AI Physical Security

Defense-grade AI is pulling physical security capital into mega-rounds

Anduril's $1.3B growth round at a $90B valuation (and a reported $1.3B employee tender at the same valuation) shows how far the capital center of gravity has moved toward defense-adjacent sensing and autonomy. Series D+ rounds account for 7 deals but $13.3B of the last 90 days' funding, while 16 Series A deals total $5.2B. Helsing, Alta Ares (AI air defense), Fortem and D-Fend (drone detection) sit in the same orbit, and the 27 deals in the last 28 days came to $13.76B. Late-stage capital is concentrating in a few defense-linked names rather than spreading across commercial camera analytics.

Nvidia leads the investor table with 50 deals, far ahead of Blackstone (8) and Goldman Sachs (6). It is acquiring Reflection for $30B, planning stakes in four French startups, and shipping open multimodal models such as Nemotron 3.5 Super VL. Its stack reaches the edge, with GR00T N1.7-3B used in robotics research and a full RTX 5090 running onboard the Ulysses Mako vehicle. Private credit firms such as Blackstone and Apollo back the surrounding infrastructure.

Why it matters · Physical-security vendors building on Nvidia's stack gain distribution and funding access, but they take on supplier concentration risk.

Flock Safety, last valued at $8.4B, is cutting 270 people (18%) amid backlash over unencrypted stored images, clear-text camera credentials and a police interface with no authentication. Its CEO is publicly resisting terminating customers where abuse is limited. This puts pressure on license-plate-reader and body-cam analytics vendors such as TRULEO and ShotSpotter, and advocacy groups such as EPIC are active in the space.

Why it matters · Security and privacy posture is becoming a procurement gate, which favors vendors with auditable, encrypted architectures.

Startups such as Conntour (a natural-language search engine for camera feeds), ARGU (Vision Agents that need no model training) and Coram AI (an AI-native platform integrating video, access and emergency workflows) are reframing cameras as queryable systems. Multimodal open models like Nemotron 3.5 Super VL, which handles text, image, document and video inputs for agents, lower the barrier to building these interfaces.

Why it matters · The value is shifting from camera hardware to the intelligence layer, which exposes incumbents such as Verkada and Motorola Solutions to software-native challengers.

Seed rounds total $5.35B across 7 deals and Series A $5.24B across 16, which points to outsized early checks. Weekly capital has been lumpy, with $10.98B in the week of Jul 27 and $8.46B in the week of Sep 21. It then fell to $313M across 10 deals the week of Oct 5 and no recorded deals the week of Oct 12, even as mentions stayed near 100 per week.

Why it matters · Averages are skewed by a handful of giant rounds, so investors should read the recent drop in deal flow cautiously against persistent attention.