In The Arena by TechArena - Inside MiTAC’s Turnkey AI Infrastructure for Data Centers

Episode Date: June 23, 2026

In this Data Insights episode, co-hosts Allyson Klein and Jeniece Wnorowski chat with guest Raymond Huang of MiTAC to explore how AI infrastructure is rapidly evolving to meet the demands of modern workloads.

Welcome to Tech Arena, featuring authentic discussions between tech's leading innovators and our host, Allison Klein. Now, let's step into the arena. Welcome in the arena. My name's Allison Klein. Today is a Data Insights episode, so I'm here with Janice Norowski. Welcome, Janice. Hi, Allison. Thank you very much. So, Janice, I know that we've got a really exciting topic today with one of the foundational players in the Data Center. Tell me who you, you brought with you and what is our topic? Yes, very foundational indeed. Today we had Raymond one, so excited to talk with you, Raymond, learn a little bit about what you guys are up to and dive into some of the key segments that you're targeting with Mitak. So welcome to the program.

Yeah, thank you having me. Yeah, today we'll be a love to join discuss how we enable the AI industry providing the turnkey solution to the market. So Raymond, do you want to just talk a little bit about your role at MyTac and how you are working to bring data center capabilities to fuel AI infrastructure? Sure. My name is Raymonda Wong. So I serve as a general manager in MyTac Computing. I need a function team overseas our strategy execution across the AI infrastructure. We're working closely with our manufacturing, ecosystem partner, engineering to bring scalable, high-performance solution to the market. And basically it's all about the turnkey AI infrastructure solution. And ultimately, my role and our mission as a company is to help customer accelerate time

to AI reduce the deployment risk and scale with confidence as their demand grows. Amazing. So Raymond, my tech is positioning itself as a full AI infrastructure stack and provider. What is driving this evolution beyond traditional server manufacturing, you think? Mytech evolution is driven by few things. First, I think, is the technical necessity that relates to the power, cooling, and density. And it's also driven by ignited pressure. People are talking about margin and also differentiation. And the third things, I think, is also driven by the customer demand. They all want something faster to deploy turnkey deployment. For example, we are working on the Diamond Cool server, direct-to-chip liquid cooling, red-level power distribution,

pre-integrated infrastructure. And the AI factory is also something as a new category we're working on. So basically, my tech is a line around modular AI fluster as a building block. And the strategic that is, We don't just build server. We deliver AI capacity. Now, you have an end-to-end rack design, and you have integration capabilities that are clearly a differentiator in the market. How does this vertical approach

from design to manufacturing help accelerate deployment and customization for your customers? Yeah, so from my tech perspective, we provide the pre-integration, which eliminates the slowest step, which is the system bringed up. In the traditional model, the customer have to stitch together servers and GPU from whether it's Nvidia, AMD, together with networking fabrics, power distribution, and also dealing with cooling systems.

Now, that all-integration phase is also considered where months disappear or where the failure often showed up. What we do is the fully integrated rack-level solution, which components are pre-valided. together, firmware or networking and thermals are tuned as a system. So now Rex arrived effectively, deployed ready. So to customer, this is a time to market, right? Then this is where they ship from one-stop integration, now to days of weeks just for installation inside, and they can be able to scale a scale.

Second thing is also for design for manufacturing, which means now is faster scaling. Because MITAC we control both design and manufacturing, so we be able to standardize the modular rack architecture or preview repeatable AI cluster units, which can quickly ramp up production, you know, at a design which is validated. Now customer can scale from pilot to full cluster without redesigned the system each time.

So this is really something with innovated, And then most importantly, I think, is what we type coupling or cooling, power, and compute. With vertical integration, we'd be able to modify the configuration with a non- or validated power design envelope, whether it's 2.8, 415, 480 volt, or even the upcoming 800-volt DC. So this is something we consider as the turnkey for the customer. So Raymond, to help customers overcome GPU density, power, cooling, storage, of course, and just overall complexity challenges, could you introduce a little bit more about how MITAC turnkey solutions address all these things? Sure. So I think how we address the turnkey solution for customers first start from the GPU density, right? So that means hacking more compute without breaking the system.

