of excitement going on. There are a lot of momentum. We are focused on building a great company which solves customer problem. >> I want to get into the transformation and the culture but I first want to get out there that um a lot of people are learning more about semis every day as we see the mainstreaming of it. FPGA explain what that is and why it's important because people are now realizing that the role of architecture in systems there's a lot of old concepts in computer science and engineering that are now coming in and being re not refactored but like reconfigured from a system architecture standpoint that's producing some significant value and I think FPGA is one of them. Explain what it is and why it's so important right now. Yeah. So, FPGA is is a part of the uh compute heterogeneous compute solution. So, if you look at the full continuum of the compute on one hand, it is general purpose CPUs and then GPUs. Um and uh on the other hand, it's it's a very optimized application specific uh semiconductor which is called ASIC. Uh FPJs are between between them. In other words, FPJ provides um a ability to implement the hardware. Um but it is in a flexible manner which means even after deploying the product in the field if you need to change the uh you know property of the hardware you can change. So think of it FPGA as a as a programmable application a specific semiconductor uh programmable even after deployment and that is why FPG has a unique position in the industry especially in the industry which is going through the transformation or evolution and right now we all know the whole world is going through transformation what we what we believe yesterday is completely changing today right so in in this in this changing environment and changing market the the role of FPJ is even more important and I'll give you various reasons for that. The first of all because everybody is trying to optimize the the data movement for the AI. So compute is one part. So but data movement is the biggest biggest problem to be solved and this is where even in data center FPJs are a very good companion for the GPU and CPU. um because FPJs provide the ability to implement any non-standard data movement protocol or any non-standard data connectivity or even the standard connectivity which is needed for the next you know generation of the product but AS6 cycle takes time it takes 15 to 18 months so while you are developing your ASIC you still want to don't want to miss the window of of you know utilizing the best way of your compute so what you do you FPGA connectivity solution. You go through that and then you your ASIC is coming 15 months later which will swap the FPJ out but then you while you're doing that while you're designing the next ASIC you still need another FPJ to catch the next wave of the of the data movement right this is where it's it's a beautiful place you can call it glue logic or you can call it you know advanced uh you know data movement logic whichever way you want to call it but there is a very defined role of FPGA in a data center. Now things become even more interesting when you go into the physical AI. In a physical AI you know you you need to gather data from lots and lots of sensor I mean just if you look at let's say a robot humanoid robot there are tons of camera there are lightar there are radar there are you know force sensor there is there is the you know temperature sensor pressure sensor all those type of things. So all of these sensors are streaming data um to the to be processed. Now um these data has different data format different data interfaces different latencies and uh for for brain for GPU to understand and and process it properly it needs to have a coherent view of the of the world what's going on which mean all these data needs to be gathered some is change you know coming at a thousand you know ticks per second some some is really 30 frame per second they all have different format they need to be timed properly because if they are not timed properly the the you will create a perception of the reality which did not even exist right so all these data needs to be in digested digested needs to be gathered needs to be cleaned needs to be you know putting in the format easy to understand for the GPU need to be synchronized with each other and that's that's all the work which FPGAs are great in in doing it and providing it to the to the >> so versatility adapt adaptability. So what you're saying is is that the old way was okay, we have a chip and a change in the software, wait for the next version. What you're saying is the the the value is that you can actually adjust to the current chip with programmability and then swap out the next version, adjust it. So it's more is that kind of >> it's getting the right right solution much faster number one but more importantly a solution which gives you a deterministic low latency data movement that is the key the key is the determinism low latency security and safety especially in the physical AI right so so the the value is that you should be able to get the data and you should be able to react on it in a in a you know very very um deterministic manner. I mean I I give it as a high level example that GPUs are you know uh G GPUs make robots to think but FPGA make sure that they act and behave in a timely manner in with safety and security. >> Yeah. And all the talk here at Stanford for the hot chips event is about disagregated serving. you're starting to see demand on connectivity and capacity and token sizes. So, a lot's going on in the tech um puts you at the front and center of the market. So, take us through the culture because you know, okay, semiconductors, FPGAAS are out there. You got customers that putting them in in PCs, devices, data center, servers. Now, the world is spun to integrated systems. Hundreds of models potentially running on robots. A lot of