Powering AI with next-generation DPUs
Powered by Kalray's MPPA® architecture, our Data Processing Unit technologies unlock a new class of intelligent AI systems where data movement, networking, storage, and security are offloaded, accelerated, and processed with unmatched efficiency and scalability.
From cloud to edge: a foundational layer for the AI era
Kalray believes the future of AI is no longer defined by compute power alone, but by how efficiently data is moved, processed, and secured across distributed systems. DPUs are the foundational building block of that future, and we have spent 15 years architecting them.
The AI breaking point
AI is reaching a breaking point as the scale, distribution, and dynamism of workloads outpace the capabilities of traditional systems. Generative AI has already placed unprecedented pressure on compute, networking, and data movement, and agentic and physical AI are about to make the problem worse, not better.
Compute is no longer the only bottleneck
GPU underutilization, rising costs, and inefficiencies are now driven by the data path, not compute performances.
Workloads are interactive, latency-sensitive, distributed
Agentic AI and physical AI introduce fundamentally different requirements: continuous reasoning, real-time decision-making, and exponentially growing data flows across cloud, HPC, datacenter, and edge.
The bottleneck shifts to data movement
Moving, processing, and securing data efficiently across distributed AI systems becomes the binding constraint. A new class of processing is required, data-centric, massively scalable, and continuous.
Three processors, each in its optimal domain
CPU
OS & orchestration
Tools & pipelines
GPU
Inference Compute
Tensor operations
DPU
Multi-tenant isolation
Fabric & storage offload
The architecture accelerating DPU development
Kalray's patented Massively Parallel Processor Array (MPPA®) is the open, scalable, modular foundation behind our DPU solutions. Its programmable cores, networking engines, hardware accelerators, and high-speed interconnects can be tailored and assembled with flexibility to create DPU solutions precisely matched to customer requirements across AI, HPC, cloud, and edge environments.
A unique approach to DPUs
Kalray's patented MPPA architecture
High-performance, energy-efficient processing tailored for data-intensive workloads. Flexible design enabling efficient data-path processing and specialized compute.
Proven technology and unique expertise
15+ years of DPU innovation. World-class expertise combined with production-proven architectures engineered for data-intensive workloads.
Open and flexible by design
Flexibility and control to design, customize, and scale infrastructure. Through open standards, flexible licensing models, and tailored design services.
Tailored solutions for specific customer needs
From architecture definition to silicon design, software development, and full system integration, Kalray partners across the entire value chain.
GigaOm DPU Sonar Report
Recognized as a Leader and Fast Mover for innovative architecture and strong performance across AI and data workloads.
ISO 9001 certified
Interoperability validations across the storage, AI, and networking ecosystem.
Deployed at 50+ customers
Opening the DPU era, transforming how data is moved, processed, and secured
Talk to Kalray about deploying DPUs, licensing IPs for your custom silicon, or co-designing the next generation of data-centric infrastructure.