AI factories are becoming the essential production facilities of the digital economy
In the ongoing transformation of artificial intelligence from a software curiosity into a physical industrial force, three companies — Pegatron, Together AI, and 5C — have formalized a partnership this week to construct what the industry now calls 'AI factories': purpose-built computing environments powered by NVIDIA's most advanced accelerators and anchored in American data centers. The collaboration reflects a deepening recognition that the ambitions of the AI era cannot be sustained by code alone, but require the same kind of deliberate, large-scale manufacturing and infrastructure investment that once defined the industrial revolutions of the past.
- The computational hunger of modern AI models has grown so vast that a single training run can demand thousands of GPUs working in unison, pushing companies to rethink infrastructure from the ground up.
- Pegatron, Together AI, and 5C are responding by dividing the challenge across their distinct strengths — manufacturing precision, software efficiency, and data center operations — to offer a complete, end-to-end deployment solution.
- Production is already underway, with NVIDIA GB300-based systems being built in Texas and liquid-cooled HGX B200 racks already installed and running at a 5C facility in Maryland.
- The partnership positions all three companies to capture demand across North America's rapidly expanding data center landscape, with new NVIDIA HGX B300-based systems already in development.
- Pegatron is set to present these solutions at Supercomputing 2025 in St. Louis this week, signaling that the race to industrialize AI infrastructure has moved well beyond the prototype stage.
Three companies with complementary expertise announced this week that they are joining forces to build and deploy next-generation AI computing infrastructure across American data centers. Pegatron, a Taiwanese server manufacturer, is partnering with Together AI, a cloud platform for AI development, and 5C, a data center operator, to scale production of specialized hardware capable of running the most demanding AI workloads.
The collaboration spans two core hardware platforms. One centers on NVIDIA's GB300 NVL72 accelerators integrated into Pegatron server systems being manufactured in Texas — a deliberate choice to anchor production domestically and serve American customers more quickly. The second pairs NVIDIA's HGX B200 accelerators with Pegatron's liquid-cooled rack systems, already deployed at a 5C data center in Maryland, alongside an air-cooled variant installed at the same facility.
What makes the arrangement distinctive is its division of labor. Pegatron contributes manufacturing precision and thermal management expertise. Together AI brings software capabilities for training and running AI models efficiently. 5C provides the physical data center environments and operational scale. Combined, the three companies say they can deliver a full-stack solution — from rack design and cooling integration to networking and live deployment.
The partnership reflects a broader industry shift toward what executives are calling 'AI factories': massive, purpose-built facilities designed to train and run AI models at scale. As models grow larger and more complex, the infrastructure supporting them has become as critical as the software itself. Pegatron's CTO Dr. James Shue framed the collaboration as a direct response to this reality, while Together AI's infrastructure lead emphasized that deployment at scale is now inseparable from the underlying hardware.
5C's CEO Jonathan Ahdoot praised Pegatron's speed and responsiveness, describing them as a true innovation partner rather than a conventional vendor. The announcement coincides with Pegatron's showcase at Supercomputing 2025 in St. Louis, and the company is already developing systems based on NVIDIA's next-generation HGX B300 accelerators — suggesting the partnership is built to grow alongside the hardware it depends on.
Three companies with complementary expertise announced a partnership this week to build and deploy the next generation of AI computing infrastructure across American data centers. Pegatron, a Taiwanese manufacturer of server systems, is joining forces with Together AI, a cloud platform for AI development, and 5C, a data center operator, to scale up production of specialized hardware designed to run the most demanding artificial intelligence workloads.
The collaboration centers on two specific hardware platforms. One uses NVIDIA's GB300 NVL72 accelerators, built into Pegatron's RA4802-72N2 server systems. These machines are being manufactured in Texas, a deliberate choice by Pegatron to anchor production in the United States and provide faster service to American customers. The second platform pairs NVIDIA's HGX B200 accelerators with Pegatron's AS401-2T0-8H1 liquid-cooled racks, which have already been deployed at a 5C data center in Maryland. A third system, the AS800-2T0-8H1, uses air cooling instead of liquid and has been installed at the same facility.
