Parasail to Combine NVIDIA AI Infrastructure with d-Matrix Accelerators to Achieve 10x Faster Token Generation

Inference service provider to deliver faster, more cost-efficient tokens by pairing NVIDIA Hopper and Blackwell GPUs with d-Matrix Corsair accelerators

Parasail, the inference cloud for AI-native startups, and d-Matrix, a pioneer in low-latency AI inference compute platforms for data centers, announced Parasail is deploying d-Matrix Corsair inference accelerators alongside the NVIDIA Hopper and NVIDIA Blackwell architectures to deliver up to 10x faster, more cost-efficient inference services to its customers.

Parasail’s Corsair deployment marks one of the first commercial-scale examples of heterogeneous disaggregated inference in production, where NVIDIA AI infrastructure and d-Matrix’s purpose-built Corsair inference accelerators will operate in concert, each performing computing tasks in which they excel. Using this approach, Parasail aims to improve inference economics for select workloads by combining NVIDIA GPUs for compute-intensive prefill with d-Matrix Corsair accelerators for latency-sensitive decode.

With approvals and construction buildouts of new data centers a multi-year process, Parasail is leaning into a heterogeneous compute approach with d-Matrix to obtain even more value from the NVIDIA AI infrastructure already in its data centers. By pairing Corsair with its Hopper and Blackwell fleet, Parasail is extending fleet performance and passing the gains directly to its customers. For the developers and enterprises relying on Parasail’s platform, the result is simple: faster, more economically efficient tokens, today.

“We’re relentless about delivering the best inference performance per dollar,” said Mike Henry, founder and CEO of Parasail. “Everyone’s focused on the next GPU generation, but we run large Hopper and Blackwell fleets today and pairing them with Corsair lets us deliver our customers the edge they need while also extending the life of the hardware we’ve already deployed.”

“Parasail is proving what we’ve believed from the start: the future of inference is heterogeneous,” said Sid Sheth, founder and CEO of d-Matrix. “GPUs and purpose-built inference accelerators aren’t competitors; they’re better together. By deploying Corsair alongside their NVIDIA fleet, Parasail is embracing a heterogeneous inference approach designed to optimize each phase of the workload.”

Also Read: AiThority Interview with Matej Bukovinski, Chief Technology Officer at Nutrient

The Technology Innovations Driving Maximum Efficiency
Parasail’s automatic kernel optimization technology plays a critical role in this deployment, dynamically routing workloads to the right hardware at the right time to extract maximum performance and value from each model across a heterogeneous fleet. The result is a cloud-based token service that delivers premium inference economics to Parasail’s customers at scale.

d-Matrix Corsair’s performance advantage stems from d-Matrix’s foundationally unique Digital In-Memory Compute (DIMC) chiplet architecture, which integrates compute with memory on the same silicon. In conventional systems, data must travel back and forth between separate memory and processor chips during inference, which consumes time, power, and cost with every transfer. Corsair eliminates that penalty entirely, enabling up to 10x faster interactive inference and up to 3x better energy efficiency versus traditional approaches. Built on TSMC’s N6 process with organic substrates and LP-DDR5 memory, Corsair’s architecture is designed to support efficient, scalable inference deployments for customers like Parasail.

The companies plan to share detailed performance results and case studies following the first series of deployments and explore expanded integration across Parasail’s global fleet of over 40 data centers in 15 countries.