Liqid Unveils Massive Scale-Up AI Platform Powered by AMD Instinct™ MI350P GPUs

Business Wire

Liqid® (www.liqid.com), the global leader in software-defined memory and GPU pooling infrastructure, today announced a strategic collaboration with AMD to deliver next-generation AI infrastructure solutions built around the new AMD Instinct™ MI350P GPUs. The collaboration pairs Liqid’s GPU pooling solutions with AMD PCIe-based Instinct GPUs to give enterprise and cloud customers scalable, efficient AI infrastructure, without rebuilding their datacenters around it.

As AI inference workloads continue to grow in size and complexity, organizations need infrastructure that delivers higher performance while maximizing GPU utilization and lowering the cost of every generated token. Powered by AMD Instinct MI350P Series GPUs, Liqid enables customers to dynamically pool and scale-up 30x GPUs to a server, creating large, pooled accelerator fabrics that can deploy models of the largest size. The result is breakthrough AI performance, higher GPU utilization, and industry-leading token economics; delivering more tokens per second, more tokens per dollar, and more tokens per watt without requiring specialized cooling or a complete datacenter redesign.

“Liqid has established itself as the leader in GPU pooling and scaling,” said Rick Hegberg, CEO, Liqid. “The AMD Instinct MI350P Series gives us the ideal PCIe-based GPU to build the next generation of AI infrastructure, allowing us to deliver solutions tuned for enterprise AI inference where utilization and cost per token decide the economics.”

“The AMD Instinct MI350P PCIe GPU delivers exceptional AI inference performance without requiring specialized infrastructure,” said Suresh Andani, corporate vice president, Compute and Enterprise AI Group, AMD. “Together with Liqid’s GPU pooling solutions, customers can scale GPU resources on demand to achieve higher utilization, lower infrastructure costs, and industry-leading AI inference economics.”

Liqid UltraStack 30 – AMD MI350P

Specification	Configuration
Server	AMD EPYC™ 9005 Series CPU, Dual-Socket
GPUs	30× AMD Instinct™ MI350P PCIe
AI Performance	69 PFLOPS (FP8)
GPU HBM Memory	4.3 TB
Total System Power	~22 kW

Scale up to 30x MI350P GPUs within a single server through Liqid’s GPU pooling solution.
Scale more than 4TB of aggregate HBM3E for deploying large AI models in one system.
Support multiple parallel model deployments with native Kubernetes support.
Reduce the infrastructure cost of deploying frontier and large-context models.
Target up to 65% lower deployment cost and 50% lower power consumption.

Optimized for AI Inference

This solution is designed for AI inference, where infrastructure efficiency directly impacts deployment cost and business value. By pooling and sharing AMD Instinct GPUs over Liqid’s fabric, customers optimize the metrics that matter most:

Up to 3.7x higher tokens per second — significantly greater inference throughput than traditional servers.
Up to 2.1x more tokens per dollar — generate up to two times more AI tokens from the same capital investment.
Up to 1.8x better tokens per watt — double inference efficiency, cutting power consumed per generated token.

Key Metric	Value
Performance per kW	~3.14 PFLOPS/kW
HBM per kW	~195 GB/kW
HBM per PFLOP	~62 GB/PFLOP

Memory Pooling Ready System

The Liqid platform, powered by AMD, is designed to be CXL memory pooling ready, providing a future path to memory expansion and shared memory access across multiple servers. As CXL memory pooling and sharing mature, customers will be able to dynamically allocate terabytes of memory for KV cache and other memory-intensive AI workloads. Combined with large-scale GPU pooling, this next-generation architecture will support larger AI models, higher inference throughput, and significantly improved AI infrastructure economics, delivering unprecedented performance, efficiency, and scalability.

Use Cases and Who It’s For

By combining AMD Instinct MI350P GPUs with Liqid’s GPU pooling platform in this new solution, customers can scale AI infrastructure beyond traditional server limits to run larger models, more models per server, and deliver industry-leading inference economics. Key enterprise workloads include:

Enterprise agentic assistants — sales, HR, marketing, legal, and coding assistants running on-premises, where data stays private and costs stay predictable.
Retrieval-augmented generation (RAG) and long-context inference that benefit from more than 4TB of HBM3E per system.
Scientific and technical workloads — drug discovery, materials science, and mechanical design that need GPU capacity that flexes with demand.
Mixed model fleets — small, medium, and large models served from one shared pool to enable higher model per server density.

The platform is built for organizations deploying AI at scale, NeoCloud and AI service providers, enterprise AI teams, and high-performance computing and research institutions. Pairing AMD AI accelerators with Liqid’s GPU pooling infrastructure allows customers to build flexible AI environments that adapt as workload demands change.

Liqid and AMD are committed to an open, scalable AI infrastructure platform that accelerates AI adoption while maximizing performance, efficiency, and return on investment. More detail on the companies’ joint solutions will follow as development milestones are reached, and customer deployments begin.

Additional Resources

Liqid UltraStack 30 – AMD MI350P

Liqid Matrix Software

Liqid GPU Pooling Solutions

Liqid Memory Pooling Solutions

Connect with Liqid on LinkedIn

About Liqid

Liqid is the global leader in software-defined memory and GPU pooling. Liqid builds flexible, high-performance, and efficient datacenter and edge solutions that deliver superior tokenomics. Liqid powers next-generation AI inference, KV cache, in-memory databases, VDI, virtualization, and HPC workloads. Liqid solutions are trusted by organizations across cloud service providers, financial services, higher education and research, healthcare, media and entertainment, and government and defense.

Liqid enables customers to manage, configure, reconfigure, and scale essential memory, compute, accelerators (GPU, DPU, TPU, FPGA), storage, and networking into physical bare-metal server systems in seconds. Liqid customers can optimize their IT infrastructure and achieve up to 100% memory and GPU utilization for maximum tokens per watt, tokens per dollar, and tokens per second. Learn more at www.liqid.com.

Performance figures referenced for the AMD Instinct MI350P are based on AMD published specifications and engineering projections as of April–May 2026 and are subject to change. Liqid solution metrics are targets based on internal projections and may vary by configuration and workload. AMD, the AMD Arrow logo, AMD Instinct, CDNA, EPYC, ROCm, and combinations thereof are trademarks of Advanced Micro Devices, Inc. PCIe is a registered trademark of PCI-SIG Corporation.