As AI models scale over 1 trillion parameters and computing clusters expand to hundreds of thousands of GPUs, traditional interconnections are struggling to keep up. Next-generation AI requires much higher levels of connectivity, while significantly reducing power consumption.
Enter Hyperlume, a Canadian high-tech startup that tackles these infrastructure restrictions. Hyperlume replaces high-performance computing with aged copper-based solutions in AI data centers and high-bandwidth, low-latency optical interconnects.
Today, Hyperlume announced it has raised $12.5 million in seed funding to provide optical interconnect technology to the market. The round is led by BDC’s Deep Tech Venture Fund and Arctern Ventures, and was supported by strategic investments from Muus Climate Partners, SOSV, Intel Capital and LG Technology Ventures.
Push AI infrastructure beyond limits
Founded in 2022 by Mohsen Asad and Hossein Fariborzi, Hyperlume has developed high-bandwidth low-latency optical interconnects designed to replace aging copper-based solutions in AI data centers and high-performance computing environments. Masu. Unlike traditional interconnects that face bandwidth bottlenecks and power inefficiencies, Hyperlume’s technology leverages ultra-fast microrings and advanced low-power circuits to move data faster and increase latency. Reduce energy consumption dramatically.
By shifting away from traditional electronic interconnections, Hyperlume allows for 10x computational performance, 5x power savings, and 4x lower costs compared to existing solutions. This efficiency improvement is important as AI workloads continue to grow exponentially and increase power usage and cooling requirements in large data centers while demanding more demands from the infrastructure.
“We’re looking forward to seeing you in the future,” said Mohsen Asad, co-founder and CEO of HyperLume.
Today’s AI models are already pushing the limits of existing interconnections. As the industry expands, new solutions are needed to handle the exponential growth of data movements while reducing power consumption.
Scaling AI without energy trade-offs
Hyperlume’s optical interconnect offers an alternative designed to efficiently handle large AI workloads. With this funding, the company plans to strengthen its partnerships with hyperscalar, chip makers and AI infrastructure providers while expanding its product development, engineering and R&D efforts. The startup is also preparing for production to meet the demand for interconnections of 800g and 1.6T.
“The data center carbon emissions are expected to more than double by 2030, making energy-efficient networking solutions important,” said Murray McCaig, managing partner at Arctern Ventures. “Hyperlume’s ultra-low power microring-to-ring connection technology dramatically reduces energy consumption while providing top-class bandwidth performance.”
Srini Ananth, Managing Director at Intel Capital, pointed to the rising demand for AI power and growing tensions over legacy infrastructure. “Hyperlume’s technology eliminates bottlenecks that hinder AI performance and sets new benchmarks for data center energy efficiency,” he said.
As AI adoption accelerates, the industry needs breakthroughs that not only expand performance, but also reduce energy consumption. Hyperlume bets that optical interconnections are key to making AI infrastructure faster and more sustainable.
Source link
