The AI Infrastructure Race: Why Storage Is the New Competitive Advantage

Speaker A: What I have observed is that I have turned into a lot of the execution heavy tasks into something that is quick and simple and give you quick feedback. Right. So I would say that me as a leader, I try to be open minded and always bring in young talents. Who is getting into this AI world with Fresh Mind, no prior experience. How do we go back to their ground zero and said what is the best way with the help of the AI that can do this much more efficiently. So that's what I always remind myself. Where can I learn that experience especially for the people that is coming forward. Fresh Mind knows much more about the AI.

Speaker B: Welcome to the AI advantage where modern leaders decode the future of tech. I'm Solomon Williams, founder of Solonox. So here it's no noise, no hype every week. Sometimes it's just me and sometimes it's the brightest minds in tech breaking down what it really takes to grow smarter with AI and automation. Let's dive in. Welcome to AI uh Vantage. I'm your host Solomon Williams. We have a special guest today. We're joined by Michael Wu. He's the general manager and the president of uh, Bison Technology usf. So they are a global leader in semiconductor and storage solutions. Michael, I'm really excited to have you on today. How are you?

Speaker A: I'm doing fine. Good to very honored to be invited in this podcast. So looking forward to today's conversation. Solomon.

Speaker B: Perfect, perfect, thank you. So I was real curious on it because you know, in today's time of uh, where how the market is, you hear so much about AI and a lot of on you know, what it can do, the new software, it can do the capabilities. Can you tell us a little bit about how did you start with Fison? How did it come out to where you're now on such a leadership position? Can you kind of walk us through your journey?

Speaker A: Of course, yeah. So I actually come to United States for the first time when I come here to study college, get my bachelor degree here, master degree here at Virginia Tech. So it was a uh, good learning to kind of get my foundation in. After I graduated, spent couple years as a test engineer at ARF Co. Learning how to kind of test all the new devices on the RF side. And then later on I joined Fison where Fison does not have, even though we are the inventor of a USB controller back in 2000, we have no presence in the US so the goal is how do we grow not only support the local customer in the US but also grow into a strategic location where a lot of the new technology, new demand, new requirement can be collected from the US customers and then make the company better and make a better product. So I've been here for almost 20 years starting as a uh, FAE engineer in supporting the customer. Now we have a full grown team that has enterprise validation. We have a corporate marketing, cto, office business development and just you know, making day to day decisions, benefiting the customer, the headquarter to develop a better product.

Speaker B: With you kind of coming straight in and you know you're having to start the, I can only imagine just the, your journey through it all like as you're building the company in more. On the, on this US side what are the key aspects that you've learned throughout? What were the key things that you've had over your two decades you would say Michael?

Speaker A: Uh, yeah, yeah I would say that you know it may sound very simple answer but I would always start with customer experience and starting as a FAE where every inquiry that comes to you is about feedback about your product, how to make it better, how to solve the problem for the customer. After you finish a project you want to promote another product to the same customer. So you kind of go into a very you know, reinforcement, kind of feedback loop of what the customer tells you and how to make your product better. But along the way you found out that ah, okay, now we got one customer here, now let's grow to 10 customer, 20 customer and I'm shorthanded start to grow the team Fae team and then I guess the next cycle will be how do I create a team of experts that really can collect all the requirements and make decision on uh, what product is best to be developed and then collaborate with the headquarter engineering and make the product out. So just kind of starting with the customer experience and overall grow into a organization where it's end to end from getting the requirements, develop the product and be able to validate the product and then finally how to market a product. As you know we started with a company that we do a lot of private labeling. We don't have a brand, we had a brand behind the brand. But how do we now grow into a, a brand company, you know, later on in the stage of the company

Speaker B: and with that as you know coach, you have to be able to understand more on the requirements and uh, obviously sell you know, through the time that you've had and as you've grown into more on that leadership, bringing in the right team, you know, getting to make sure that you're, making sure that you're positioning the product the right way. What are the, what would you say like the biggest shifts that you've seen during the time as you've progressed and gotten you know, the company uh, with within uh the US region to a good state. What are the demands and the biggest shifts that you've seen thus far in your leadership?

