Microsoft-Backed D-Matrix Commences Full Production of AI Chip Corsair

AI chip manufacturer D-Matrix, with Microsoft among its investors, has announced the commencement of full-scale production for its highly anticipated Corsair inference platform. The company claims its new chip offers significantly higher performance and energy efficiency for AI inference workloads compared to existing graphics processing units (GPUs). This development has the potential to intensify competition in the AI hardware market, currently dominated by Nvidia (NVDA).

D-Matrix states that its Corsair chip can run inference workloads 10 times faster and with five times less energy consumption than standalone Nvidia GPUs. These claims specifically target the inference stage, which is crucial for the rapid and efficient operation of large language models (LLMs). Utilizing its Digital In-Memory Compute (DIMC) architecture, Corsair adopts a memory-centric approach designed to overcome the memory bottlenecks encountered by traditional GPUs. This innovative approach holds the potential to reduce the total cost of ownership and energy consumption for running AI models in data centers.

The company secured $275 million in Series C funding in November 2025, led by a global consortium including Temasek and Microsoft's M12 venture fund, bringing its total raised capital to $450 million. This funding round valued D-Matrix at $2 billion. The participation of prominent investors such as the Qatar Investment Authority (QIA) and EDBI has enabled the company to advance its product roadmap and support its global expansion. With the Corsair platform now entering volume production, D-Matrix is commencing shipments to hyperscalers, neoclouds, enterprise, and frontier AI labs.

However, there are nuances to D-Matrix's positioning as an “Nvidia challenger.” Some analyses suggest that the Corsair chip performs optimally in a disaggregated pipeline alongside Nvidia Blackwell GPUs. In this scenario, Corsair handles the sequential token generation (decode) phase of inference, while Nvidia GPUs manage the initial input prompt processing (prefill). This indicates that D-Matrix is positioning its solution as complementary to existing Nvidia infrastructure rather than a complete replacement. While the company states its SRAM-based architecture bypasses DRAM bottlenecks, experts like Rick Bahr from Stanford University caution that SRAM's capacity for managing extremely large models with trillions of parameters might be limited.

D-Matrix CEO Sid Sheth anticipates that the AI inference market could reach a valuation of $1 trillion. With Corsair, the company aims to capture a significant share of this large market by providing low-latency, high-throughput, and energy-efficient AI inference solutions for data centers. As markets closely monitor this new competition in the AI chip sector, where Nvidia maintains dominance, the potential for widespread adoption of D-Matrix's offering and its long-term market impact will become clearer in the coming period. The company's strategy of offering a complementary solution to existing infrastructures may provide a more attractive transition path for data center operators.