d-Matrix Ties Up With Gimlet Labs to Power 10x Faster, Lower-Power Agentic AI Inference

New updates have been reported about d-Matrix.

d-Matrix is expanding the reach of its Corsair low-latency, memory-optimized AI accelerators through a new integration with Gimlet Labs’ multi-silicon Gimlet Cloud, positioning its hardware as a core engine for next-generation agentic AI inference. Under the collaboration, Gimlet will deploy d-Matrix Corsair PCIe cards alongside GPUs in its data centers, targeting an order-of-magnitude improvement in both inference latency and throughput per watt versus GPU-only stacks.

The joint architecture splits inference workflows so that GPUs handle the phases they are best at while d-Matrix accelerators execute memory-bound and latency-critical stages, such as speculative decoding that underpins highly interactive AI services. By offloading these energy-inefficient GPU tasks to Corsair, the companies expect to deliver materially faster token generation and significant power savings, directly addressing the industry constraint that power availability is capping AI growth. Gimlet’s software stack dynamically maps segments of agentic workloads across heterogeneous accelerators from multiple vendors and generations, leveraging high-speed interconnects in its data centers to orchestrate this division of labor at scale.

For d-Matrix, the deal validates its inference-only strategy and positions Corsair as enabling infrastructure for frontier model providers and AI-native enterprises seeking lower operating costs and higher performance in cloud deployment. The Corsair architecture, optimized for high memory bandwidth and low latency, is designed to plug into existing data centers via standard air-cooled PCIe cards, which should support faster customer adoption and minimize capex friction for Gimlet and its clients. Management at d-Matrix frames this partnership as evidence that homogeneous GPU infrastructure is giving way to specialized, power-aware inference platforms, and that “doing more with less” will increasingly guide procurement decisions.

The combined d-Matrix–Gimlet solution is slated to be offered to select customers on Gimlet Cloud in the second half of 2026, creating a near-term commercialization milestone and a potential recurring revenue stream for d-Matrix if uptake among frontier labs and AI-native firms is strong. Early-access programs are already being promoted through d-Matrix and Gimlet’s channels, alongside a technical write-up detailing how Corsair accelerators enhance low-latency speculative decoding. For executives, the key implication is that d-Matrix is moving from proof-of-concept hardware into integrated cloud deployments with specialized inference providers, which could strengthen its ecosystem position and provide a platform for future products such as its JetStream IO accelerators and Aviator software.