Speaker A: Yeah, I would say the, the, the very first shift is going from a retail or retail kind of mentality going into an uh, OEM mentality. That is a big shift for the company. And the biggest difference is when you develop product for retail there's more opportunity to be flexible, to change things, to change the firmware, to update the firmware. But when you go to the OEM side everything has to be more fixed, more rigid, more careful about designing and validating. So that's the big shift. The other big shift for Fison that uh, is really changes the company is how do we go from a follower to a leader. Right. You have to understand in this storage world there are about seven company that makes the name flash today, right. They, they go end to end, right. They have, from you know, they have the memory, some of them do controller integration and they have a brand to the product. So for Fison who has our uh, core technology is the controller. But we are relying on other people's name at the beginning of the company. We are trying to see how do we catch up to the leader which happened to be a name makers right that time our number it was Samsung. So in 2019 that all changed. We had an opportunity to do a product gen 4 where nobody is there to jump into it because the gen 4 market is unproven at that point. But we took a chance with amd. We developed a product ahead of everyone else, have an uh, 18 month leadership to Samsung at that time or whoever followed and, and be the only player for about 18 months. So that mentality changed entire philosophy. How we go about develop a new product and how we want to lead the market. Right. So I think that's probably the biggest shift that we have personal experience.

Speaker B: So once you've gotten to that place once, you know, because that's the key part in always doing competitive analysis of your competitors. And once you've brought that forward to it, where's the part to where you have to make sure that you present that to the market and to your current partners and to upsell to get into that next stage. What, what have you seen that's really worked well for folks that are listening in other leaders, what have you learned that you would Give back to help with that.

Speaker A: Yeah, that's an excellent question. And, and uh, from our experience is when you want to be the first to market, you have to accept that there will be parts of the design that will not be perfect. Right. And also so, so you have to take a very good, I would say educated design and estimation. Sometimes customer doesn't know what they want actually. Right. So you just have to come out with a solid foundation that okay, this is what the market should be on. Um, the first market knowing that there are some parts that may not be perfect. Right. Nobody's going to tell you do exactly this because your competitor is this. So be better. Right. So there's some risks you have to take but you make your uh, best educator guess. You get the product out in the market and when people catches you in about six months you already working on the improvement of your first generation where you get actual feedback from the customer. You also learn about what your competitor is doing for the first generation and then you come back with the follow up design. So that's the winning formula that has worked with us. After Gen 4 we did it again in Gen 5 and we try to make it in Gen 6 again.

Speaker B: When I get your take on this as well is you hear a lot with how AI is coming forward. It's really, it's changing the dynamic and the whole paradigm shift of throughout it all. And do you feel that a lot of uh, leaders and other folks are having the technology lead the business versus the business leading the technology? Because, and I just want to get your thoughts because of there's new features and there's new, you know, cloud with MCP protocols and being able to connect to different applications. Do you feel that it's kind of been missing the goalpost of where is the actual pain point? And then utilizing AI and using the technology to fix that or to make it more streamlined. What are your thoughts on it? I'm curious.

Speaker A: Yeah. Well first of all we kind of had this discussion prior to the call. I think this, Yeah, I think this is a time where everyone, including you and me, right, if we have the luxury, we should just pause the work that we're doing and just really relearn how we do the things that we used to do in the past with the help of the AI. What I have observed is that uh, they have turned into a, you know a uh, lot of the execution heavy tasks into something that is quick and simple and give you quick feedback. Right. So I would, I would say that you know me as a leader I try to you know, be open minded and always kind of and bring in young talents. Who is getting into this AI world with Fresh Mind, no prior experience. How do we kind of go back to the ground zero and said what is the best way with the help of the AI that can do this much more efficiently. So that's what I always kind of remind myself. Where can I learn that experience? Especially for the people that is coming forward. Fresh Mind knows much more about the uh, AI. Where can I learn from the younger generation? So I think this is very important for the leadership to be able to say hey, my past experience may not be applicable right now.