Air cluster today are constrained by how tightly you can pack GPU from vendors like M Media or AMD without heating the thermal or power ceilings. MyTax turnkey approach is we provide pre-designed, high-density racks, optimized for specific GPU configuration. For example, a GPU system, or 32 GPU from air cool, per rack all the way to up to 96 GPU per liquid cooling rack system. So instead of customer figuring out how many GPU per rack, we can safely run my tech deliver the fully validated density.

So this is saving the much time and also effort. And now they be able to have higher usable GPU density without trial and error theory. So the power delivery is something we also are very focusing on. So we make the high density actually usable. Today you see GPU Ratt easily 30 kilowatts all the way to 130 kilowatts per REC. So the power isn't just about capacity. It's about stability and efficiency. So one might take turnkey solution is we provide a rec level power distribution for engineers, for AI workloads,

and it's also redundant, high efficiency power architecture. So all these, you know, we matched power profile aligned with the GPU or CPU configuration. So bottom line, the outcome is customer now avoid the costly issues like power instability or derating or underutilized tracks. So we solved all those kind of bottleneck for customers at times. Now, I know that one of the things that we talk about on this show a lot is that as AI infrastructure becomes where data intensive, performance of those applications is increasingly being tied to how compute and storage work together. How is MITAC working with partners like Solidime to integrate high-performance SSDs into its systems and support data-intensive AI workloads?

One of the partnership that we have is together with Solidine is also on a Turkey solution is MITAC, DDN, and Solidine. We kind of solve the data bottleneck, and we enable the high-performance, low latency storage solution. Basically, even the best GPU cluster today is useless, right? If the data cannot keep it, what we provide is the optimized AI workload to pair with our server together with Solidine, I think is the try of the D7 SP10, MEME DRI, so we be able to provide something so-called a turnkey solution. This kind of eliminated IOTLINX, and at AI training, it requires a massive

data set. So it's only come strand on by the bottleneck of the IO. But with the solution that we put up together, MyTech Server, Sautodine, NMEDRI, and DDN software, we keep a GPU fully utilized and avoid idle compute due to the store storage. So the result now, enterprise customer be able to get the performance they pay for. And I think there's something very important. Yeah, I agree. I think collaboration is key. And, you you mentioned DDN being one of the folks that you guys collaborate really closely with. And keeping on the topic of collaboration, can you tell us Raymond a little bit more about your work with Rafa and maybe other Neoclouds and how you guys focus on kind of orchestration

and how do these collaborations with Neocloud simplify your AI deployment for customers? I think for Rofé partnership, it basically from the solution perspective, that they simplify the orchestration and operation. AI infrastructure without the orchestration quickly becomes unmanageable, especially when you are dealing with a multi-node GPU cluster, continualized workload, or multi-teams sharing resources. I think that's where Rafi fit.

And what they bring up to the table is managed Bernadis and GPU orchestration, also multi-cluster lifecycle management, as well as the self-service environment for the AI ML teams. So now customer be able to have faster cluster bring up. Instead of manually configure Kubernetes crossbrex or this cluster, Rockby provides a call the pre-integrated cluster template and automated provisioning align with the MITEC hardware. So customer go from Power-on to now usable much faster, basically.

And from AI environment, it typically often involve data scientists, ML engineer, or platform teams. What we provide together is enable workload isolation, GPU sharing, and scheduling also the road-based assets. So customer be able to avoid building complex internal tooling and now just to manage the assets. I think that's something also very important. And one last thing is consistent operation at scale, right? So basically whether it's one rack, five racks, or 500 racks now the policies, the overall updates that monitoring can be all centrally managed, and this is significantly reduced operational complexity to the customer.

