d-Matrix, with the support of Microsoft, raised 275 million dollars.

d-Matrix, a semiconductor company backed by Microsoft, announced on November 12 that it raised $275 million. The company claims that its artificial intelligence chips perform better than Graphics Processing Unit (GPU) based systems.

The company conducted a Series C financing round led by Bullhound Capital, Triatomic Capital, and Singapore's Temasek sovereign wealth fund with a valuation of $2 billion. Microsoft participated among the investors through its venture arm M12. This funding brings the total raised by d-Matrix since its founding in 2019 to $450 million.

Based in Santa Clara, California, the startup stated that its model inference platform offers ten times better performance, three times cost efficiency, and five times energy efficiency compared to GPU-based systems.

The founder and CEO of d-Matrix, Sid Sheth, and the Technology Director Sudeep Bhoja established the company six years ago with a competing vision. While the industry focused on the training of artificial intelligence models, they developed technology aimed at running these models at scale.

Sheth said, “When we founded d-Matrix, training was viewed as the biggest challenge for artificial intelligence, but we knew that new challenges would soon arise.” The founders have previously manufactured over 100 million chips and built a team of 250 people across Santa Clara, Canada, Australia, India, and Serbia.

With the latest funding, interest from investors across various continents was attracted. New participants include the Qatar Investment Authority and the Singapore Economic Development Board Investment Division, while existing investors include M12, Nautilus Venture Partners, Industry Ventures, and Mirae Asset.

Following the funding, M12 Managing Partner Michael Stewart stated, “The explosion in demand for artificial intelligence inference demonstrates that efficiency and scalability can significantly contribute to revenue generation and profitability for hyperscales and AI factories.”

d-Matrix's Corsair accelerators provide a latency of 2 milliseconds while producing 30,000 tokens per second on the Llama 70B models. This speed enables the operation of 100 billion parameter models on a single server rack.

d-Matrix positions its technology as a solution to the sustainability challenges of artificial intelligence. The company claims that a single data center can meet the needs traditionally served by 10 facilities. Jeff Huber, General Partner at Triatomic Capital, stated, “Artificial intelligence inference is becoming the dominant cost in manufactured artificial systems, and d-Matrix has managed to deliver scaled performance and sustainable economies.”

d-Matrix introduced the SquadRack reference architecture in collaboration with Arista, Broadcom, and Super Micro Computer. This ecosystem approach aims to accelerate adoption among hyperscale, enterprise, and sovereign customers.