interactions, a lot of networking happening. Moving the data is the key. Talk about how the culture shift has changed at the company. Now you're in the center of the action. Does that change hiring, product strategy, deals? Take us through what the transformation of Altera is going through right now. >> Very good question and it's all over the place. So let me let me explain. So you know in a in a large corporation many times things are being done through the through the you know financial analysis and the and it takes it takes a long long long time to gather information from various groups and various things and it it's a long decision cycle right um and many times um you know large companies when you are in that mode they they get into this mode that hey we need to understand exactly what needs to be done and we will design and then customer will use it. Right? Throughout my semiconductor experiment, I have done completely opposite. I believe in work closely with the customer, understand their problem, solve their problem with the engineering way, with the technology and and make them create value, make them happy. And if they're happy, they will be they will give you gifts. And you know what is gift is the is your revenue, right? So it's it's really the whole in my view the whole sales process is not trying to you know kind of >> sell them something. It's not about misguiding them or tricking them into your thing. It's about understanding the problem and solving their problem. Once you understand their problem, solve their problem, work as a single team, then automatically value is created and value is created is is for everybody. So there's no there's no >> so on on the customer solving problem that's which by the way is a true formula everyone knows if you work on the customer they'll pay you but there's also feedback we're in a highly iterative design process so >> so we work very closely the culture that I have is this work with the customer as a single team we we try to really work as a single team understand their problem and and get a very fast reaction so this is another thing which which is important and again comes back to the culture where How how is your decision process? How fast can you make the decision? Right? Um I believe there are mainly organiz in in a typical business there are there are people who you know create the product create the you know technical solution and there are people who sell and support. So when I came in altera there were many many groups and the layers between the customer actual customer and the actual engineering were several layers right that's not a winning solution. So what first thing I did I I collapsed these organization. So we have sales and support and then we have engineering. In fact we give direct engineering access to the customer. Now you know a lot >> of the buzz words for deployed engineers but that's what you're basically doing. >> That's exactly because in my mind engineers need to know what is customer real problem because two things first they will design the right thing or if they have designed the wrong thing in the last cycle they will learn from that and then correct in the next time. Right? So it is very important to understand you have to be on the ground to understand the reality and and that is where I believe this org structure itself matter because then people have the right ownership and accountability. They know what they are supposed to do. They know their their goal their goal is to solve customer problem. It's it's not just oh I created a beautiful product and customer is wrong that they don't understand. >> Yeah. They're pedalling wares as they say in the old days. talk about the um inside out versus the outside in. It's a you know people say, "Oh, we're an inside out organization. We take our process and push it to the customer." You're saying the opposite. You're outside in. You're going to the market. Pull that back in. >> I'm saying embrace market, embrace customer problem. Think of it that you are their engineering team. You do not need to think them. It's usually it's us together. Us and customer are one team, right? And when you think from that point of view and engineers they are very logical people. When you think from that point of view you will come up with a logical answer. you will not really give people run around right and this is the this is the value you can create by having that environment that having that mindset that yes we can instead of you know many time in the past in I have come into a situation where people are analyzing things for let's say 3 months and I'm saying folks amount of time we took to analyze this thing you would have been done >> you're going to build a solution >> you would have built the solution [laughter] move on so get things done is our moto. Yes, we can and get things done. Because unless you go with that mindset that yes, you can >> you will not be able to do. If if if you have doubts about your ability, you have lost half of the battle anyway. [laughter] Right. >> Yeah. >> So, so that mindset that growth mindset that not only yes we can but also if anybody in the world can do something we can do. So which means instead of comparing yourself with your last year hey last year we did this so this year we are doing 20% better so we are great. Yeah, it's a nice thought but in my mind always compare what is the best of the best in the world and how are we comparing with that and if they can do it why can't we do it right so we should we should really you know keep encouraging ourself to keep improving every day because it's a in my opinion it's a it's a continuous improvement >> you know engineers are very logical and if there's a layers between them that just gets lost in translation that's one it's obvious that's an obvious thing but also engineers like to solve hard problems and so if to service the problem. They