Speaker B: I think it goes into how are we creating a new culture that's driven AI native thinking that can be able to excel and complement as you know the new generations added on with the skill set. So I think that's. Do you think that's really the goal point for us is in the leadership side to how do we implement that for the, for the younger folks that are wanting to get into the workforce?

Speaker A: Uh, yes, definitely pay attention to you know, how to make it better. Why, why not today? Right. And then when you kind of have that mindset then you're starting to trigger you okay, what are the available tool that allows me to have a next milestone discussion tomorrow instead of two weeks later.

Speaker B: And then we were talking a little while back of you know now is a big discussion and this is what Fison is, is really doing with the, the new ssd uh for the AI chip is you're hearing a lot of repatriation like of uh people. The paradigm shifting back to people wanting to move away from the cloud now because of either compliance or data sovereignty. You know we're kind of seeing it go back into somewhat of an on prem side. So for this discussion on that you know where, where are the, the leaders? And now with this bare metal on the SSD cars for, for Fizen, what do you see really the, the you know the I should. How do I the market on um, how it's trending and so far what are your thinkings of it for the next few years?

Speaker A: Yeah, so a very good question. I, I think you have seen that uh the whole world is you know being remade. We're in a new industry evolution where every, every big companies, AI companies and big CSB they are building out AI data centers right in the you know, 23, 24. It's very much about how do I create that large knowledge base, the uh, big large language model. Right. That allows that get all the human knowledge and data and latest information into a big brain that allows, you know, the, that allows critical thinking, decision making and all that. But then in this year and then moving forward to the next five to ten year is really about how to use those knowledge into something that we could apply from the day to day basis. Right. So what have we seen that for certain technology to be very widely valuable. Right. Of course a lot of things can be done on the cloud. Right. But with the AI especially there is a interaction now just beyond computers. There's also physical AI that's interacting the robots and with so much information that is available that needs to make a quick decisions. We believe that uh, we need to have an on prem strategy that allows every devices or every robots or every computer to be able to process, uh, interact with the human beings with the AI information. And then the other thing is, uh, especially with the globalization, with the geopolitical atmosphere, we think that every country and every organization should have their AI, uh, being controlled within their premise. Right. So that makes the AI has to be more private, more on prem. And when that safeguard is in place, there will be much more usage across the board for AI. And that actually would regenerate new demands on the cloud. Right. So I think that, you know, as you see that the smartphone, when it's created in 2007, it really exploded the whole Internet era because of the physical on prem devices. Right. That everyone has access to. Yeah.

Speaker B: You know, I think it goes back to. I've had to kind of go back into my foundations of how it work, how it works. Um, I'm more on the bare metal of how are we when we put AI into it. You kind of how we have to look at it in kind of a 01 term of the IOPS, the input and output of the SSD, of how it's pulling in the, the latency of the networking up that goes in. Do you feel that it's. Those are those core structures and that's what Bison is really bringing into. On bringing the security, as you said, you know, more on the governance and the risk assessments of it. When people are moving to the core lot, uh, moving back on prem or utilizing more of a hybrid approach of keeping maybe mission critical applications on, you know, their. On their. On prem. You know, what are your, what are your thoughts on that?

Speaker A: Yeah, so maybe I'll go back to a little bit of how inference works. Right. Right now in order to keep. Have your device on prem and be able to process a large language model right there needs to be the data and a lot of the data you were talking they call the context memory right. It has to store in a KV cache. Right And KV cache today is typically stored in a small density expensive volatile memory. It's called dram. Okay that's today right so and then typically the larger density much more cost effective. It's called NAND Flash. So what we have done in a ah NAND Flash SSD provider we were thinking how do we combine both into a large pool so how can I take a large SSD SSD to be part of the KB cache memory so that devices can actually oversight and start to process a large model on prem. Everything on prem. Right. So that's the technology we're bringing to the world today is that you know using what we do best on the SSD and combined with the DRAM to make many devices that's able to. One of the most popular killer apps right now is OpenCloud. How do we run OpenCloud on premise so that it's more secure and also more cost effective while also give the flexibility for people when there's a more complex task. Right. Have access to the cloud. So this hybrid model is our kind of next level innovation to take from being a commodity SSD provider into uh integration of a system that allows both on prem and on cloud assets for the AI.