What distinguishes this arrangement is the division of labor. Pegatron brings manufacturing precision and expertise in thermal management—the challenge of keeping powerful chips cool without wasting energy. Together AI contributes software capabilities for training and running AI models efficiently. 5C operates the physical data centers where these systems live and manages the infrastructure at scale. Together, the three companies say they can deliver what they call an "L12 total solution capacity," meaning they can handle the full stack of requirements: designing the racks, integrating the cooling systems, networking the machines together, and deploying them in working data centers.
The timing reflects a broader shift in how companies think about artificial intelligence. Rather than treating AI as a software problem alone, the industry is increasingly focused on building what executives call "AI factories"—massive, purpose-built facilities designed to train and run AI models at scale. As AI models grow larger and more complex, the computational demands become staggering. A single training run for a cutting-edge model can require thousands of GPUs working in concert, which in turn demands specialized infrastructure, careful power management, and sophisticated cooling.
Dr. James Shue, Pegatron's chief technology officer, framed the partnership as a response to this shift. "AI factories are becoming the essential production facilities of the digital economy," he said, emphasizing that the collaboration combines Pegatron's manufacturing and thermal expertise with NVIDIA's accelerator technology and the software capabilities of Together AI. Mahadev Konar, who leads infrastructure engineering at Together AI, echoed the point: as AI applications grow more complex, the infrastructure supporting them becomes critical to whether companies can actually deploy and scale their systems.
Jonathan Ahdoot, CEO of 5C, highlighted the speed and responsiveness Pegatron has demonstrated in previous work. "From our first collaboration, Pegatron has proven to be more than a vendor—they're a true innovation partner," he said, noting that the company's ability to move quickly to market and support customers at scale made them essential to 5C's strategy. The partnership also signals confidence in NVIDIA's newest chips, particularly the GB300, which represents the company's latest generation of AI accelerators.
The announcement comes as Pegatron prepares to showcase its AI infrastructure solutions at Supercomputing 2025, the world's largest high-performance computing conference, being held in St. Louis this week. The company is also developing additional server systems based on NVIDIA's HGX B300 accelerators, suggesting the partnership will continue to expand as new hardware becomes available. For companies racing to build AI capabilities at scale, these kinds of integrated partnerships—combining hardware manufacturing, cloud platforms, and data center operations—are becoming the standard way to move from prototype to production.
Citas Notables
AI factories are becoming the essential production facilities of the digital economy. Through our partnership with Together AI and 5C, combined with NVIDIA AI infrastructure, we are delivering the thermal management expertise, manufacturing precision, and system integration capabilities required to build and operate these next-generation AI factories.— Dr. James Shue, SVP & CTO of Pegatron
From our first collaboration, Pegatron has proven to be more than a vendor—they're a true innovation partner. Their speed to market and responsive support have made them a key partner for delivering NVIDIA HGX B200 liquid-cooled racks at scale.— Jonathan Ahdoot, CEO of 5C
La Conversación del Hearth Otra perspectiva de la historia
Why does it matter that Pegatron is manufacturing these systems in Texas rather than overseas?
It's about reducing friction. If you're running a data center in the U.S. and your hardware fails or needs service, having the manufacturer nearby means faster repairs and better support. It also signals to American customers that Pegatron is committed to the domestic market, not just treating it as a secondary market.
What's the actual difference between liquid-cooled and air-cooled systems?
Liquid cooling is more efficient at removing heat from densely packed chips, which lets you fit more computing power into the same physical space. But it's more complex to maintain. Air cooling is simpler and cheaper, but takes up more room. You'd use liquid cooling when you need maximum density and performance; air cooling when you have space and want simplicity.
Why do all three companies need each other? Couldn't Pegatron just sell hardware directly to data centers?
Pegatron could, but they'd be selling boxes. Together AI brings the software layer—the tools and platforms that actually let customers train and run AI models efficiently. 5C brings the operational expertise and the customer relationships in the data center world. Together, they're selling a complete solution, not just hardware.
What does "L12 total solution capacity" actually mean?
It's a technical specification for how much computing power they can deliver end-to-end. The number itself matters less than what it represents: they're claiming they can handle the full pipeline from design through deployment, not just one piece of it.
Is this partnership temporary or permanent?
The language suggests it's ongoing. They're already deploying systems in Maryland and Texas, and they're developing new server lines based on newer NVIDIA chips. This looks like a long-term alignment, not a one-off project.