AI Cluster Power: Liquid Cooling and Exascale-Ready Solutions from MiTAC at SC25 - Six Five In The Booth

The evolution of AI clusters is clear: liquid cooling and rack-scale integration are becoming the new blueprint for exascale infrastructure.

‍

From SC25, host David Nicholson, Global Technology Advisor of The Futurum Group, is joined by MiTAC Computing Technology Raymond Huang, GM and VP of Sales & Business Development, for a conversation on MiTAC’s latest rack-scale AI clusters and liquid-cooled solutions for exascale and HPC applications. Find out how MiTAC’s vertically integrated approach is resulting in record-setting performance, sustainability, and operational flexibility.

‍

Key Takeaways Include:

‍

🔹Strategic Focus at SC25: MiTAC is spotlighting its high-density, liquid-cooled AI racks—featuring the debut of systems powered by AMD Instinct MI355X GPUs—to address the growing performance and efficiency demands of exascale workloads.

‍

🔹Differentiated Liquid Cooling: MiTAC’s liquid cooling solutions stand out due to their purpose-built, rack-scale thermal management, enhancing cooling efficiency and supporting the requirements of AI, HPC, and large language model clusters.

‍

🔹Roadmap for AI and HPC: The company is adapting its product strategy to the latest trends in AI and HPC, prioritizing scalable architecture, energy efficiency, and instant global deployment.

‍

🔹New Product Launches: Highlights from SC25 include the launch of exascale-ready, liquid-cooled racks and success stories in large-scale deployments that demonstrate real-world customer gains in efficiency and scaling.

‍

🔹Commitment to Sustainability: How MiTAC is focusing on sustainable infrastructure for AI and HPC, leveraging its end-to-end capabilities to support evolving customer needs in the data center.

‍

Learn more at MiTAC Computing Technology.
‍

Watch the full video at sixfivemedia.com, and be sure to subscribe to our YouTube channel so you never miss an episode.


Or listen to the audio here: 

David Nicholson:

Welcome to Six Five Media in the Booth. I'm Dave Nicholson, and we're here in St. Louis, Missouri for SC25. And I'm joined by Raymond Huang, General Manager and Vice President, Sales and Business Development at Mitac Computing. Raymond, welcome. Thank you for having me. So Raymond, let's start out by kind of introducing Mitac Computing for those who are not familiar with Mitac Computing. What is Mitac Computing all about? Sure.

‍

Raymond Huang:

Mitac is the ODM server highway manufacturer. We've been doing the business for more than 20, 30 years, and it's all about everything server related. We do everything, start from the motherboard, barebone system, fully integrated system, and also rack level design. And nowadays, it's not just air cool, but also liquid cool, server, rack, up to the cluster build level. plus the on-site service deployment.

‍

David Nicholson:

This isn't your first SC show, supercomputing show. How is 25 different for you and what sets this year apart?

‍

 Raymond Huang:

for you at Mitac. I think this year we are showcasing not just the hardware level design, we are showcasing a fully liquid-cooled based rack level and cluster build capability. And our capability not just include designing the hardware, we kind of give you a flexible fully integrated rack level And that's kind of something plug-in for a customer that needs tailor-made. It's not just, hey, here's the menu, choose one, two, three. It's more like, what do you need? I can be specifically tailor-made to meet your workloads for both HPC and AI.

‍

David Nicholson:

Yeah, and you just touched on it a little bit, but go a little bit deeper into what differentiates you from the other folks that are here. There are a lot of folks talking about cooling at SC25. It's a hot topic, if you will. But what really differentiates my type

‍

 Raymond Huang:

I think definitely. So liquid cooling, as you know, a lot of people get involved doing that. What differentiates ourselves is not just on the system level, rack level, is everything we do is homegrown, in-house, versus some others might be going to leverage some others to make it for them. For example, my tech today, we still design and manufacture for some other server, OEM, for example. So that kind of differentiates the capability in terms of position. We are the one behind the scene, making ledger for ourselves, but also for others. Second, what differentiates ourselves is we also have the OCP platform, which is Open Compute Project. That's not something everybody has, but MITEC also has that for many years, more than a decade. So we have both an EIN 19-inch rack, as well as open platform rack, OCP rack, 21-inch.

‍

David Nicholson:

And you've seen this evolve over the years. I imagine that you evolve your roadmap based upon changes that you're seeing in the market. What are you doing to meet those needs in terms of specifically large language models and HPC environments? How have things changed and how is MITAC evolving to meet those needs?