Together with Mitag, Raffa and also DDN, Solidine, we provide an end-to-end AI platform solution to the market. That's help customer be able to scale faster and quicker. How is MITAC's strong partnership with AMD enabling the creation of next-gen AI systems from their epic processors and instinct MI-355XGPUs to rack-scale liquid-cooled solutions? And how do you ensure the data pipeline,

including storage, keeps pace with that compute performance? But my tech, we had a very long-term partnership with AMD. Then it's fundamentally about the type co-engineering across the full AI stack. So from the silicon, the CPU, GPU, to the system design, networking, and cooling. And that's what enables us to move beyond the server, now into true rack-scale AI infrastructure. So I always say it about it's the AAA solution nowadays. AMD CPU, AMD GPU, and AMD MIIC. So it's the Epic CPU with the Instinct GPU.

Now in our 4U server design, for example, the liquid cool or the AU design, the airpool. So from the system design perspective, we have the DLCs, the red liquid cooling at the no level. At the red scale integration, we also provide not just a single rat, but a pot or cluster. So whether it's a 48 U, 52U, liquid cooling from all the way 96 GPU, right, MI355XGPU. And we also provide so-called plot cluster building block. Easily 256 GPU, 512, 1024, or 2048 GPU as a cluster. So large AI model, it requires the massive HBM memory. So I think that there's a 10-turbide per rack which enable that, you know, help out the inference or training faster.

So that's the efficiency there at scale and also the faster deployment cycle. So pre-integrated rec system means the scale and scale makes cluster-ready deployment, basically. So what about your liquid cold? I think it's a G48-26-Z-5 system, which is designed specifically for state. scalability and efficiency. And how does this particular system manage and power, the density, and thermal performance as workloads continue to grow? Our 4U system, the G of 4826.35,

is the dual AMD Epic, 9,005 CPU to pair with the 8 Instinct MI355 GPU. It's a support up to 6 terabyte DDIL5 memory in just the volume fun factor. So this is designed for extreme power density and also is the DLC-based liquid cooling, which enable and removing basically up to 95% generated heat via liquid-cooled loops.

And you're also allowing a much higher watt-tribute ratio than traditional air-cool system. So it support very high-dense GPU packing without the thermal interference. So without the DLC, this kind of, the density wouldn't be even thermally impregal. So basically, the system is no performance throttling, even during the long AI training runs. And it's all because of the liquid cool.

So in practice, the system sustain the peak GPU utilization instead of the cycling down due to the heat. This is critical to the large language model training or HPC simulation, multi-AI workload jobs. So efficient data flow equals to less wasted power, which also equals less heat. Now, I think that the other thing that I've noticed about my talk is that you've got incredible leadership and sustainability for AI data centers. Can you briefly introduce a couple of key technologies, including Akasha's diamond cooling solution and Tenomea's containerized AI factory solutions and how they relate to your sustainability objectives? Sure. So we are very excited to have a few our own consider innovation and also innovative solution as a turnkey to the market. One thing is the partnership with the Akash. So the diamond is the world's most thermally conducted material with unmatched thermal conductivity. It's 2,000 watt per meter Kelvin. So it's more than five times greater than industrial standard copper.

So Diamond Cool server solution basically is designed to enhance the GPU performance. We observe up to 22% higher slots per watt. We also targeting to improve the energy and capital efficiency, which also enable customer to scale and height density AI infrastructure. So the Diamond Cool is actually paired with AMD MI 350 GPU. So we observe it consistently remain up. to 10 Celsius degree cooler than the stock GPU and it still be able to remain throttle-free in the various hot BN environment. So basically the solution is what we are putting together

with Akash for using the diamond cool technology. So ultimately as you can imagine, less energy spend on cooling and higher compute per watt. So that enabled higher density without increasing the energy footprint. The other solution we are also pointing together is something for the AI factory. It's partnered with the Tonomia. So basically it is also leveraged MyTex server hardware and RACs to pair with their solution is so-called containerized solution, GPU racks, storage, battery, liquid cooling, all in one.