Speaker B: I'm glad you brought it up on openclaw in your perspective of it with being with enterprises because when you first heard of Open call the big part of was obviously it did so much on the consumer what people really wanted AI to do but there was always the security factor of prompt injection of people coming in and you know taking the was a double edged sword for it you know but as obviously it was one of the biggest downloads within get history. It was big uh, and GitHub uh, it was pulling through. So the market was obviously this is what the people wanted. So when you saw that just on yourself on your personal side and how to integrate this in. But you understood the security aspects and the naysayers how is it from when those pieces come up how do you balance and how do you leverage the trade offs of this is. But this is what the market wants. How do you as a you know, C suite or for a leader how do you make that assessment and find the alternatives? Does that make sense?

Speaker A: Yeah, yeah I think definitely it's one of the most viral piece of software that's ever happened in a GitHub and you know we, it's, it's you know you know Jensen Huang mentioned about agentic AI is coming and that's the realization of agentic AI that's really happened right in front of us. Right. So with any gene technology there's always security risks, right. So we understand especially it's doing a lot of the. This uh, function is trying to operate your local machine as if you were operating it. Right. So the companies are putting safeguards on where it can have access to locally on the data. Right. But most of the so called interaction with the brain, right, with a brand I see that when it's access. Well, many companies are focused on how to safeguard or create boundary within your computer of what data can be accessed to. They forgot that a lot of data is exchanging with the cloud. So how do we prevent that? You have to have the model, the brain on the print on premise. But you have another problem, right? Most of the machine, most of the PC does not have a capability to have a local large language model store right on your computer because it requires a lot of dram. And so how we come in play is that how do we use SSD as DRAM so that your local machine now all of a sudden uh, possess the capability of a large language model does not require to own on the cloud. All your tokens are all paid for because it's all on prem and by the same time have flexibility to access the cloud while it's absolutely needed.

Speaker B: So to that point for organizations really trying to keep you know they're having it the hub within bring local for security and the sovereignty. What are the biggest trade offs from your experience, hardware trade offs. They need to understand around the ssd, the dram, you know, the latency and the cost.

Speaker A: Yeah, I would say that the biggest trade off is people want to be hassle free. They will just sign up for the cloud, right. They just go to the cloud, they pay a fee, recurring fee and in the past it works, right? Company have subscribed to 10 different software that does certain things by certain software, certain things by certain software. AgentIC AI changed through the whole landscape, right. It's something that you could do ten things now, right. But what people do not realize is that well they can easily have this call experience. Everything is all paid for, it's natural language. But wait until you get the bill, right? The token cost, right. That's the new currency, the token cost, right. So the trade off will be how much you would invest. You don't invest in your physical infrastructure and just rely 100% on the cloud and that the more you use it, the more you have to rely on it. So your destiny or budget is tied to the cloud. That's one trade off people are making. The other trade off is that okay. People say okay, I'm going to do it on um prem. I'm just going to spend on the infrastructure. But with the current hardware is you just need to keep buying more dram, more gpu. And you buy the GPU not because of GPU computing, it's because of the, the memory. Right. Because now memory officially become a bottleneck. So that's a trade off they are making is spending way too much on the hardware just for the memory. So that's why it's timely that using uh, SSD as a memory concept which is also endorsed by at the ces. Right. They call it the uh, inference context memory storage. That is going to be a game changer. How do we use a storage now become part of the memory for the AI inference.

Speaker B: I really liked how you brought that because I think a lot of even small to medium businesses they think of on the cloud, but I don't and I'm curious to get your thoughts. There's not really a lot of talk into going on prem because you think of the capex like I need to spend a large amount of money to bring in the racks of servers or bring in so much time. Do you think that we should be talking about that more?