love it. Now take that to the market today because we are in a problem-solving systems architecture game. It's not about the chip, it's about the system that's now solution. >> That's the narrative, the solution. So, so take us through some of the things that have changed because I was just looking at um some things about concurrency that's being talked about this week at hot chips and concerns concurrency and other things coming. Those were papers in the 70s and 80s. There's stuff in computer science. So things are being applied to the new era. So you're starting to see this AI architecture and systems. What are some of the hard problems that you're seeing? And as you look at the market, there's a whole new set of requirements. You can mix and match some science, new software layers, connective tissue, glue layers. There's ways to kind of do it. We see graphs and different approaches. Google's got pods. Nvidia's got density. Serious has got the big wafer. I mean there's different like approaches. what the big problem in my that's where I think the reason I say is we are living in the most exciting time because because we are trying to solve a humongous problem we are trying to solve the problem of how do we bring the intelligence at the most power and costefficient manner and that is why as you talk about we are talking about papers which were written before like a decades before and at that time we did not have the right technologies now we have right technologies we analyze it. We can look into modeling it and and applying it a real because the problem the the cost is so big and the problem is so big even applying various specific solution may create value. I mean you you already have seen in in the data center between training and inference there are many many type of solutions and guess what each one of them have their own value ad their own area which we can they can solve the problem. So, so I really believe in this era where this this intelligence uh is growing at a fast pace and the and the data has become I would say more important than gold right >> everything minor even if you do a minor modification it's a huge value so it's not a zero sum game it's it's everybody will get benefit out of it as long as we are focused on creating value through technical enhance advancement and this is where I believe Again FPG is a very in a very unique position. You know why? Because all those new ideas that you think of you can use FPJ to implement it and test it and in reality deploy it in production in a very fast manner. You do not need to do hundreds of millions of dollar investment right to get to because these days especially in advance advanced node >> developing an ASIC is is a several hundred million dollar you know investment and it takes time right so while all these creative minds are coming up with creative ideas FPGA provide them platform where they can implement these ideas see the result and the result look good even deploy in mass volume. >> I think the speed is key and also the capabilities with the advancements of all the systems around it. You can get stuff out that actually perform and take those old concepts of science and engineering and apply them and grow them. We're seeing that ontologies and data platforms. We didn't have the supercomputing capability. Now you do it changes the game. Now, one area that I think you guys are going to play well in besides the enterprise, which I think is perfect for Alterara, which you just highlighted, but I love the physical AI piece because that expands and who doesn't love robots? Who doesn't love space? Who doesn't love kind of the science there? But robotics isn't just humanoids running around like we see on the videos. It's manufacturing. It's a lot of other things at the edge. So edge and AI factories and physical AI, our real world is going to be integrated into our lives with intelligence. This will be a big thing. What's your vision on physical AI and how that's going to play out from an intelligence standpoint? Cost, form factor. >> To tell you truth, the reason I mean I was I was doing great at Marvel. I was president of Marvel and I was in the middle and center of this data infrastructure. Um the reason I chose uh which I chose myself to to uh you know run Altera is because the physical AI in my mind is the most exciting part of the next phase of AI expansion and this is where FPGAs have a very critical role to play. um in in in my opinion um FPJs are built for that to solving that problem. I mean I give example that hey if you look at if you look at humanoid if if brain when human being a brain is the GPU then the whole nervous system and cerebrum is FPGA because because a brain without an ability to perform the action in a deterministic manner with safety and security and in a low latency way it's it's useless right so you need you need that that determinism you need that SE safety and security and you need that latency and and and when it comes to robots, robots are not only things which we see moving around us. I believe anything any device which can do autonomous decision is a robot, right? Um a surgical equipment is a robot. Um you know a drone >> agents are software robots. >> Yeah, there could be agents or software robots, right? How do you make those agent faster work faster? could be in another area right. Um same way you know the the machine various [snorts] kinds of machine which are making autonomous decision um even the um in in near future I expect even the radio you know radio based station are used they will be robots because they will be making decision and giving information to autonomous vehicle giving information to other infrastructure. So in my opinion any any thing which can make autonomous decision is a robot. It's it's extended information. >> It's I can see your excitement. I love I share that with you by the way. But you mentioned