Lotus Microsystems has launched its first vStrata module for AI data centre infrastructure, marking the company's first commercial product built around a vertical power delivery approach.
Engineering samples of the first module, known as LSC0580, are due to ship in the third quarter of 2026. The Copenhagen semiconductor startup described it as a combined power delivery and thermal management platform for AI accelerators and other high-current compute systems.
Demand from AI workloads has increased pressure on data centre operators to manage rising power draw and heat output from processors. That has shifted attention toward designs that shorten the electrical path between power delivery components and the processor, in an effort to cut losses and improve efficiency.
Lotus Microsystems said its platform places power delivery beneath the processor using what it calls silicon Power Interposer Technology. It said this addresses electrical, thermal and mechanical constraints within a single design, rather than treating power conversion and cooling as separate parts of the system.
The first module has already completed tape-out work for what Lotus Microsystems described as leading xPU and AI infrastructure partners. It did not name those partners, but said it is working with Tier-1 hyperscalers and other companies through an early access programme.
According to the company, the LSC0580 is designed to cut power conversion losses by more than 50% and achieve point-of-load efficiency of up to 96%. It also said the module can reduce operating temperatures by as much as 25C in optimised configurations, while its thin design could allow operators to fit more boards into each rack.
Those claims target a central problem in AI infrastructure, where gains in processor performance often run into limits set by power delivery and cooling. As accelerators draw larger current loads, conventional board-level architectures can struggle to respond quickly enough to sharp shifts in demand without adding more components and generating more heat.
Lotus Microsystems said the LSC0580 has been engineered for kiloampere-class power demands and can support load transients exceeding 10 A/ns without external capacitors. It added that the module is designed to work with established reference designs and existing power management controllers, which could reduce redesign work for customers considering adoption.
System constraint
The company framed the launch as a response to a broader architectural challenge in AI computing, where system design rather than chip design alone is becoming a limiting factor.
"We have reached a point where AI compute performance is constrained by physical architecture," said Hans Hasselby-Andersen, Chief Executive Officer, Lotus Microsystems.
"vStrata was developed to address the reality that power delivery and thermal management have become inseparable system constraints. By integrating both into a single co-engineered platform, we help data centres increase compute density while reducing the power and cooling overhead that drives AI infrastructure cost," said Hasselby-Andersen.
Lotus Microsystems said its approach is based on multi-domain co-design, bringing electrical, thermal and mechanical engineering into one architecture. That reflects a broader trend in advanced semiconductor packaging, where chipmakers and system suppliers are looking beyond the processor die to improve overall system performance.
Design pressure
AI servers have become a focal point for those efforts because of their unusually high power density. Operators are under pressure to contain both electricity use and cooling costs as they build out larger clusters for model training and inference workloads.
Lotus Microsystems argued that the bottleneck lies not only in delivering more current, but also in handling sudden and unpredictable swings in processor demand. Reducing the distance between the power source and the processor is intended to improve response time while managing heat at the same location.
"The industry has been trying to scale AI compute using an architecture that was never designed for this level of intensity," said Yasser Nour, Co-founder and Chief Technology Officer, Lotus Microsystems.
"The real bottleneck isn't just delivering power - it's how the system responds to rapid, unpredictable load changes. By collapsing the distance between the power supply and the processor, vStrata is designed to deliver current where and when it's needed, without compromising thermal behaviour or transient stability," said Nour.
Founded in 2020, Lotus Microsystems is based in Copenhagen and focuses on low-profile vertical power delivery and thermal management products for next-generation compute systems. The company said it holds multiple patents covering the underlying technology.