Speaker A: Yes. So the world that we envision is that uh, why do you need a server to run a inference for the organization. And the only way to do that is how do we redesign hardware architecture and leverage the power of the SSD as a cache. Right. So very much that we are able to have many demonstration of compacting a record server as opposed you know, compacting a record server. The cost that requires to build that server into a single workstation with couple GPUs. With the trade off of time. Right. There is a trade off of time. It's going to be slower but at the same time. Right. It will kind of change how people think about getting into owning your own devices where you could process everything locally. Right. So that's the pricing in the past history we see robot arms, we see electric cars. When the pricing gets to a certain point you will just create a new generation of use case. Right on prem that uh, people do not realize today.

Speaker B: And I do agree with that. So you know with you Michael, like what is your vision, you know that you've been able to leave Fison, what is your goal for, for the US side on the operation for the next year or two would say oh yeah,

Speaker A: we very much believe that uh, Fison needs to be moving upward on the upstream. Meaning we are not only looking at just on the storage as a component but we have to look at the system with the software with it to create a value add ah, system for the end customer. Right. So that's how we envision. We want to recreate the computer architecture to create more you know, kind of a, A, a total solution devices for the end customer to you know, better use AI processing and also you know just a storage solution overall.

Speaker B: You know. And, and with that I'm, I'm curious on having it because that's, that's the future of it and having it to where as we move forward with AI native and you know new, new companies are coming forward and utilizing a brand new insightful ideas and creatives you know with you from, from you starting off as you said, you know, you came from, came into the US and you've had to really grow your, your your own hero story really to get to where you are now with Fison. You know and I, that's kind of a question I was going to ask earlier in the, in the interview is what have you for folks that are wanting to get forward to get to the next stage of that small and more getting into that medium size or me coming from that medium to that enterprise side more on that human level on, on the thinking of it. What have you learned now? And having to deal with now with Bison being a uh, billion dollar corporation, having to deal with more on the international components. What are the things that you've seen that really helped you in your career?

Speaker A: Pay attention to how the market is evolving when that there's different stages where you know, in 2000 there's a lot a bunch of removable devices getting into the smartphone and then later on 2013 with the Ultrabook where it requires SSD and then all the way to you know, 2020, you know the enterprise pay attention to where the market is evolving and where the margin of the product is moving into have a strategy. So that's one part. The other part is a global strategy. Right. You know, a formula that works in the US may not work in China, may not work in India, may not work in Europe. Right. I already see that Europe there's a lot of new cloud company and they're all the place they're small and medium. They want to have their own on premise server. Even though there could be just one giant company that support the Entire European. So it's a different dynamic within this region. And for example China and India, we decided to partner with the local partner to use local to support the local brand. Right. Instead of forcing a US or Taiwanese brand into those region. Right. So I think just observe the market dynamics and also geopolitical kind of dynamics as well and know where your strategic supplier is all the time. You know, being a, a, a, a system integrator that uh, had to rely on strategic supplier on the memory side. Those are the element. I think it will help a company evolve into a billion dollar company.

Speaker B: I really love that because I think it's, you get, you don't really take that assessment of, of where I can put my company into, you know, another uh, sphere, another culture. You're, you're saying, you know, really understand locally where they're at, what, you know, take away the compliance or security, what is it that's really needed? Is that what you're saying? And then bring in how do we adjust and fit to meet their needs? Is that what you're saying?

Speaker A: Yes, yes, that's right.

Speaker B: I always ask this, Michael, as we're coming in for AI really shifting, we're seeing so much, you know, it's continuously getting better and more innovative. So if you had a magic wand, you're seeing how, how the market is going in the next 10 years, or, you know, which what would you love to accomplish? Whether it's on the business side or whether it's for your own personal use, what would you love to see? AI

Speaker A: I think that what I would very much want AI to do is not to remove all my creativity. Right. What's so convenient is that when you ask for something, it create like 5 different version for you and you just pivot.

Speaker B: Right.

Speaker A: Your brain stop working. The creative side of things start to be, okay, you give me five answer, I'll just choose the best. Right. So I would say that if there's a way, a metric one that will allow to kind of extract the elements of my creativity to create something that is truly personalized and not uh, generic. Right. I, I think that's the experience that I will be looking for because at the end of the day, people in China, people in Taiwan, people in us, they all assess all the same brain, the AI. Right. But how do we make it personalized for everyone? It requires a lot of work to be done on the AI development side to help and assist asking the right questions and extracting the data to be able to make the AI personality different for everyone.