the comment earlier about data movement. Yes. >> So first of all brains aren't one thing. They're like the frontal lobe you got multiple sections of the brain. You mentioned rising tide. You can FPGAAS you got GPUs. So your brain can have a lot of things in it and connecting that and then moving the data around that >> becomes a very key thing. And you mentioned that earlier. Tie that in to why physical AI needs to have high quality low latency data movement that's on point. >> So, so as I as I mentioned that even in our body brain doesn't we don't use brain all the time. We we touch a hot surface we react quickly. We don't think analyze okay is it hot enough or not hot enough or you know all [laughter] those things. In other words, there are in our body there are thousands of fine-tuned models are running. Same is applicable in the robot. In a in a typical robot, dozens or hundreds of models fine-tuned for a specific actions are running. There are separate model for perception, vision, perception. There are different model for 3D extraction. There are different model for you know the force you know sense of force. There are different model how to apply pressure. There are different models you know temperature sensing. So all those and then how to react on those thing. These robots are are you know when when coexists with human being they need to make a very quick decision and those decision are in microscond right and that is why you need a processing of those sensor aggregation processing and act in a very very deterministic manner and this is where I I would say the FPJs are the best because it gives a hardware deterministic behavior with the safety and security, right? And a low latency, right? Now, of course, as these things evolve, uh there will be certain application where you will be able to justify to do the ASIC if the volume is so high for that. But there will always be many other things where you you still the models are changing so fast that you want the hardware that you can update even after deployment. So in my mind there are need for FPGAs for for at a you know very low end which I call which which simply take in you know basic sensor transform it in a in a known protocol. there is a mid-range FPJ requirement which really takes those sensor aggregation synchronize them reformat them clean the data organize the data and present right and then there is there is even um higher FPJs which are needed for let's say to aggregate you know 10 stream of 4K vision and then do 3D extraction and then and then present it to the GPU CPU because all these operations um are not very well suited for the GPU and CPU to Just like all >> like your nervous system example our nervous system our brain >> put your hand on the stove you your brain didn't have to think do I leave my hand there it's latency is not going to round trip you just hand comes off >> so brain is good in processing large model and giving optimized uh you know kind of focus model as you know as you may have read the book like in tennis the good tennis players they have this optimized model how to react they don't think while playing and the fact is the moment they think they will lose because latency becomes higher right Same thing is applicable in in the entire robotics industry where you need to constantly keep evolving these model keep fine-tuning and keep upgrading in a deterministic manner low latency manner and that is why it's exciting and then the I think there's a room for everybody right >> I mean even the enterprise I was talking to a CEO off camera he said hey I'm trying to build a brain for my company so I can be have a knowledge layer and let everyone be successful I mean That's essentially what he was getting at like he needs to have all that. That's his mode. Yeah. Things will be built around software be recreated. >> Exactly. So >> I mean AI is helping every even in our company our our you know chip design cycle has improved a lot. I mean our I mean I'll give credit to my team. They have adopted the whole AI in in every aspect and we are doing things at a much faster pace inside inside our company than we what we used to. I mean I in prior interview I said that hey in the last one year we did six step out which I think for the longest time Altera did not do that in a year. >> Yeah you're getting faster. >> Yeah. So we are get getting things done faster improving quality improving our overall portfolio and and everybody is super energized super focused to achieving one goal which mean solving customer problem in the most efficient and fastest manner. >> Well I appreciate you taking time out of your busy day. I know you got a lot of things going on. Final question. And what are you optimizing for now as CEO? Culture shift is happening. You're seeing the markets growing significantly. What's your focus? >> Well, the the focus again as I say goes back to how can we bring a leading product portfolio to solve customer problems in a most efficient and fastest manner. Everything we do is focused on that. Everything we do internally and externally is focused on that. And the mindset and the culture that we are building is around that one customer focus. >> Well, congratulations. Thanks for coming on again. Enterprise is hot, agents are hot, physical AI's about to boom. We think next year will be a very big physical AI market. As the real world connects with digital, we'll be interacting with more data injecting into intelligence into machines, humans and machines. A lot of action. >> You you you got it. You're exactly the point. And this is this is the world that we have been uh you know getting ready for and focus on and and we are super excited about it. >> Awesome. Thanks for your time. I'm John Furrier host of the cube here in Peloto for the NYC wired AI factory series of cube and AI together with NYC Wired Peloto to Wall Street. Thanks for watching.