‍

 Raymond Huang:

So this year we have a very specific solution. It's called the AI solution. In particular building a cluster, we kind of have a mostly liquid-cooled, rack-level solution. And it's not just the rack level, it's the cluster build level. So meaning, customers can scale much quicker, much faster, and at speed. And that's about something my tech we are providing, not just the hardware design, but also the full scale of the deployment.

‍

David Nicholson:

Along that line, what are the key specific things that you're announcing here that you want people to know about?

‍

 Raymond Huang:

We are very excited to debut our very first AI solution liquid cool rack. It's called the MR1100 series. It supports up to 256 GPUs. We're also debuting our very first 4U, MGX-based, that's supporting NVIDIA RTX 6000 Pro, server-edition GPU, that kind of gives you more flexibility and also an option as alternative for either with large language model or inferencing or HPC AI workload.

‍

David Nicholson:

If you had a sort of a takeaway message when people come and visit you here at your booth, at the show, or just about MyTac in general, if I were to meet someone in the elevator and ask them, what do you know about MyTac? What would you like them to understand about MyTac?

‍

 Raymond Huang:

I will explain a few aspects about who we are, what we do, what we have. First, we are not just a server hardware manufacturer. We kind of design, manufacture, comprehensive from the system all the way to rack level, and that's just the hardware. On top of that, we have the global footprint in terms of North America, EMEA, and APAC. So that kind of meets customers' global deployment needs. It's not just one region or one country. For example, we have a customer that deployed to more than 300 parts worldwide, and he's all using our red-level solution built by Mitac, designed by Mitac. And on top of that, we also, I would say, on-site service is very critical to the customer nowadays. It's no longer just the hardware. It's also about who actually has the power capacity, production capacity to keep up the growing demands for the AI hype. Today, you see the large language model have super big demands everywhere here and there, but who actually have the power capacity for the production and also power capacity for those production, integration, testing, et cetera. I think that's also one thing that differentiates ourselves from others. And we also have a very big footprint locally here in the US, all in the Bay Area, including several production buildings, integration facilities, some even like a full liquid dedicated for liquid cooling only production environment. So I think those are something others have and can compare.

‍

David Nicholson:

Would you like to share any kind of real-world specific stories, or is that about as detailed as you want to get in terms of the Bay Area? Any customer real-world examples that you wanted to share?

‍

 Raymond Huang:

I would say from a very large CSP customer, We've been doing the business for more than two decades, all the way to today. Transition from just the motherboard, bare-bone system, all the way to the system and fully integrated racks. And it's for a very large scale, I would say. Give you an example. We're now a day in the Newark area, California. One facility, we built out more than 1,000 racks for this particular customer. 1,000 racks a month? A month. Okay. So at scale, you can imagine those kinds of volumes. That tells you something already. Yeah.

‍

David Nicholson:

So not just a point, not just a point solution, but comprehensive.

‍

 Raymond Huang:

It's not about the hardware. It's not about pricing. It's more like who actually has the full scope of support in terms of a functional team. And if you can fulfill their global deployment needs in different regions, different countries, and the logistic-wise, arrangement-wise, shipping-wise is very complex. including certification, for example, VAT or tax, the tariff, so who actually have those kinds of global footprint be able to mitigate the tariff or the tax or local just-in-time needs, those kind of, you know, demands. And I think Mitac We are in a very good position in having all those to meet our customer global needs.

‍

David Nicholson:

I want to go a little bit deeper on the variety of ways that you offer these solutions. And I'm also curious about how you go to market. What is your sort of entry point? into your customer environments? Are you designed into the ground floor buildup of a data center? Are you integrating yourself into existing data centers? So it's sort of a two part question. I'm curious about more details like some of the rack level solutions, for example, but then also this question of going to market. What can you share with us about that?

‍

 Raymond Huang:

From my perspective, I would say we cover pretty much a very diversified market as well as the customer base. For example, we have the typical channel distribution. distributor for our SI, VAR, reseller, et cetera. We also have the hyperscaler customer cloud service provider, which deploy custom-made ODM models. So I would say we have both off-the-shelf products as well as the ODM tailor-made business model. It just depends on the customer, their business model, specific needs, and then we can base on that to specifically fulfill.

‍

David Nicholson:

Raymond, any final thoughts on SC25 and Mitac Computing?

‍

 Raymond Huang:

I think today, this year, for SC25, our theme is the AI cluster power. Cool fast, scale faster is our mission. So we want to deliver the message to the customer, whichever needs the most critical compute and AI training needs using our server hardware design at a global deployment level and also at scale.

‍

David Nicholson:

Very cool stuff, pun intended. Raymond Huang from MiTech Computing, thank you so much for joining me. I'm Dave Nicholson from Six Five Media in the booth. Thanks for joining us and stay tuned for more content.

‍