Data centers consumed 4.4% of U.S. electricity in 2023 and are estimated to use up to 12% by 2028. The majority of the energy data centers suck up is used to help transfer data from chip to chip. A company called Hyperlume is looking to make that process more energy-efficient while also speeding it up.
Ottawa, Canada-based Hyperlume created a version of microLEDs that can transfer information faster than the copper-based connections commonly found between the racks in data centers. These microLEDs also require less energy to transfer data than copper wires.
Hyperlume co-founder and CEO Mohsen Asad told TechCrunch that the company was a “logical extension” of the work he and his co-founder Hossein Fariborzi were doing before founding the company. Asad’s background in electrical engineering led him to a career focused on figuring out ways to transfer data between chips and between racks. Fariborzi has expertise in low-power electrical circuit design.
“I was working on microLEDs, I was working in data transfer, and this boom of AI and the requirements for sending information from chip to chip, power consumption — all things came together naturally,” Asad said. “We found a huge market opportunity.”
Energy consumption and latency have always been problems for chip-to-chip communication in data centers, Asad said, but they’ve been exacerbated by the rise — and breakneck pace — of AI. Solving the latency issue, or data delay, will not only speed up existing links between chips but also unlock chip capacity that wasn’t previously accessible due to the latency bottlenecks, Asad added.
“If we can solve this latency issue practically, we make [chips] work faster together,” Asad said. “When you have large language models […] you need the chips to communicate with almost zero …