The most dangerous thing a soldier or security operator can see in the dark is nothing at all. For decades, the solution has been bulky, expensive, and monochrome: analog image-intensifier tubes that amplify scant light into a grainy green glow. Now, a San Francisco startup is betting that the next generation of night vision won't come from a better tube, but from a smarter algorithm. Deepnight is embedding AI models into cameras to transform near-total darkness into vivid color video, and it has already convinced some of the Pentagon's most demanding customers to take a look.
Founded in 2024 by ex-Google engineers Lucas Young and Thomas Li, Deepnight has moved with the urgency of a company that found its market on day one. Within its first year, the company reported booking approximately $4.6 million in contracts with federal agencies, including the U.S. Air Force and Army [TechCrunch, February 2025]. Its software is now part of integrated counter-drone systems protecting military installations, a rapid validation for a technology that aims to upend a multi-billion-dollar industry long dominated by defense giants.
The Algorithmic Wedge
Deepnight's core proposition is a software wedge. Instead of manufacturing proprietary hardware, the company builds AI models that can be integrated into standard, low-light camera sensors,the kind found in smartphones. These models process the extremely faint signals available in conditions as dark as 0.1 millilux, a level described as overcast, moonless starlight [Perplexity Sonar Pro Brief, retrieved 2026]. By aggregating photons temporally and reconstructing the scene, the software claims to deliver color video with a wider field of view and better motion handling than traditional Generation 3 image intensifiers, and at a fraction of the cost [deepnight.com, retrieved 2026].
This digital approach creates a dual-use pathway. The primary and immediate market is defense, where the technology slots into goggles, helmet systems, and surveillance platforms. But the underlying software architecture is designed for any platform that needs to see in the dark.
- Defense and Security. Deepnight's announced partnerships are telling. Its software provides the low-light vision for Circle Optics' hemispherical imaging systems deployed for Counter-UAS (Unmanned Aircraft Systems) capability at U.S. Air Force bases [Circle Optics, June 2026]. Another partnership with Picogrid brings "AI-enabled digital night vision to protect critical military sites" [PRWeb, 2026].
- Broader Horizons. The company's stated roadmap includes integration into drones, automotive systems, and maritime cameras [Perplexity Sonar Pro Brief, retrieved 2026]. By starting with the stringent requirements of military contracts, Deepnight aims to build a robust, battle-tested core that can later scale into commercial sectors.
Funding and Early Traction
The technical ambition and early customer traction attracted notable investors. After participating in Y Combinator's Winter 2024 batch, Deepnight closed a $5.5 million seed round in February 2025 led by Initialized Capital [TechCrunch, February 2025]. The round included angels like Kulveer Taggar, former In-Q-Tel partner Brian Shin, and musician Matthew Bellamy. The capital appears to be fueling both R&D and business development, with the company growing to an estimated 11-50 employees [LinkedIn, retrieved 2026].
The following table outlines the company's disclosed funding to date:
| Round | Date | Amount | Lead Investor |
|---|---|---|---|
| Accelerator | January 2024 | $500,000 | Y Combinator [PitchBook, 2024] |
| Seed | February 2025 | $5,500,000 | Initialized Capital Management [TechCrunch, February 2025] |
The Competitive Field
Deepnight enters a field with entrenched incumbents and a handful of tech-forward challengers. Its success hinges on convincing buyers that digital AI processing is not just a cheaper alternative, but a superior one. The competitive landscape breaks down into a few key segments.
- Legacy Defense. Companies like L3Harris represent the gold standard in analog night vision, with decades of contracts and deeply embedded supply chains. Competing on pure performance claims is one thing; displacing an entire procurement ecosystem is another.
- Digital Pioneers. Firms like Sionyx have commercialized digital night vision sensors using specialized silicon. Raytron is another player in the thermal and low-light imaging space. Deepnight's differentiation here is its focus on AI reconstruction software that works with commodity sensors, potentially offering a cost and performance advantage over other digital approaches.
- In-House Development. Large defense primes and tech companies could theoretically develop similar AI vision capabilities internally. Deepnight's answer is speed and specialization,moving faster than bureaucratic R&D cycles to deliver a production-ready product.
The startup's early government contracts suggest it is making headway, but the true test will be moving from initial prototypes and pilot programs to large-scale, recurring procurement.
The Path Ahead and Inherent Risks
For a hardware-enabled software company in defense tech, the road is paved with both opportunity and formidable hurdles. Deepnight's next twelve months will likely focus on executing its current contracts flawlessly, expanding its partnership network with system integrators, and pursuing additional military program-of-record opportunities. A Series A round seems a plausible next step to scale manufacturing support and commercial market exploration.
The risks are substantial but not unique. Defense procurement is famously slow and political; a promising technology can stall in the "valley of death" between prototype and production. Furthermore, while AI models excel in controlled training environments, edge cases in real-world battlefield conditions,extreme weather, smoke, deliberate countermeasures,could reveal limitations. Deepnight's bet is that its software-first, iterative development model will allow it to adapt and improve faster than hardware-bound competitors.
For the warfighter or security officer, the clinical need is unambiguous: situational awareness in low-light environments to ensure mission success and personal safety. The current standard of care is a mix of Generation 3 image intensifier tubes, which provide a high-resolution but monochrome and sometimes fragile view, and thermal cameras, which see heat signatures but lack detailed visual context. Both are expensive, and the former imposes significant logistical and maintenance burdens. Deepnight is proposing a new standard,digital, color, and continuously improvable via software updates,that aims to address those gaps directly. If the technology holds up under the pressure of real-world use, it could redefine not just how militaries see at night, but how all machines perceive the dark.
Sources
- [TechCrunch, February 2025] YC grad Deepnight nabs $5.5M for AI night vision software that disrupts a multi-billion-dollar industry | https://techcrunch.com/2025/02/27/yc-grad-deepnight-nabs-5-5m-for-ai-night-vision-software-that-disrupts-a-multi-billion-dollar-industry/
- [Perplexity Sonar Pro Brief, retrieved 2026] Deepnight company briefing | (Source from research snippets)
- [deepnight.com, retrieved 2026] Deepnight - Nextgen Night Vision | https://deepnight.com/
- [Circle Optics, June 2026] Deepnight partnership announcement | (Source from research snippets)
- [PRWeb, 2026] Deepnight and Picogrid partnership announcement | (Source from research snippets)
- [PitchBook, 2024] Deepnight accelerator funding data | https://pitchbook.com/profiles/company/593014-87
- [LinkedIn, retrieved 2026] Deepnight company page | https://www.linkedin.com/company/deepnight-ai