Breaking

Nvidia and d-Matrix to Deploy Combined GPU-Inference Chip System

July 9, 2026 at 14:17 EDT

Infra & Chips
Funding & M&A

A "heterogeneous" inference system combining Nvidia's GPUs with accelerators from inference startup d-Matrix is heading to commercial deployment. Inference cloud provider Parasail will be the first commercial customer, with launch planned for late 2026.

July 2026 · d-Matrix + Parasail

Nvidia Pairs Its GPUs With a Rival's Chip to Chase Faster Inference

A "heterogeneous" system splits the job: NVIDIA Hopper & Blackwell GPUs handle input processing (prefill), while d-Matrix's Corsair accelerator handles token generation (decode). First commercial customer: Parasail — launching late 2026.

10×

target faster token generation for the joint system

$275M

Series C raised in Nov 2025 to power AI inference

2400

TFLOPS 8-bit dense compute on the Corsair PCIe card

Corsair vs. H100 — d-Matrix's projected gains

Llama-70B, 4K context, 8-bit inference. Figures are projections / internal benchmarks — independent verification still limited.

1×

H100
baseline

10×

interactive
speed

20×

throughput

20×

lower
latency

3× cost-performance

3–5× energy efficiency

How the disaggregated pipeline splits the work

NVIDIA GPUs

Hopper & Blackwell handle compute-heavy prefill

→

d-Matrix Corsair

3D in-memory compute handles latency-sensitive decode

WHY IT MATTERS

As AI demand shifts from training to low-latency inference, Nvidia hedges by absorbing rivals rather than shutting them out. Corsair, in mass production on TSMC's 6nm, won a 2026 AI Breakthrough Award.

THE CAVEAT

Many performance figures

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.