d-Matrix is a Santa Clara semiconductor company building inference-first compute hardware designed to make generative AI economically and energetically sustainable at scale. While much of the AI chip world was optimized around training, d-Matrix bet early that inference — the act of actually running models in production — would become the dominant and most cost-sensitive workload. Its architecture is built from the ground up for that future.
The company's core innovation is Digital In-Memory Compute (DIMC): performing matrix operations directly within memory using a digital (rather than analog) approach, which sidesteps the energy and latency cost of constantly moving data between memory and processing units. d-Matrix pairs DIMC with a chiplet-based design, assembling smaller silicon dies into larger packages to improve yield, scalability and cost. This combination underpins Corsair, the company's flagship inference compute platform unveiled in November 2024, which d-Matrix bills as one of the most efficient AI computing platforms for datacenter inference.
Corsair followed a series of earlier silicon milestones — the Nighthawk, Jayhawk-I and Jayhawk-II chiplets — that progressively validated d-Matrix's architecture. The pitch to customers is compelling: comparable or better inference throughput than leading GPUs at a fraction of the energy, translating into lower total cost of ownership for the hyperscalers, enterprises and AI providers serving large language models to millions of users.
d-Matrix has assembled an impressive capital base and investor roster. Its $110 million Series B in September 2023 was led by Singapore's Temasek with participation from Microsoft's M12, Industry Ventures, Ericsson Ventures, Mirae Asset, Samsung Ventures and others. In November 2025 the company raised a further $275 million to power the age of AI inference, bringing total funding to well over $450 million and cementing its place among the best-capitalized GPU-alternative startups.
As inference spending balloons and operators hunt for relief from GPU scarcity and power constraints, d-Matrix offers a differentiated, energy-efficient path. With proven silicon, a clear inference-first thesis and deep-pocketed strategic backers, it is one of the most credible challengers in AI datacenter compute.