D-Matrix, located near Nvidia's headquarters, asserts that its Corsair chip can execute inference tasks 10 times faster and with five times less energy than Nvidia's standalone GPUs, provided the workloads are small. The company, founded in 2019, has raised approximately $500 million, achieving a valuation of around $2 billion.
D-Matrix has secured commitments from various high-profile customers, primarily in the U.S., and plans to begin shipping its chips this month. The Corsair chip utilizes a unique memory integration approach, relying on SRAM instead of DRAM, which is currently in short supply.
While this design allows for low-latency performance, it may struggle with larger AI models that require extensive memory capacity. D-Matrix's strategy includes a plug-and-play system that differentiates it from competitors like Cerebras and Groq. The company is also developing its next chip, Raptor, expected to launch next year.
This competitive landscape suggests that as AI demand grows, innovative startups like D-Matrix could carve out significant market share, challenging established players like Nvidia