Speaker B: You're Kind of seeing it because all of our everything is data. I was even seeing with Claude. Now that it can connect now it's a beta. Now that it can connect to my chart, it can connect to all of my, you know, oura ring, you know, now that it's able to track my heart rate. So it has all our lives are now is just data and then now it's metadata, just breaking it down. So I love that to have it to where now we can get to a full personalized way and amplifying how my creativity is, how I am. You know, I think that's where we're headed in that time. You think?

Speaker A: Right. And I'm um, just very much looking forward, you know where AI is taking us to so called level one unique personality. How do we even improve beyond that?

Speaker B: Right.

Speaker A: And have more kind of stylized personality and you know, with the help of the AI. Right. Yeah. So I think that's what I'm looking very much forward to.

Speaker B: You know I've really enjoyed this conversation Michael. I think really getting your perspective really with you know, how the market's moving and just honestly just from your expertise of going from the ground up, your journey from to where you are now and where you started. So you know, for folks that the, the listeners having in you know where can they be able to 1 follow your journey and then 2 for companies or for the leadership that's listening in on this, on um the interview, you know what is the key factor that you should, what you would ask them or what they need to think about to if the fisons for your new SSD card.

Speaker A: Right.

Speaker B: For their, for their company and for where their. Their their market.

Speaker A: I would invite you guys to go to our company website um www uh.fison t h I S O N where we have the pool portfolio of our product offering. Is there a little bit about also our company. And then to see our brand new enterprise product line go to haskari.com that's the name of our enterprise brand Haskari P A S C A R I uh dot com. And the technology I mentioned to you earlier about using SSD as a cache memory, we call it adaptive. So that's all in the website that you could see.

Speaker B: And then for the audience listening, we'll make sure to have that down in the description so you can easily be able to click on and be able

Speaker A: to see Michael Wu. Look for Michael Wu Faizon. You should be able to find it.

Speaker B: Perfect. Perfect. And then so for the folks listening in, thank you so much for tuning in to this conversation. And then, Michael, I, uh, have one last question. You know, I've been having it for, for the series so far. For your next question that you would have to ask for, you know, a leader or anyone else, whether it's a founder, what would you be the keep in mind for an advice or what they should ask themselves as they go forward for the next guest.

Speaker A: So you asking what my advice is for the next leader?

Speaker B: Yeah. What question they should really think about, like whatever is in their business or what just in leadership or just whichever. Like what question they should ask themselves that's helped you on it? Just on a simple question.

Speaker A: Oh, okay. That's a good question. Oh, you're good. Yeah, yeah, yeah.

Speaker B: Okay, I'll keep in mind. Well, you can reach out to me later and then I'll. I'll put it in for them to think of.

Speaker A: Yeah. Do I still. It's okay to say it now or.

Speaker B: Yeah, of course.

Speaker A: Oh, okay. Because you're gonna edit it anyway. Right? Okay. So the advice I would give to my fellow leader is that dream big, know who to hire that can dream to have the same big dream as you. Right. I think that the execution side of things, a lot of it can be facilitated with the help of the AI but it's the people that has the same vision that uh, could make the same decision or have the, the make certain critical decision for your organization, uh, that is going to change your organization.

Speaker B: I think that's really a good point because it brought back to what you were saying beforehand, Michael, on, you know, how you brought your team up in the US side and that bringing together on the right of people to be able to really figure out how do we get forward and push past, you know, Samsung, as you said back in 2019. So, Michael, thank you so much for tuning in. For the folks listening in, we'll catch you next time on the next episode of the AI Advantage.

Speaker A: Thank you. All right, thank you.

Speaker B: Thanks for tuning in to the AI Advantage. This episode sparked an idea. Share it with your network.

Speaker A: Subscribe for more.

Speaker B: No fluff conversations on how smart tech drives real outcomes. And if you're ready to future proof your business, let's connect@salonox.net.