#What Makes D-Matrix’s Technology Stand Out in Inference?
D-Matrix, a Silicon Valley-based AI hardware startup, is stepping into the competitive arena dominated by Nvidia with its innovative Corsair inference accelerator platform. Founded in 2019, D-Matrix claims that its technology significantly enhances performance, achieving up to ten times the speed of standalone Nvidia GPUs while also consuming as much as five times less energy.
This ambitious platform began volume production in June 2026, a clear indication that the product is more than just theoretical concepts presented at conferences; it is actively being shipped and utilized in the market.
#How Does the Corsair Platform Function?
D-Matrix addresses a critical problem known as the memory wall, which inhibits the efficiency of AI inference. The core issue is not in the computational capability but rather in the movement of data. Often, chips spend more time retrieving information from memory than performing calculations. To overcome this, Corsair utilizes an in-memory computing architecture, processing data where it resides rather than constantly transferring it.
Each Corsair PCIe card comprises 6,400 mm² of silicon and features 4 GB of what D-Matrix refers to as Performance Memory, allowing for an impressive bandwidth of 300 TB/s. Additionally, it supports up to 512 GB of capacity memory per card. In configurations utilizing dual cards, Corsair achieves peak computation figures that can reach 4,800 TFLOPs for MXINT8 and 19,200 TFLOPs for MXINT4 precision formats.
#How Is D-Matrix Positioning Its Product in the Market?
D-Matrix has crafted the Corsair platform to function effectively both as a standalone solution and in hybrid setups with Nvidia's Blackwell GPUs. This dual compatibility allows data center operators to integrate Corsair without the need to completely overhaul their existing GPU infrastructure.
Priced in the tens of thousands of dollars, the Corsair card is not aimed at being the cheapest option available. Instead, it focuses on delivering significantly higher inference throughput per dollar spent and per watt consumed.
In a strong show of confidence, D-Matrix secured $275 million during its Series C funding round, attracting notable investors such as Microsoft’s M12 venture arm and Singapore’s sovereign wealth fund, Temasek. Ceo Sid Sheth, who has led D-Matrix since its founding, showcased the revolutionary technology at the Hot Chips 2025 conference. Production ramped up in June 2026, responding to rising demand from major hyperscalers and cloud service providers.