Infineon Technologies AG and d‑Matrix today announced a strategic collaboration that advances AI inference performance, power efficiency, and system integration for data‑centre environments.

The partnership pair’s d‑Matrix Corsair inference accelerator is designed for highly interactive, low‑latency AI applications with Infineon’s OptiMOS TDM2254xx dual‑phase power modules, enabling true vertical power delivery and a class‑leading power density of 1.0 A/mm² on high‑density boards.

Meeting the demands of real‑time AI

AI inference, applying trained machine learning models to new inputs to generate predictions, classifications, or decisions, is rapidly shifting from batch processing in the back office to real‑time, interactive user experiences. That transformation imposes strict requirements on compute architecture: sub‑millisecond to low‑millisecond latencies, predictable performance under load, and tight power and thermal budgets in dense server racks.

Corsair was purpose‑built for that moment. d‑Matrix engineered the Corsair inference accelerator to deliver sub‑2ms token latency for interactive large language model (LLM) workloads and other real‑time AI use cases, while achieving energy efficiency multiples above traditional approaches. To sustain such performance at data‑centre scale, Corsair requires high‑density, low‑loss power delivery that can be tightly integrated into board and system design.

Infineon’s broad portfolio spanning silicon (Si), silicon carbide (SiC), and gallium nitride (GaN) power semiconductors enables tailored solutions across both AI inference and training markets. From grid‑level power to board‑level delivery, Infineon’s devices help designers balance peak performance, latency constraints, and operational efficiency. The company’s early and sustained collaboration with inference‑focused customers has reinforced its position as a trusted partner for leading AI hardware developers.

Leadership perspectives;

Infineon has been collaborating with customers specialising in inference processors, such as d‑Matrix, from the early days when the industry was mostly focused on training hardware, said Raj Khattoi, Vice President and General Manager of Consumer, Computing and Communication at Infineon. These early, strategic engagements have positioned Infineon as a leader in the inference hardware industry, further extending our leadership in powering AI with semiconductor solutions for both inference and training processors.

AI is rapidly moving from back‑office experimentation to a real‑time interactive experience, and that shift demands a fundamentally different compute architecture, said Sid Sheth, founder and CEO of d‑Matrix. Corsair was purpose‑built for this moment: delivering the sub‑2ms token latency that interactive applications require, at multiples better energy efficiency than traditional approaches. Infineon has been a design partner since the inception of our platform, and their power semiconductors are a meaningful contributor to our ability to deliver what the market demands.

Practical applications and market impact
d‑Matrix’s Corsair accelerator, enabled by Infineon’s power modules, targets a range of inference use cases that require low latency and reliable throughput:

• LLM response generation for chat, assistants, and customer interaction systems.
• Agentic AI for automation workflows and decision agents that must act in near real‑time.
• Predictive analytics in finance and healthcare, where timely insights impact outcomes.

As AI workloads continue to proliferate across enterprises and cloud providers, data centres face increasing pressure to deliver greater compute density without proportional increases in power and cooling costs. Solutions that tightly integrate power electronics with accelerator design can provide measurable reductions in energy consumption, improved thermal profiles, and simplified system scaling advantages that translate into lower total cost of ownership for operators.

For organisations evaluating inference platforms, the signal is clear: power architecture is not an afterthought; it’s a strategic differentiator. As interactive AI moves from experiment to expectation, partnerships like Infineon and d‑Matrix demonstrate the integrated engineering required to meet tomorrow’s performance and efficiency targets.

Please feel free to access our most recent updates on the relevant report: https://semiconductorinsight.com/report/inference-ai-chip-market/