AI Factories Turn Infrastructure Into a Business Advantage

Penguin Solutions is the AI factory platform that designs, builds, deploys, and manages AI factories, combining compute, memory, and software, and services to help organizations deploy AI at scale.

Q: What is the main difference between traditional data centers and modern AI factories? 

A: The fundamental difference lies in the complexity of power and networking requirements, as well as the specialized expertise needed to build them. Designing an AI factory must begin with a specific use case or workload in mind, working backward to build an infrastructure capable of supporting the dense, high-speed throughput demanded by Graphics Processing Unit (GPU) acceleration. Traditional Information Technology (IT) administrators are accustomed to standard enterprise workloads, whereas AI clusters require designing entirely new architectures from the ground up or retrofitting existing infrastructure with strategic investments to generate advanced decision-making outcomes.

Q: How has Penguin Solutions’ position changed as the market has shifted from experimenting with AI to deploying AI workloads at scale? 

A: We have been running High-Performance Computing (HPC) and AI workloads for decades, long before the AI global phenomenon gained traction. This extensive tenure positions us at the forefront of the current revolution, serving as the central hub that connects the complex components required to build modern AI clusters. As enterprises shift from experimentation to large-scale deployment, our deep engineering expertise is highly sought after across the Americas. In regions like Mexico, close economic collaboration with the US enables a rapid exchange of subject matter expertise, empowering local IT professionals who are eager to expand their skill sets to tackle advanced ventures.

We are part of a select group of companies that is able toexecute the full AI factory lifecycle, from design and build to deployment and management. By integrating these four critical phases, we find ourselves without a direct competitor. We utilize Original Equipment Manufacturer (OEM) products that represent our proprietary intellectual property, allowing us to deliver comprehensive, end-to-end outcomes. Furthermore, we actively manage more than 100,000 GPUs and have accumulated over four billion hours of GPU runtime, an experience level that has driven us to develop our own specialized software and compute solutions.

Q: How do you measure the return on investment associated with AI infrastructure? 

A: We evaluate return on investment by directly asking our customers how they define success for their specific operations, which often revolves around timing, the cost of deployment, and "time to token" metrics. Based on their performance requirements across colocation facilities, on-premises data centers, or hybrid models, we engineer the solution to meet those exact financial targets. Sitting at a crucial intersection of hardware availability and platform integration, we have multiple levers to drive efficiencies. For example, by integrating innovative memory appliances into our platforms, we can achieve advanced cluster outcomes at a significantly lower price point, thereby maximizing the financial return for our clients.

Q: Which services are most requested by your Latin American customers and what are the biggest pains these companies are facing? 

A: While the initial design architectures are generally well understood, Latin American companies primarily request our expertise for deployment and operational execution. The greatest demand lies in professional services required to physically integrate racks into data centers, cable the systems, and connect liquid cooling infrastructure to manage intense power loads. After we build and test the racks in our California factory, we deploy them on-site and configure the networking. Additionally, day-two management is a significant pain point because GPUs are fragile components that require constant troubleshooting and maintenance. 

Q: Penguin Solutions Full-Stack AI Factory Platform integrates solutions like ClusterWareAI, MemoryAI, ComputeAI, OriginAI, and end-to-end services. How does this offering translate into a concrete advantage for a B2B customer? 

A: Integrating these solutions eliminates the hidden bottlenecks that frequently compromise accelerated GPU workloads and diminish capital investments. Our software suite, particularly ClusterWareAI, proactively controls and remediates performance drops or hardware outages, ensuring the entire cluster maintains optimal health and throughput. If a cluster is built using disjointed components, an organization might only realize a fraction of its intended efficiency, wasting substantial resources. 

Q: What problem does the MemoryAI KV Cache Server solve and what improvements can it deliver for customers? 

A: This solution was developed to address the escalating demand for massive memory capacity in accelerated workloads amidst global component shortages. Traditionally, achieving specific performance targets might require a large, exclusively GPU-based cluster, which entails high costs and long procurement lead times. By integrating MemoryAI appliances, which utilize standard Central Processing Units (CPUs) to drive memory efficiencies, we can significantly reduce the number of GPU nodes required to achieve the exact same operational outcome. This approach delivers a smaller initial investment, shorter lead times for hardware delivery, faster time to token, and an inherently simpler architectural footprint.

Q: Why is the “AI Factory” model becoming an operational necessity rather than simply a new technological architecture? 

A: The rapid evolution of specific applications has transitioned the technology from an experimental interest to an operational imperative. Forward-thinking enterprises have housed vast amounts of proprietary data in their storage networks for years; the AI factory model finally unlocks this latent value through agentic AI and inference workloads. By automating complex analytics and reducing duplicate efforts, organizations can instantly extract relevant insights that were previously inaccessible through traditional database management. This capability magnifies the effectiveness of human teams, directly delivering highly contextualized data to decision-makers and transforming raw computing power into measurable business productivity.

Q: As an AI Factory evolves, new governance challenges arise, like cost control, security, and energy constraints. How prepared are Latin American markets to operate this model, and what gaps in maturity have you identified? 

A: The governance challenges surrounding power and resource allocation are not unique to Latin America; they mirror the same conversations happening across North America and the European Union. Enterprises globally are working to balance the economic demand for new capabilities with the maintenance of existing infrastructure. While there is substantial built-up demand and investment flowing into the Latin American market, closing the maturity gap will require time and continuous innovation to improve operational efficiencies and reduce costs. Ultimately, preparedness is highly localized; the ability to successfully manage power constraints and workload allocations will depend on bespoke, city-level infrastructure planning rather than a uniform regional approach.

Q: What factors could determine whether Latin America succeeds in transitioning from consuming AI services to developing and operating competitive AI infrastructure at scale? 

A: The primary determining factor will be human capital and the willingness of IT professionals to adapt their skill sets from traditional virtualized environments to complex AI ecosystems. Success depends on local talent actively partnering with specialized firms and local integrators to design and manage these clusters internally, rather than merely consuming externally provided models. Furthermore, the region must develop a deep understanding of the global supply chain. Latin American countries have a significant opportunity to drive local investment by identifying how they can contribute to increasing the worldwide supply of critical infrastructure components.

Q: The evolution toward inference and Agentic AI is changing the requirements for computing, memory, networking, and operations. Where is Penguin Solutions focusing its innovation efforts? 

A: Our innovation efforts are intensely focused on resolving the memory bottlenecks that hinder advanced workloads. We are engineering superior solutions for both memory bandwidth and capacity to support the heavy demands of inference and agentic AI. The next critical frontier in the industry is transitioning away from massive model training toward applying those trained models within secure environments.