So this is what we are providing turnkey solution to allow customers, to deploy much quicker and be able to enjoy something in a timely manner compared to typical deployment, the data center, customer had to wait for the data center build up to be ready, and it takes forever to have the power from, for example, the PGA to enable it also take forever. Right now we provide so-called containerized solution that be able to quit the market, to deploy, to be able to plug in, and just let it all power out. So this is something we think is addressed the new energy usage case. And all these low carbon on-site renewable energy helps us to deploy something like this.

This is something we bring up to the market as a sustainable solution and a turnkey solution. So with the explosive growth in the AI market drive me demand, what are my tax expansion plans, especially in the United States, around production capacity to kind of meet needs for localized? scalable AI infrastructure? Sure. I think from MyTech, we are focused on localize the AI supply and global multi-region manufacturing strategy.

So MyTech, we are scaling aggressively across capacity, geographic, and also deployment models to meet the surging AI infrastructure demand, especially by strengthening our North America manufacturer footprint. we actually build in America, yeah, infrastructure. That's one of our major focus to combine local build, local fuel, local support,

those kind of demands. So by doing that, it's also reduced the lead time and logistic risk and potentially the tariff exposure. So aligned with the CSP, Neo-Cloud, enterprise demand for localized supply chains.

Today, we had a few thousand rack monthly capacity to support our customer. So I think the local resource, local fulfill, local build in this region is crucial to our customer's success, particularly for those customers. They want to deploy globally in every single country or different region. So we'll be able to fulfill that locally and to each customer's demand. Now looking at Had, how do you see AI data center design changing over the next few years? And where does my tech want to lead that transition. Power becomes primary constraint. I think that's one thing to consider. Operators are moving on-side generation, hybrid energy, for example, using gas or renewable energy, and even energy islands,

right? So new electrical architecture like high voltage DC improved the efficiency and also reduce the conversion losses. So the data center now becoming more actively participates. in this energy ecosystem, not just a consumer. The other trending things I see is the cooling shifting from air to liquid everywhere. So air cooling is heating its limits already. And liquid cooling, direct to chip, even immersion, is becoming mainstream moving forward. And events technology like microfluorloids and evaporated membranes are emerging to handle extreme heat loads. So the design and replication is the cooling is no longer a facility at-on.

It's a co-design with the silicon and also REC, which we are very focused on collaborating with the partner, too. And one last trend, I believe, is modular, scalable, cluster first, this kind of architecture. The AI data center upviewed as modular GPU cluster that scale incrementally. So we integrated RAC system is something very crucial to their success. So the speed of deployment and repeatable building block matter more than just best-buck builds. So there's something what we see as trending.

And of course, what my tech we are aiming to do is deliver of reference architecture that actually be able to work at scale. And also we solve the cooling plus the density challenge. faster than industry. And one more thing is provide a modular system that softened deployment cycle. I believe in the next three to five years, we will expect AI data center design to cover some liquid cooling and very high-dense cluster. Each rack may be 100 kilowatt all the way to one megawatt per wreck. The hybrid on-site power generation is also what we see is upcoming.

And most important, I think the standardized modular built up like Lego blocks or AI capacity, this is an extremely critical or point to speak needs. So this is something I consider the trending things and what my tech is also very focusing on to enable market. I think Raymond, this has been an amazing conversation. Our listeners, as Alison knows, will want to get information and learn more about MyTac. where can they go to connect with you and to just read up on your innovation? You can go to the MyTechcomputing.com,

which is our official website for more detail and just search me from LinkedIn at the way to reach out. And we welcome all the questions and collaboration. And we are very looking forward to working with industry. Thank you so much for being on, Raymond. It was such a pleasure. And Janice, that wraps another edition of data insights.

Thanks so much for the, the collaboration. Thank you, Alison. Thank you. Awesome, man. Thank you, guys. Thanks for joining Tech Arena. Subscribe and engage at our website,

techorina. All content is copyright by Techarena.