Penguin Solutions (NASDAQ:PENG) is positioning its business around a full-stack “AI factory” platform that combines infrastructure, memory technology, software and managed services, Senior Vice President of Global Marketing Mark Adams said during Rosenblatt Securities’ Sixth Annual Age of AI Scaling Tech Conference.
Adams said the company’s strategy builds on more than 25 years of high-performance computing experience, including its early work commercializing scalable computing cluster technology that originated in NASA-related projects. He described Penguin’s evolution from a primarily hardware-oriented high-performance computing provider into a company focused on AI infrastructure and large-memory systems.
AI Factory Platform and Vendor-Agnostic Approach
Penguin’s platform includes OriginAI architectures, ClusterWareAI software, ComputeAI systems, MemoryAI appliances and services spanning design, build, deployment and management. Adams said the company seeks to evaluate customer workloads first and assemble systems using the most suitable components rather than applying a single standardized architecture.
The company recently became an initial participant in NVIDIA’s AI Factory Specialized Partner program, according to Adams. Penguin had previously been an NVIDIA DGX-ready managed services partner and has accumulated more than 4 billion hours of GPU-cluster runtime experience, he said.
While highlighting Penguin’s NVIDIA experience, Adams said the company supports other computing options, including AMD-based systems and other accelerators. He cited a cluster delivered to Sandia National Laboratories called Spectra, which uses Kneron accelerator silicon selected for the customer’s workload.
“We’re equally happy to bring to a customer that needs it an NVIDIA cluster, an AMD cluster, or other clusters,” Adams said.
ClusterWareAI Focuses on Deployment and Uptime
Adams described ClusterWareAI as Penguin-developed software designed to deploy hardware as a cluster, monitor systems continuously and automate remediation of certain issues. The software incorporates operational data and lessons from managing large GPU environments, he said.
According to Adams, the platform can monitor millions of data points and potentially identify signs that a component, such as a GPU or network adapter, may be at risk of failure. That can allow a node to be removed from a cluster before a failure disrupts a user job, he said.
Penguin licenses the software on a subscription basis, Adams said, often alongside managed-services agreements. He said the company views the platform as differentiated because it combines deployment, monitoring and AI-driven remediation capabilities rather than addressing only one part of that operational process.
MemoryAI Targets Inference Economics
Adams also outlined Penguin’s MemoryAI offering, which is aimed at AI inference and agentic AI workloads. The platform includes a KV-cache server appliance using Penguin-developed Compute Express Link, or CXL, memory expansion cards.
He said a MemoryAI server can provide 11 terabytes of DDR5-based memory storage for tokens that have already been computed during AI workloads. Storing and retrieving those tokens can reduce the need to return to GPUs to repeat work when similar requests arise, potentially improving response time and reducing compute requirements, Adams said.
Adams contrasted the approach with cache systems based on flash storage, which he said can be substantially slower. Penguin’s use of DDR5 memory is intended to reduce latency and improve time to first token, he said.
Enterprise, Sovereign AI and Neo Cloud Demand
Penguin is pursuing enterprise customers, sovereign AI deployments and neo cloud providers, Adams said. He characterized neo cloud opportunities as potentially larger and faster-moving because those customers are often adding capacity at scale. Sovereign deployments may require more design work and customer engagement, particularly where workloads must remain within a specific geography or within an enterprise’s private network.
Adams said enterprises are increasingly moving from experimentation toward production deployments of agentic AI and inference applications. These companies are generally less focused on training models than on applying trained models to business use cases, he said.
For enterprises considering on-premises systems rather than rented cloud capacity, Adams pointed to data security, proximity to internal systems and cost predictability. Companies moving from experimentation to sustained AI usage may view infrastructure as a long-term enterprise asset, he said.
Adams also discussed Penguin’s collaboration with SK Telecom, which is both an investor and customer. Penguin designed, deployed and manages SK Telecom’s Haein cluster in South Korea, which Adams described as a sovereign neo cloud deployment operating at full capacity. He said SK Telecom has announced plans to expand its AI business across the Asia-Pacific region, creating potential opportunities for further collaboration.
In partnerships with Dell Technologies, CDW and other companies, Adams said Penguin contributes cluster software, memory systems and managed services alongside partners’ hardware, distribution and customer relationships. Penguin was named Dell AI Partner of the Year this summer, Adams said.
Looking ahead, Adams said Penguin expects to remain focused on infrastructure while expanding its involvement in software, services, Kubernetes, AI orchestration and purpose-built customer applications as enterprise AI deployments mature.
About Penguin Solutions (NASDAQ:PENG)
Penguin Solutions, Inc engages in the designing and development of enterprise solutions worldwide. It operates through three segments: Advanced Computing, Integrated Memory, and Optimized LED. It offers dynamic random access memory modules, solid-state and flash storage, and other advanced integrated memory solutions for networking and telecom, data analytics, artificial intelligence and machine learning applications; and supply chain services, including procurement, logistics, inventory management, temporary warehousing, programming, kitting, and packaging services.
