• Deepnight raises $5.5M to revolutionize military night vision with AI technology.
• It claims current night vision relies on outdated hardware, limiting its effectiveness and cost.
• Night vision goggles are still "esoteric hardware," Deepnight's cofounder told BI.

AI is rapidly changing the face of military technology, enabling innovations like autonomous drones and software for managing field logistics.

One area that still remains outdated, however, is military night vision — a critical capability for navigating covert operations, reconnaissance, and targeting adversaries.

A startup tackling the problem through AI has caught the attention of notable investors.

Deepnight made a splash last week by announcing it had raised $5.5 million in funding. The round was led by Initialized Capital, with participation from Y Combinator, Vladlen Koltun — a computer scientist who coauthored "Learning to See in the Dark," a seminal paper on using AI for low-light imaging — angel investor Kulveer Taggar, Brian Shin, a former partner at the CIA's venture capital arm In-Q-Tel, and Matthew Bellamy, lead singer of the band Muse. The company has also landed $4.6 million in contracts from the US Army, Air Force, and private sector companies Sionyx and SRI International.

Deepnight's innovation combines low-light cameras with a novel AI image processing model that improves low-light imagery and can run on just a smartphone. The company plans to use its technology to mass produce higher-performance digital night vision goggles at a fraction of the cost of current technology.

Google alums Lucas Young and Thomas Li launched the company through Y Combinator in 2023. The founders are childhood friends from New Jersey who stayed in touch as they attended college in California. Young went to Cal Polytechnic State University, where he specialized in computer vision, and went on to research computer vision problems at Meta and Google. Li went to the University of California, Berkeley, studied electrical engineering and computer science, and worked on machine learning systems at Lawrence Berkeley National Laboratory, Lyft, and Google.

When hardware becomes software

The US Army began using night vision during the Vietnam War in the form of "starlight" scopes. These 6 lb. scopes were bulky, hard to use, and relied solely on ambient light and basic hardware to enhance the visual contrast of images. The technology has progressed since then, but incumbents like L3Harris still rely on hardware like image intensifiers to convert available light into electrons, amplify them, and then convert them back into visible light to improve images.

Young told Business Insider that current night vision goggles fall into a category he calls "esoteric hardware." He said, "military night vision currently is sort of stuck in the same technology valley as analog record players or film cameras."

Deepnight saw this as an opportunity for reinvention. The key, Young said, was reframing it from a hardware problem into one that could be solved through software by improving how cameras encode light into a digital signal.

"We're now doing the digital transition," Young said. " That digital signal is super corrupt and limited, so that's why we've made an AI model that processes that and rectifies the corruption of that."

Deepnight aims to cut the cost of night vision goggles — which currently run tens of thousands of dollars — down to $2,000, Young wrote by email. The company aims to build technology for both defense and commercial customers, where it's seeing growing interest in applications for consumer drones, smartphones, and advanced driver assistance systems in cars. An "AI model is the cheapest possible way to see in the dark and is thereby scalable to many applications besides military," Young wrote to BI by email.

The simplicity of the company's concept has attracted investors, but Young said skepticism has also been a natural part of the process. " It should be incumbent upon us to just prove that this works with undeniable, awesome demos that are rigorously metricized," he said.

Deepnight has turned model evaluation into a company ritual. Every new moon, the team drives to the outskirts of San Francisco to collect nighttime footage, using it to improve the accuracy and reliability of its models, one lunar cycle at a time.

See the pitch deck Deepnight used to raise money.