Takeaways by Saasverse AI
- d-Matrix | Series C | $275 Million | AI Inference Chips with Memory Computing Architecture.
- Led by Bullhound Capital, Triatomic Capital, and Temasek, with participation from QIA, EDBI, and M12.
- Valued at $2 billion, d-Matrix aims to disrupt GPU-dominated inference systems with a 10x performance boost and 3-5x efficiency gains through its Corsair™ and JetStream™ platforms.
d-Matrix, a leading innovator in AI inference hardware, has successfully raised $275 million in a Series C funding round, pushing its valuation to $2 billion. The round was led by Bullhound Capital, Triatomic Capital, and Singapore’s sovereign wealth fund Temasek, with additional backing from new investors such as Qatar Investment Authority (QIA) and EDBI, alongside existing stakeholders like Microsoft’s venture fund, M12. This latest round brings d-Matrix’s total funding to $450 million, underscoring its pivotal role in shaping the future of AI inference.
The company’s groundbreaking platform integrates memory computing technology, high-speed networking, and inference-optimized software to deliver unmatched performance for large-scale AI models. d-Matrix’s Corsair™ inference accelerator, JetStream™ NIC, and Aviator™ software collectively enable customers to achieve up to 10x performance improvements, 3x cost reductions, and 3-5x energy efficiency enhancements compared to conventional GPU-based systems. The platform can process up to 30,000 tokens per second for Llama 70B models, with a latency of just 2 milliseconds per token. With its high-density design, d-Matrix empowers enterprises to run AI models with up to 100 billion parameters on a single rack, a feat previously unattainable with traditional infrastructures.
Sid Sheth, CEO and Co-Founder of d-Matrix, highlighted the company’s long-term vision: “From the start, our focus has been on inference. Six years ago, the industry was fixated on training as the primary bottleneck, but we foresaw that the real challenge would emerge later—when trained models require continuous, large-scale operations. Over the past six years, we’ve built a new architecture to make AI inference scalable and ubiquitous.”
The Corsair™ accelerator is at the heart of this innovation, leveraging digital in-memory computing to embed processing directly into memory, a stark departure from the traditional GPU design where processors and memory are separate entities. Each Corsair PCIe card is equipped with two custom chips, featuring 1GB of SRAM each and a RISC-V-based control core. The system employs DMX Link interconnect technology to seamlessly integrate components, supported by 256GB LPDDR5 memory for data storage.
To scale these breakthroughs, the JetStream™ NIC enables the construction of AI inference clusters, with d-Matrix’s SquadRack reference architecture supporting up to 8 servers per rack, each hosting 8 Corsair cards. This setup allows a single SquadRack to run models with up to 100 billion parameters in SRAM, delivering a tenfold performance advantage over HBM-based solutions.
Looking ahead, d-Matrix plans to launch its next-generation Raptor inference accelerator next year. Raptor will feature a cutting-edge 3D stacking design that integrates RAM directly atop compute modules and will transition from a 6nm to a 4nm process node to further enhance performance and efficiency. This funding will fuel d-Matrix’s global expansion and support the deployment of advanced AI clusters for its customers.
“ d-Matrix’s ability to address the growing demands of AI inference positions it as a formidable challenger to GPU incumbents like NVIDIA. By focusing exclusively on inference and pioneering memory-computing-based architectures, d-Matrix has seized a critical opportunity to reshape the hardware landscape. Its solutions not only promise dramatic performance gains but also tackle the rising costs and energy inefficiencies associated with AI workloads at scale. ” Saasverse Analyst comments
Saasverse Insights
The AI hardware ecosystem is undergoing a seismic shift as specialized architectures like d-Matrix’s gain traction. The Corsair platform’s ability to natively support massive language models while drastically reducing latency and costs reflects a broader trend in AI: the need for domain-specific hardware that optimizes for inference rather than training. With global AI workloads expected to soar, d-Matrix’s advancements could catalyze an industry-wide pivot away from GPU-centric designs, spurring innovation in memory computing and alternative architectures. However, the company’s ambitious roadmap, including the upcoming Raptor accelerator, will need flawless execution to maintain its edge in this highly competitive market. If successful, d-Matrix could emerge as a cornerstone player in the next generation of AI infrastructure.