Neurophos to Test Optical AI Chip in Norway

Neurophos, a US startup, is gearing up to test its optical AI chips in a datacentre located in Norway, with the trial set to take place in 2027. Collaborating with Terakraft, the company aims to pilot the technology in a green AI datacentre that runs entirely on renewable hydropower and utilizes natural lake water for cooling. This setup boasts an impressive Power Usage Effectiveness (PUE) rating of below 1.1, positioning it as one of the most energy-efficient facilities in the industry.

As part of a commercial early access program scheduled for 2027, Neurophos and Terakraft are planning to conduct a pilot of the Neurophos accelerated AI inference platform. This initiative will serve as a practical testing ground for the integration of in-memory optical compute with GPUs, showcasing the potential of this cutting-edge technology.

• The art of cooling
• Blaming drought on old emails is madness

Neurophos has successfully developed and demonstrated its miniaturized optical modulators that leverage metamaterials to significantly reduce energy consumption. These advancements enable optical processing units to deliver computational power equivalent to 100 GPUs while consuming just 1% of the energy, a feat validated through comprehensive end-to-end simulation results.

Giorgio Sbriglia, Chairman of the Board of Terakraft, expressed enthusiasm about the collaboration, stating, “By incorporating Neurophos’ highly efficient optical chips into our eco-friendly data centre for select enterprise clients, we are not only minimizing our environmental impact but also setting new standards for energy-efficient AI infrastructure. Our overarching goal has always been to drive the future responsibly, and this partnership is a tangible realization of that vision.”

Patrick Bowen, CEO and Founder of Neurophos, echoed this sentiment, remarking, “Terakraft’s dedication to renewable energy and innovative solutions perfectly aligns with our mission to democratize high-performance AI. Through the deployment of our 100 times more efficient inference chips within Terakraft’s sustainable data centre, we are demonstrating that the exponential growth of AI can be achieved in an environmentally conscious manner, marking a significant step forward for the industry.”